JP2006500643A - Active visual recognition method and apparatus for feature extraction and recognition by multi-class histogram calculation and analysis of one-dimensional / multi-dimensional parameters in processing apparatus by dynamic replenishment of apparatus - Google Patents

Abstract

The present invention relates to an active visual recognition method and apparatus for characterizing and recognizing an object with varying spatial resolution for object identification and local identification. An object is formed of a sequence of digital signal sequences and sub-sequences, and the method changes the spatial resolution of the signal object in time Gaussian in three or more sequences. This change includes a phase of increasing resolution from a decreasing value to an optimal basic value, during which a difference with a smoothing effect is obtained, and when the difference exceeds a threshold, the Gaussian difference A signal representing the change is obtained, and further a signal representing the change in the hierarchical Gaussian difference that can be represented in the form of an analysis tree of the object. The variation can be obtained by mechanical or electronic filtering. The invention further relates to a one-dimensional or multi-dimensional multi-class module that can be dynamically supplemented for histogram calculation and processing.

Description

The present invention provides for the feature extraction and recognition of objects by means of dynamic replenishment of the device, as much as possible of multi-class histogram calculation and analysis of one-dimensional / multi-dimensional parameters on the processing device, in particular identification or identification and For local identification,
It relates to an active visual recognition method and apparatus.
For active visual recognition, the present invention performs a gradual increase in resolution, in particular a gradual increase step, especially after a rapid increase step in the spatial resolution of the perception of the observed object.

Therefore, instead of showing all the details at the same time that are difficult to examine, interpret and classify and present the object to be observed, the minimum resolution (at the end of the resolution reduction phase and before the start of the resolution enhancement phase) and ( By gradually increasing the spatial resolution between the maximum resolution at the end of the resolution increase phase when searching for the base resolution before the start of the resolution reduction phase, the details of the object to be observed are the most characteristic details. To the least characteristic details can be organized and revealed in stages and hierarchies, which allows the details to be examined, interpreted and classified in order of importance.

Instead of using all detail data at the same time (before any change in resolution), it is most important only to reduce the resolution and maintain one or very few very important details The method is seemingly paradoxical because it arbitrarily removes low details. In fact, according to this method, all of the details or a small number of details mentioned above can be separated gradually with increasing resolution to gradually reveal non-characteristic details.
Tag or label against it by presenting details in a hierarchical order of occurrence in the resolution increasing process, for example in a tree format illustrating the parent-child relationship between a zone and a part of the zone By assigning, it is possible to extract features of the object.

  The object can be observed by means of sensors, in particular by means of, for example, MOS-type or CCD-type sensors, which generate a digital signal in the form of television and represent an image of the object including its surroundings at that moment. Contains a continuous frame or sequence. Each frame or sequence consists of a series of horizontal scan lines, and each horizontal scan line contains a series of pixels.

  Modifications, in particular the increase in resolution, seem to act optically on the front optical system of the sensor before converting the optical image into an electronic image (ie a frame) or Can be realized by electronically processing the coefficients of the filter that processes the digital signal generated by (ie, successive frames). In either case, the processing is Gaussian or pseudo-Gaussian.

In essence, the present invention performs the following three sequential steps:
* Increase resolution gradually and substantially Gaussian.
* A method and apparatus according to the following international publication by the same inventor as the present invention is implemented to differentiate with the effect of smoothing between two consecutive sequences in the above increase in resolution. Thereby, a digital signal representing a Gaussian difference between two successive sequences is derived.
WO98 / 05002

And
* Deduces hierarchically organized details from the difference between the above derived signal values of successive sequences at the same pixel location, from the most characteristic details to the few characteristic details.

Preferably, the third process described above realizes a module called STN (spatial and temporal neuron) disclosed in the following two patent documents by the same inventor as the present invention.
WO01 / 63557 French patent application FR01 / 002539

  This type of STN module can be schematized in FIGS. 19 and 22, respectively. FIG. 19 is a one-dimensional (ie, univariate linear) module being processed, and FIG. 22 is for two input parameters (eg, two coordinates, Cartesian coordinates x and y, or polar coordinates α and β). Module with one two-dimensional (ie bivariate linear) input parameter. Within the inventive STN framework, a three-dimensional (ie, three-variable linear) STN module or three input parameters (eg, three color components) can be used. In summary, such an STN module performs logic processing represented by the function fog on input parameters and outputs it using one or several registers (reg) under the control of the command API. Generate an output signal at D.

The one-dimensional and multi-dimensional parameters processed by the STN module are characteristic of objects in the space-time domain and are represented by sequences and sub-sequences of data that develop over time.
The present invention can be applied to the analysis of video in the form of video data for identification and local identification of objects in the screen. By this analysis, the object is determined according to the shape of the object and / or the size of the object and / or the direction of the object and / or the position of the object relative to the screen and / or the position of the object relative to other objects in the screen. Local identification of things is possible. The application of the present invention to active visual cognition is specifically detailed herein. The invention can also be implemented with other types of data that can be represented in the form of sequences and subsequences of data, such as sounds.

Methods and devices that allow local identification of objects in an image are already known. In one known example, a pixel of a video digital signal sent from an observation device is statistically analyzed to realize an effective device capable of operating in real time.
In recent years, it has been proposed to configure the above-described apparatus by combining information processing apparatuses having similar properties that respectively address specific parameters extracted from video signals for analysis. These correspond to the above-mentioned WO98 / 05002, French patent application FR01 / 002539 (filed on February 23, 2001), WO01 / 63557, the following international applications, or the following documents.
WO01 / 11609 Electronique, Juin 2000, No. 104, "Le mecanisme de la vision s'integre sur une puce"

In the above document, a plurality of blocks or units or modules used here in the same meaning are provided to form a plurality of calculation and processing histograms that form truly electronic spatial and temporal neurons, each with one parameter. Analyze. Here, the parameters are processed with a given function (fog) to generate output values individually. All of these output values together form the feedback available on the bus for use in the block being analyzed. At the same time, each of the histogram calculation and processing units configures and updates an analysis output register that provides statistical information about the corresponding parameters. The selection of parameters to be analyzed by each histogram calculation and processing unit, the contents of the analysis output register, and the function (fog) that it meets are determined by the partial software running in the application programming interface API.

In the histogram calculation and processing block disclosed in WO01 / 63557 (described above), for a given parameter, from the histogram calculated and stored in memory, the maximum value RMAX of the histogram, the position POSRMAX of the maximum value, Determine a number of points NBPTS in the histogram. In addition, a classification limit is determined that can accurately define the zone of interest for the parameter. Taking the maximum ratio of histograms, eg RMAX / 2, as a criterion for determining the limit, and obtaining the limit by scanning the data in the memory from the origin looking for the limit in the zone corresponding to the criterion Has been suggested.

  The application of the STN block is described in particular detail in the French patent application FR01 / 002539 (described above), in which the object to be localized in terms of its nature is decomposed hierarchically, for example, first of all relatively It has been proposed to determine the overall contour of a moving object against a stable background and then look for characteristic elements inside that contour by its tone, color, relative position, etc. Yes. Such a method allows for rapid elaboration of a number of applications associated with local identification of objects. These applications are either from pre-normalization processing that extracted important features of the object, or by a learning function that examines the screen that can extract the feature parameters of the object when the object is in the screen. Can be developed.

Therefore, the above-mentioned application FR01 / 02539 suggests a method for local identification of a shape in a space represented by pixels that together form a multidimensional space i, j that is time-dependent and represented by a sequence of moments. Yes. Each data attached to each of the time parameters A, B,... Is in the form of a digital signal, and DATA (A), DATA (B),... Are a space instant t and a position i in the space. , j, an n-bit binary sequence Aijt, Bijt,. The signals Aijt, Bijt,... Are received at a given moment.

The above application describes the following.
a) localize the zone of interest in space with respect to statistical criteria applied to the time parameter;
b) ban main zones so identified locally,
c) Repeat steps a) and b) to localize other zones of interest inside the non-prohibited zone in space,
d) When the zone of interest corresponding to the statistical criteria can no longer be generated from the remaining unbanned zones in the space, ie when the number of points in the zone of interest has decreased (below the threshold), Cancel,
e) For each successive valid frame, increment the counter for the cloud centroid of the point of each localized zone of interest as described above.
f) Collect the point cloud centroid for each zone of interest as specified above.
The above application discloses the advantage of providing two subassemblies of a histogram calculation processing block that each receive a signal and generate a classification value. The first subassembly receives a signal carrying a first time parameter, the second subassembly receives two spatial signals, and the classification value of the first subassembly is processed by the second subassembly. The number of points is n ₁ and the classification value of the second subassembly enables the values of the parameters processed by the first subassembly, The subassemblies cooperate to generate a binary signal that represents the zone of interest and a signal that represents the value of the time parameter within that zone. The application further discloses the advantage of incorporating a third subassembly in the apparatus for receiving a signal carrying a second time parameter. This third subassembly operates in the same manner as the first subassembly, and operates on behalf of the first subassembly when the number of enabling points in space is n ₂ greater than n ₁ .

The application further provides three or more subassemblies that receive spatial signals that can validate three or more consecutive groups of points in space, and are controlled by part of the API software to control the data bus and feedback bus. To implement an apparatus having a histogram computation processing block in a set coupled together.
Even though a device with a histogram calculation processing block can identify an object in the screen and localize it reliably, there are some constraints depending on the classification limit determination mode and the type of parameter being analyzed. is there.

One of the objects of the present invention is to provide an improved method and apparatus for implementing a histogram computation processing block that can increase the effectiveness and flexibility of histogram-form analysis.

  In connection with the above method, rather than using more than two blocks each processing basic parameters, instead of being able to process multidimensional parameters directly, combining partial results and scanning the module's memory from the origin Obtained from the input parameters of a complex histogram calculation processing block, i.e. from a combination of basic parameters according to predetermined criteria and / or programmable criteria, by a block that determines the classification limit from the maximum position of the histogram From the input parameters, it has been found that performing the analysis is more effective.

Similarly, it has been found that in a given block, it is more effective to determine more than two vertices for classification with respect to the histogram. Here, each of the vertices corresponds to a specific class. Thus, a multi-class functional unit can be realized in the histogram calculation processing block.

The object of the present invention consists in deriving a representative input digital signal from which the object of the environment is composed of a succession of representative sequences of successive figures of the object of the environment. Recognizing that each representative of a location was related to the time domain (each said sequence consisting of inheritance of the results) placed after each other of the sequence, and therefore provides a method for the space domain First featured in it

Over a period of substantially several sequences (worldly variations in the spatial resolution of the object of the input digital signal), including a phase of Gaussian increase in resolution from a decreasing value to an optimal basic value While 1 runs
For the smoothing effect between two consecutive sequences, the difference between the substantially absolute value of performing differentiation, for each same spatial location of the derived signal, exceeds a certain threshold In order to obtain a derived digital signal representative of the variability of Gaussian differences between both these sequences, Gaussian resolution is increased.

And from the derived signal by comparison of successive sequences between values of this derived signal corresponding to the exact same location of the result (hierarchy-organized details), 1 is the most characteristic of the object Deduct that it starts with something.
Preferably, the input digital signal consists of a video sensor, the sequence in which a continuous image of the video signal is formed, the result in which a line of the video signal is formed, and the pixel location of the video signal. A video signal from a location, wherein the characteristic of the information transported by the digital signal relates to an object present in the scene observed by the sensor.

This type of method also presents other features. And that can be combined:
The Gaussian increase in spatial resolution is performed during the increase step between two successive sequences in the stage, substantially without any change in resolution between the sequences.
-Spatial resolution variations include:
From the base value, the decreasing value before the phase that the increase phase is relatively slow from the gaussian increase substantially and from the decreasing value to the base value. Until the pre-phase of sudden reduction in spatial resolution
-Spatial resolution variations include:
The phase of the phase of the slow increase in resolution, and the substantially constant resolution during the phase of sudden event reduction,
The substantially Gaussian increase in spatial resolution is realized by electronic filtering of the input video signal generated by the sensor
-Filtering formula h (w, k) = Cp. (W + 2- | k |) (w + 1- | k |) {-3k2 + (2w + 3) | k | + w (w + 3 ) For the spatial resolution given by Cp = 5 / {2w (w + 1) (w + 2) (w + 3) (2w + 3) w given by the formula}
((Standard deviation of Gaussian mean) = 0,3217w + 0481,
-Because the sensor has a varying focal length and an object with it, the resolution variation is implemented by acting on the focal length of the lens,

-Itself to correct the focal length for the object, substantially during the Gaussian increase pitch of resolution (replacement of the lens being performed at a constant speed between the two stages) from the object Move at least one lens to
-By electrical control, it is in the form of at least one lens to change at a constant rate that its focal length is substantially staged during a Gaussian increase pitch of resolution between the object and 2 To correct.
-Substantially Gaussian resolution variation from performing the recurrent world and spatial methods (known to be originally derived), which are essentially Gaussian to perform differentiation, but increasing.
-Resolution of each same spatial location of the sequence between two consecutive sequences on and not exceeding one hand (binary signal DP), where both values each represent a surpassed threshold , Threshold of absolute value, difference and
-Forming a derived digital signal, a constant representing the adaptation time re-entered to reduce variations between two consecutive sequences in the method, having a decreasing number of bits and resolution The same location of the sequence when the processed signal of the difference, the continuous value of CO, the increase in resolution, the DP represents the value of the surpassed threshold (representative) related to the continuous sequence For (digital signal CO).
-For absolute values smaller than the threshold, the resolution of the space chosen for the limit L in the histogram (sequence with basic resolution and later reduced resolution so that the number of points in the histogram includes: In the absolute value histogram of the differences in the sequence, one form decreases during the previous phase of a sudden event.

For the aforementioned phase of spatial resolution increase, the stage finds at least 75% fractional sensitivity of the total number of histogram points selected by the increase step p, which is not greater than the cheaper integer price due to the L / S defect, and the sensitivity The threshold itself is a decreasing phase at the beginning of the increasing phase of the S resolution given to the digital signal CO and a zero value at the end of the value p.
Conveniently to perform in an action recognition method, the recognition of objects in the environment and local identification other than the recognition and localization only, more than the signal derived from the elicited and fewer features Digital magnitude of the signal at least at various locations, which provide information related to characteristic details of the object, detailed by forming at least two histograms of the derived digital signal of And at least for anything else, the local identification of the location of the signal. And it supplies information related to the location of the feature details.
This previous action recognition method also has other features. And that is, it can be combined.

-As such, it sorts these values when it shows a value corresponding to a threshold that exceeds at least one pair of DPs (according to the different frequency events of CO and the other frequency according to local identification of CO). Submit continuous values of signal CO to the classification method that is essential to generate a histogram
-Based on the classification of several parameters, at least one of the histograms is multi-line type
-Based on the classification of the coordinate x and the y of the location, one of the least of the histograms that has the form of two lines
-Based on the three parameters defining the color, one of the minimum of the histogram, it has a three-line type
-A dominant color is associated with each characteristic zone, and itself defines each successive zone according to its coordinates and its dominant color at once
-Starting with the sequence of events of these zones (analysis tree including), each of these zones and their center of gravity define as well, which is less and less features of the object Determine successive zones that correspond to certain details.

The normal origin corresponding to the center of gravity of the first zone with the first decreasing resolution, which increases to the point corresponding to the resolution that increases branching to the base resolution and the center of gravity s of the zone with it
-Itself performs an axis rotation operation on the reference plane between axis x, rotation and y, Y before axis X after rotation according to the matrix formula
-A new zone centered on the center of gravity of the zone where the corner sector was previously determined, several in which one search locates in which sector the center of gravity of the new zone is Divide the sector into several sub-players-sectors, search for which sub-sector within which the center of gravity of this new zone is located in the sector, and the polar of this After defining the coordinates, the center of gravity of the previously determined zone with respect to the center of gravity from the angle depicting the features of the secondary zone containing the new center of gravity and the distance between both centers of gravity s
-Itself is the number of centroids of the zone that appears relative to the first zone that states that the object stores the coordinates of these polar regions in the form of a label representative of the object, followed by a minimum resolution and the point of time Performs determination of polar coordinates of s.

-To determine whether the observed new object matches or is stored and does not correspond to an object previously classified as 1, it is the previous means and the label of the new object by itself Decide to compare the label of this new object with the label stored in
-Determine the distance D of the object to the sensor, perform the sectorization, think that the logarithm LD of this distance D is followed by the coordinates of dyslexia by sub-sectorization, and (according to previous means and itself Doing think that the sub-zone decides that there is all values of LD that are dimensioned for the same object and the value that deforms there (ie C ()) is constant is there.
The object of the present invention is an input digital signal from which each representative of a location is composed of the succession of a representative sequence of successive figures of the objects in its environment, the presence of which is the result of the inheritance After each other of the sequences, each of the arranged sequences (and therefore related to time the area) is drawn, and thus includes means relating to the space area, in which the device is connected to the object of its environment Featured to provide for cognition, further it includes

Duration of several time sequences (worldly variations in the spatial resolution of the object of the input digital signal), including a phase of Gaussian increase in resolution from a decreasing value to an optimal basic value Means to perform during
-
The difference between the substantially absolute value of each of the derived signals for the same spatial location that the means with smoothing for performing differentiation performs between two consecutive sequences is To obtain a derived digital signal representative of the variability of Gaussian differences in successive sequences when a certain threshold is exceeded, the Gaussian resolution increases and
In comparison, deduct the hierarchy-organized details starting from the feature of the object between the extracted signal in a continuous sequence and the value of this extracted signal corresponding to the exact same location of the result means.

Preferably, said means for extracting an input digital signal is a scene in which an object comprising a careful video sensor deducts a video signal from which said input digital signal is formed, a continuous image of the video signal Are presented in locations comprising the sequence of pixels formed and the locations of the pixels of the video signal.
This motion recognition device also has other features. And that can be combined:
-Said means for cashing substantially without any change in resolution between sequences, the Gaussian increase in spatial resolution is increased by the stage during the increase step between two groups of consecutive sequences. Cause an increase
-From its basic value that the increase phase is relatively slow before its phase of Gaussian increase, to its decreasing value and from its decreasing value to its basic value. Said means for realizing spatial resolution fluctuations generate a pre-phase of sudden reduction of spatial resolution events
-Said means for realizing spatial resolution variation generates a sudden increase in the slow increase in resolution and a phase of substantially constant resolution between phases;
-The means for cashing the space substantially to increase the latter (an electronic filter from which the input is connected to the output of the video sensor in order to receive the input video signal) Including a Gaussian increase in resolution

-The filter is the formula h (w, k) = Cp. (W + 2- | k |) (w + 1- | k |) {-3k2 + (2w + 3) | k | + w (w + 3) Perform the filtering with the resolution given by Cp = 5 / {with 2w (w + 1) (w + 2) (w + 3) (2w + 3) w} given by the formula
((Standard deviation of Gaussian mean) = 0,3217w + 0481,
-For objects with varying focal lengths, the sensor is adapted to implement the focal length of this lens, and the means realizing the spatial resolution variation.
-The device has at least one lens to correct its focal length (replacement of the lens being performed at a constant speed between the two stages) during the pitch of the Gaussian increase in resolution. Including means for transferring from the object
The device comprises a electromotive object in which the means for modifying the form of at least one lens to change it from the object by a constant increase the speed of its focal length, the substantially Gaussian increase is in two stages In between the resolution.

-A substantially Gaussian resolution variation consisting essentially of Gaussian increase in said means to achieve differentiation, known to implement recurring world and spatial methods that draw on themselves ,
-Resolution of each same spatial location of the sequence between two consecutive sequences on and not exceeding one hand (binary signal DP), where both values each represent a surpassed threshold , Threshold of absolute value, difference and
-Forming a derived digital signal, a constant representing the adaptation time re-entered to reduce variations between two consecutive sequences in the method, having a decreasing number of bits and resolution The same location of the result when the difference signal processed signal, the CO continuous value, the increase in resolution, the DP represents the value of the surpassed threshold (representative) related to the continuous sequence On the other hand (digital signal CO) for
Means for the apparatus to select in the histogram during the previous phase of the spatial resolution, the absolute histogram of the differences in the sequence with the basic resolution and the subsequent sudden reduction of the sequence with the decreasing resolution A limit L including means for generating, the number of points in the histogram for absolute values less than the threshold value includes:

A fraction S of at least 75% of the total number of points in the histogram (means to choose) for said phase of increase in spatial resolution by stage (increase step p not greater than cheaper integer price due to defects in mass ratio L / S) Presence detection sensitivity threshold,
And
A zero value means to disturb the digital signal CO at the beginning of the decreasing phase and the increasing resolution phase at the end of the value p.
Conveniently, the device of action recognition further includes at least two of the extracted digital from the extracted signal so as not to allow only recognition and local identification of objects of the environment and recognition. Includes means for deducing less and less features about the signal at least at various locations (which provide information related to the features of the object), the details of the object signaled by forming two histograms , And at least for anything else, the local location of the signal. And it supplies information related to the location of the feature details.
This previous motion recognition device also has other features. And that is, it can be combined.

-Sort these values when the device shows a threshold where at least one pair of DPs has been exceeded by local identification of CO, one that follows the frequency of events of different values of CO and one corresponding to the other A means of submitting successive values of the signal CO to a classification method that is intrinsic to generating a histogram
-Based on the classification of several parameters, at least one of the histograms is multi-line type
-Based on the classification of the coordinate x and the y of the location, one of the least of the histograms that has the form of two lines

-Based on the three parameters defining the color, the least one of the histograms is in the form of three lines.
-The device includes means associated with the dominant color in the means to define each characteristic zone and each successive zone that coincides with its coordinates and its dominant color at once
The device includes means for determining successive zones corresponding to less and less characteristic details of the object as well as the center of gravity of each of these zones;

And from the order of events in these zones, means to define an analysis tree that includes:
The normal origin corresponding to the center of gravity of the first zone with the first decreasing resolution, which increases to the point corresponding to the resolution that increases branching to the base resolution and the center of gravity s of the zone with it
The device further comprises means for performing an axial rotation operation in the reference plane between axis x, y before rotation and axis X (Y after rotation) according to the matrix formula
-The device includes a means for dividing the new zone into a number of angular sectors that are centered on the previously determined zone centroid, which sector within the sector the centroid of the new zone is Means for searching for means to locate and divide the sector into several sub-sectors, means for searching for which sub-zone within which the center of gravity of this new zone is located Let's let this mean to define the polar coordinates of the previously determined zone, between the angle describing the features of the sub-zone containing the new centroid and between both centroids s Center of gravity with respect to the center of gravity from the distance
-Several centroids of the zones appearing with respect to the first zone representing the object at the time of the means for the device to store these polar coordinates, in the form of a minimum resolution and label representation of the object subsequently means for performing a determination of the polar coordinates of s.

-Determining whether an observed new object matches or is stored and does not correspond to an object previously classified as it is this new object on the previous means and previously stored label Means for determining a label of the new object by means of comparing the labels of the objects.
-The device includes means for determining the distance D of the object or sensor, means for calculating the logarithm LD of this distance D, means for performing sectorization, by means of sub-sectorization, after the coordinates of dyslexia Followed and (calculates all values of LD that are dimensioned for the same object and the value in which the previous means and sub-zones are deformed, in accordance with the means determined to be there (ie ( C ()) is constant.

-Amplitude RMAX of the maximum vertex of the histogram,
-The position of said maximum number of memory POSRMAX in the histogram,
-A set of classification limits determined for the criteria was applied to the histogram by examining the memory and the discovery of the criteria.
On the other hand, reducing the scan of addressable memory addresses from the maximum POSRMAX location to generate a first set of limits with respect to the reference, while from the maximum POSRMAX location, the reference (and thus the address A scan of the address of the addressable memory to generate a second set of limits with respect to the limit determined from the position of the maximum POSRMAX in the histogram by both scans of the addressable memory. Increasingly, in accordance with the invention, generating a set of limits 1 consistent with the means of processing the histogram.

The previous invention can also be implemented for the following features. And it can be combined according to all technically feasible possibilities:
-Judgment continues in one of the following aspects:
-Maximum POSRMAX position for counting each step that generates a limit in one scan, the counting signal being added to the address, and the subtracted,
-First, all first limit cycles of the address limit then instantaneously, the counting signal presence is the new counting POSRMAX during the first and then the second is added to the position of the maximum POSRMAX Investigate to generate a limit on two scans subtracted from the maximum number of positions to examine the signal,
-Then the counting signal that is added to the position of the maximum POSRMAX during the first, first, generating the limits of all the second limits of the two examining cycles of the address, Then, during the second scan, examine the new counting signal being subtracted from the position of maximum POSRMAX
-One tool evaluated by the means of choice to authorize to select criteria for at least one of the following:
-Maximum value RMAX,
-The threshold value supplied to the module by THRESHOLD,
-NBPTS, a lot of histogram points generated by processing the histogram and stored in the module registers
-1, select the standard among RMAX / 2, THRESHOLD, NBPTS / THRESHOLD,
-At least one tool of the STN module:

-One adder / direction checking signal DIRECTION matches the binary value of the signal COUNTER generating the addressing signal COUNTER offsets the value to the maximum number of histogram positions POSRMAX which zero A subtraction that allows adding or subtracting the value of the addressable memory signaling offset, while or at least equivalent
-Allows verification of the comparator and, on the other hand, the logic circuit receiving the mentioned data item in the addressable memory, and the register of one of the limits subject to checking the other hand, reference and direction On the fiance to raise the limit to update the signal to
-Addressable memory, said multiplexer having three inputs each receiving input signals carrying parameters, selection signal COUNTER and selection placed on the mentioned inputs of the adder / decrement output Multiplexer,

-One tool of the STN module means for prediction with the calculation of the average position POSMOY related to the parameters and calculation of the differential (of the parameters and parameters signed before the execution of the operating subunit The parameter due to the difference between the two consecutive averages to be subtracted from the measured differential was intended to generate an output signal for feedback with respect to a set of limits determined
-1 further generates an output barycentric signal for feedback with the first binary state when the parameters match the average position POSMOY and for the second binary state in the opposite case Execute a center-of-gravity device that targets
-Support for complex parameters A1A2A3, each of the input signal DATA (getting the latter by a binary combination of A1A2A3 ... Ap) of at least two basic parameters, the parameters analyzed by the STN module are complex and itself ... from the ApAp containing the P field corresponding to each of the basic parameters A1, A2, A3, ..., and from the position of the maximum POSRMAX in the histogram of the complex parameter by each scan of each memory That one tool means of the STN module to generate and store a set of limit registers P corresponding to parameters,
-P = 2, module STN (2) exists so-called two lines or two-dimensional,
-P = 3, module STN (3) exists so-called 3 lines or 3D,
-P is greater than 3, module presence so-called multi-line or multi-dimensional,
The module processing the unique parameter is consistent with the basic features of the present invention, the latter being the module receiving the basic parameter or module according to these which developed the feature, The complex parameters that may be included include:

Single field with only P = 1,
-Insertion into counting signal COUNTER, which counteracts the device digital to enable addressing in memory (data corresponding to specific fields of complex parameters), executing the offsetting device Make possible
-1 tool means for prediction with the calculation of the average complex position POSMOYA1A2 ... related to the calculation of the p-parameter set of complex parameters and differential (A1, (A2,. ..
(Ap for each p-basic parameter of the complex parameter A1A2A3 ... for feedback due to the difference between two consecutive complex averages and to each set of limits on which each of the basic parameters is determined Ap is to be subtracted from the corresponding signed differential prior to execution of the operational subunit targeted to generate the output signal of
-1 further has the first binary state when the complex parameter A1A2A3 ... Ap matches the average complex position POSMOYA1A2 ... and for the second binary state in the opposite case Executing a centroid-like device whose object is to generate an output centroid signal for feedback;

AND to perform operational sub-units including means allowing the generation of output signal classification room freely by combination, or Zor OR of classification zones, respectively Zand and each basic parameter
-1 in the form of two-state binary data in the classification memory that stores the presence or number of words in the class corresponds to the size of the histogram, the module STN, the reference to the non-discovery of the reference on the histogram and The first state corresponding to the discovery of the second state, having the first state value corresponding to the histogram zone formed between the limits, and sending the output signal of the classification memory on the feedback bus Did not store addressable memory,
-One initialise by functional calculation means that the addressable memory of the STN module during a given calculation cycle with an initialization value related to the value stored in the addressable memory at the end of the previous calculation cycle ,
-Perform a function to calculate equal initialization values, (Km-1) / Km, 2m with zero or more m in Km format, clocking stored value, memory effect and memory above m0 Allows to get m equal to zero corresponding to lack of effect.

Adjust m between calculation cycles,
-Method depends on criteria of approval (memory effect decreasing as approval method progress) performed in the system of m approvals and variations
-1 of at least a given histogram into which a set of data classifies, each class corresponding to the data including the vertices and amplitudes of the histogram and the vertex, a class corresponding to the maximum vertex RMAX One, store the position of POSRMAX in the histogram,
-1 determines the set of data classes between several STN modules and allocates them to the means to authorize to store,
The first module ST1, which determines and stores the first class corresponding to the maximum vertex of the histogram,
And determining the second class corresponding to the highest second and storing second module in the ST2 histogram, and so on the vertices on ST3 ... (determining the lower vertex of the histogram) The output of a module of a larger vertex class that prohibits a set of the following stored modules)
-A number of STN modules to be allocated to the means, the class in which the set is determined, the set of classes is remembered, and the order of the histogram vertices is reduced towards the set of modules ST'1 A set of data classes between the first modules ST'0, ST'2 and ST'3 that are sending the unit classification basis first ... Allocated between one of the classes and thus the module Enabled to determine and store data corresponding to a class, each module ST'1, ST'2 (ST'3 ... determine and store the class it received)

A set of data classes between the STN module and the application programming interface API (said means including the memory M0 of the histogram value determined by the module) arranged in the amplitude of the apex of the memory M histogram of the address allocated to the means Determining and enabling the API to enable storing the order number of the memory M2 class (class with different number of classes and register RC threshold memory M3) that allows to store:
-Initialization cycle resetting memory M0, M1, M2, M3 and register RC,
-Calculation cycle for loading on M0, the value of the histogram determined in the module,
-Cycle to update class,
The above updating cycle consists of:
(A) Decreasing the order of the memory M0 and sorting the amplitudes when storing the matches referred to in the memory M1
(B) (Search for classes by :)
Tag the class in memory M2,

And storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C)-checking the class of the memory M2 with the address of the considered class of M2 having the threshold of M3 corresponding to the class considered by comparing the values of the memory M0;
-The memories M0 and M1 correspond to the addressable memory of the STN module, and the corresponding data is placed in the single function table / POSRMAX0 of the amplitude and position memory couple RMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ... vertex of the histogram in reducing the order of amplitudes (table of functions performing class mechanical hardware sorting during the calculation step)
-One gather of devices with multi-class function in module STN:
-Table of functions of one of the amplitude and position memory couple RMAX0 / POSRMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ... Histogram in decreasing order of amplitude, table of functions performing class mechanical hardware sorting during calculation steps, addressing replaced with unique table of said functions Possible memory vertices,
-Memory M2, which allows the order number of classes in the store
-Class threshold memory M3,
-Class number register,
-Application programming interface API,
Multi-class functional device receiving at least one parameter simple DATA (A) or complex DATA (A1 ... Ap) (validation signal validation signal VALIDATION linear combination and sequenced signal INIT) CALCUL, END, CLOCK (device of said multi-class function sending each to at least one set behind the output signal matching the class Cl1 ... Clk on the feedback bus)-classification signal Are generated for a set of parameters whose histogram amplitude is greater than a threshold criterion—the API is a memory-based microprocessor or microcontroller type programmable sequencer,

-Determine and store a set of data classes with means, calculate and prepare for each class in memory M4 (number of histogram points NBPTS corresponding to the class) Save
-Determine and store a set of data classes with means, calculate, and prepare for each class further in memory M5 (mean position POMOY of the parameters for the class) Keep
The method comprises determining at least one class with a minimum of two multi-class modules (the first module operating in this world domain TD) and said operating in domain SD of space Implemented in an object approval system replenished for at least one half module of the class,
-Includes a set of histogram calculation and processing STN modules with sectorization, zone-determining module and centroid s, and several systems whose own system is centered on the corresponding centroid of the zone and its one search In the object recognition to divide the determined zone into corner sectors, which sector among the sectors in which the method is performed, looks new, and itself divides the sector into several sub-sectors, However, the method can continue to fine-tune sectorization for further uplifting

-1 performs two sectorization levels, the first level dividing the first zone into sectors and the second level dividing one of the sectors with new centroids into sub-sectors,
-At the end of sectorization, between determining the module, the angle given with respect to the straight line joining both centroids s and the at least one angle and the one corresponding to the straight lateral distance the module draws a line Both of the above, center of gravity s,
The method comprises at least two orientation devices p (p () of the axis at the input for reference axis rotation of at least two orthogonal coordinates of the input parameter (and the module determining the centroid for the input parameter) A second two lines that are executed in at least one STN module and contain the centroid BarZi by the association of the univariate linear module processing the first parameter in the system 1 and processing the coordinates Determine the first space Zi of the module, the second related second center of gravity determined within the first space Zi, the event of BarZi + 1, the first space divided into distinct corner sectors Distributed Zri0, Zri1, Zri2 ... uniformly (each cell being processed by the two-line module of the sector receiving Zi. Tha) second centroid BarZi + 1 of BarZi, further signals,
The two-line module of the sector corresponding to the second center of gravity BarZi + 1 presence is uniformly Zra, Zrb, Zrc .. in relation to the set of two-line modules of the later rank sector. Sectorization, and at least one more that puts it in order to progressively divide the sector that owns the second center of gravity BarZi + 1 into a clear-angle sector delivered Fine-tuning to determine the angle (substantially the center of gravity s BarZi and BarZi + next to the straight line and the reference axis perpendicular to the direction of the module that merges the center of gravity s BarZi and BarZi + 1 Distance j between 1 and reference according to angle (authorize to get translation invariance

-The object is a hand, the logarithm of the distance LD between the reference points and the system on at least one point of the object, and at least one size invariant device (at least the size invariant received at the input) The device can be observed from different distances and its one tool, and the module (distance between two centroids s BarZi of a substantially constant projection and corresponding to the angle s BarZi + 1 ( At least one value C (said device determining) j '(of the rotation in relation (and LD,
-, Calculate (j, ((C ()
-One tool of size invariant device means to control the distance, allowing the use of the external measuring distance freely or the internal determination of the distance,
-The object is observed by physical replacement from different distances or by a soaring effect,
-Allows the object to correct the rotation corrector angle values that can be observed according to different angles and one tool of the system and at least one rotation corrector (of the reference axis for a set of values ( (Angle and of previously determined module
-The rotation correction device makes it possible to further determine the angle of rotation ((
-, Calculate (LD, ((C () ",
-It is performed in an object recognition system that includes a set of histogram calculation and processing STN modules with at least one device for transforming the parameter reference system by rotation of the angle ((the two dimensions of the parameter Rotate in the case of a two-dimensional reference for the polar coordinate parameters of pixel X (Y) corresponding to the following matrix behavior:

-Angle (or (or so projection of one of the corresponding on-axis parameters ((p (has a variation ratio that decreases after rotation,
-The parameters are 2 in total and are chosen within a couple X (Y in pixel or LogD coordinates) (pixel distance with respect to reference point and logarithmic and straight of angle with respect to said point) On the reference line
-1 determines over time that several centroids s of the zone appearing with respect to the first zone and its own first centroid store centroidal coordinates in the form of approval data with respect to the label,
-1 performs an analysis tree concatenating different centroids s with respect to their order of events,
-To determine if the observed new object matches or does not correspond to an object with a previously remembered label, 1 is the new object and 1 approval data is stored previously It is determined by comparing with the approval data of the label.

Further, the label approval data is associated with the angle of rotation (by means of authorizing to determine the angle of rotation and the average distance LD ”((and said average distance LD”,
-1 is further associated with the dominant color C by means of authorizing the label approval data to determine the color;
-1 associated with the first label approval data to form a new label corresponding to a higher level of approval to those of at least one-half label,
-1 Analyze the labels by a processing module that can determine and store a histogram calculation and a set of categorization data for the labels.
Some of the features of the previous function relating to a device or to a method that the invention is intended to operate according to 1 in the same way.

-Amplitude RMAX of the maximum vertex of the histogram,
-The position of said maximum number of memory POSRMAX in the histogram,
-A set of classification limits determined for the criteria was applied to the histogram by examining the memory and the discovery of the criteria.
In accordance with the present invention, the device processes the histogram to generate a first set of limits with respect to the reference, by a decreasing scan of the addressable memory address from the position of maximum POSRMAX, On the other hand, generating a set of limits and, on the other hand, from the position of the maximum POSRMAX, with respect to the reference (and thus the limit determined from the position of the maximum POSRMAX in the histogram by both scans of addressable memory) Means for increasing a scan of addresses of the addressable memory to generate a second set of limits;

The apparatus of the present invention can also be implemented for the following features. And it can be combined according to all technically feasible possibilities:
-By means of processing the histogram and generating a set of limits, it is possible to determine one of the following aspects:
-Maximum POSRMAX position for each walking by address, manufacturing of counting signal in turn, and limit manufacturing in one scan of reduced walking
-By the manufacture of the limit in two scans where the first first limit cycle of the address is then instantaneously limited, the counting signal presence is subtracted from the maximum number of positions, POSRMAX between the first is then Examine the new counting signal being added to the position of maximum POSRMAX during the second scan,
-By producing limits in the two examining cycles of the address, first the second then restricts the first limit, the counting signal being applied to the position of maximum POSRMAX during the first Then, during the second scan, examine the new counting signal being subtracted from the position of maximum POSRMAX

-The device includes:
A means of selection that grants authority to select criteria for at least one of the following values:
-Maximum value RMAX,
-The threshold value supplied to the module by THRESHOLD,
-NBPTS, a lot of histogram points generated by processing the histogram and stored in the module registers
-The standard is selected among RMAX / 2, THRESHOLD, NBPTS / THRESHOLD,

-The device includes at least:
-One adder / direction checking signal DIRECTION matches the binary value of the signal COUNTER generating the addressing signal COUNTER offsets the value of the maximum number of histograms to either position value POSRMAX zero A divisor that allows adding or subtracting the value of the addressable memory signaling offset, while or equivalent
-Allows verification of the comparator and, on the other hand, the logic circuit receiving the mentioned data item in the addressable memory, and the register of one of the limits subject to checking the other hand, reference and direction On the fiance to raise the limit to update the signal to
-Addressable memory, said multiplexer having three inputs each receiving input signals carrying parameters, selection signal COUNTER and selection placed on the mentioned inputs of the adder / decrement output Multiplexer,
The device includes means for prediction having a calculation of the average position POSMOY related to the parameter and calculation of the differential device (of the parameter, and the parameter is signed before execution of the operating subunit) That the parameter due to the difference between two consecutive averages to be subtracted from the measured differential is intended to generate an output signal for feedback with respect to a set of limits determined
-When the parameters match the average position POSMOY and for the second binary state in the opposite case, the device further generates an output barycentric signal for feedback with the first binary state Including a center of gravity device

-Support for complex parameters A1A2A3, each of the input signal DATA (getting the latter by a binary combination of A1A2A3 ... Ap) of at least two basic parameters, the parameters analyzed by the STN module are complex and itself ... ApAp with a P field corresponding to each of the basic parameters A1, A2, A3, ... In that it includes a module means to generate and store a set of registers P that correspond to basic parameters.
-P = 2, module STN (2) exists so-called two lines or two-dimensional,
-P = 3, module STN (3) exists so-called 3 lines or 3D,
-P is greater than 3, module presence so-called multi-line or multi-dimensional,
The module processing the unique parameters is consistent with the basic features of the invention, the latter being the module receiving the basic parameters or modules according to these which have developed features The complex parameters that may be included include:

Single field with only P = 1,
-The device includes:
Offsetting device, the device digital allows the insertion into counting the countered signal COUNTER in the memory (data corresponding to a specific field of complex parameters) what can be addressed
-The device is related to a p-parameter set of complex parameters and a differential calculation (A1, (including means for prediction having a calculation of the average complex position POSMOYA1A2 ...) A2, ...
(Ap for each p-basic parameter of the complex parameter A1A2A3 ... for feedback due to the difference between two consecutive complex averages and to each set of limits on which each of the basic parameters is determined Ap is to be subtracted from the corresponding signed differential prior to execution of the operational subunit targeted to generate the output signal of
-When the complex parameter A1A2A3 ... Ap matches the average complex position POSMOYA1A2 ... and for the second binary state in the opposite case, the feedback with the first binary state Generating a center of gravity signal further meant to generate an output center of gravity signal for,

-The device includes:
Operational sub-units with means allowing the generation of output signal classification room freely in combination or Zor OR of classification zones, respectively Zand and each basic parameter
-The device includes:
To the non-discovery of the reference on the number of words on the histogram (the class with the initial state value that corresponds to the histogram zone formed between the limits and that the output signal of the classification memory is sent to the feedback bus) Module STN, class two-state binary data in the form of memory, storing the presence, the first state corresponding to the discovery of the reference and the histogram stored in the second state addressable memory Classification memory corresponding to the size of

-The device includes:
The function calculation is initialized by the addressable memory of the STN module during a given calculation cycle that has an initialization value that is related to the meaning value stored in the addressable memory at the end of the previous calculation cycle. enable
-The function used to calculate the initialization value clocks the generated (Km-1) / Km stored value, 2m with zero or more m in Km format, memory effect above some m0 and Allows to get m equal to zero corresponding to lack of memory effect
-m varies between calculation cycles,
-The device depends on the criteria of approval (memory effect decreasing as approval method progress) performed in the system of approval and variation of m
A device comprising means for determining and a set of data classifying a given histogram, the histogram containing the amplitude and position of the vertex and each class corresponding to at least the vertex of the data, a maximum vertex RMAX ( Remember for one of the classes corresponding to the histogram (POSRMAX)
-The means to determine and authorize a set of data classes is distributed among several STN modules,
The first module ST1, which determines and stores the first class corresponding to the maximum vertex of the histogram,

And determining the second class corresponding to the highest second and storing second module in the ST2 histogram, and so on the vertices on ST3 ... (determining the lower vertex of the histogram) The output of a module of a larger vertex class that prohibits a set of the following stored modules)
-Class-by-class classification basis in which the instrument class is determined by several modules, remembering a set of classes, and reducing the order of histogram vertices towards a module of a set of modules ST'1 A set of data that is allocated in the first module ST'0, ST'2, ST'3 ... Allows to determine and store data corresponding to the class, each module ST'1, ST'2 (ST'3 ... determine and store the class it received)

-Means determined by the module (memory M1 of the address arranged in the amplitude of the vertex of the histogram) said means including the memory M0 of the histogram value, the means class is distributed in the STN module and the application programming interface API A set of data is determined and enabled to execute the API that allows storing the order number of the memory M2 class (class number of classes and the memory M3 of the register RC threshold) that enables storage:
-Initialization cycle resetting memory M0, M1, M2, M3 and registers,
-Calculation cycle for loading on M0, the value of the histogram determined in the module,
-Cycles for updating classes, updating cycles include:
(A) Decreasing the order of the memory M0 and sorting the amplitudes when storing the matches referred to in the memory M1
(B) (Search for classes by :)
Tag the class in memory M2,

And storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C)-checking the class of the memory M2 with the address of the considered class of M2 having the threshold of M3 corresponding to the class considered by comparing the values of the memory M0;
-The memories M0 and M1 correspond to the addressable memory of the STN module, and the corresponding data is placed in the single function table / POSRMAX0 of the amplitude and position memory couple RMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ... vertex of the histogram in reducing the order of amplitudes (table of functions performing class mechanical hardware sorting during the calculation step)
-One gather of devices with multi-class function in module STN:
-Table of functions of one of the amplitude and position memory couple RMAX0 / POSRMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ... Histogram in decreasing order of amplitude, table of functions performing class mechanical hardware sorting during calculation steps, addressing replaced with unique table of said functions Possible memory vertices,
-Memory M2, which allows the order number of classes in the store
-Class threshold memory M3,
-Class number register,
-Application programming interface API,

Device of said multi-class function receiving at least one parameter simple DATA (A) or a combination of complex DATA (A1 ... Ap, feedback signal and validation signal VALIDATION linear of sequenced signal INIT , CALCUL, END, CLOCK (devices of said multi-class function sending each to at least one set behind the output signal matching the class Cl1 ... Clk on the feedback bus) -API A programmable sequencer of memory based microprocessor or microcontroller type,
-Determine and store a set of data classes with means, calculate, and prepare for each class further in memory M4 (number of histogram points NBPTS corresponding to said class) Keep
-Determine and store a set of data classes with means, calculate and prepare for each class further in memory M5 (mean position POMOY of said parameters of said class) Save
-Said apparatus is operating in a spatial domain SD, including determining at least one class with a minimum of two multi-class modules (the first module operating in this world domain TD) In an object approval system that is replenished for at least a half module in class.

-In an object recognition system that includes a set of histogram calculation and processing STN modules, the device determines the sectorization, the zone and the center of gravity s and thereby the determined zone corresponding to the zone Dividing into several corner sectors centered on the center of gravity, which sector within the sector that allows you to search in, to new sub-sectors for new and visible and upstairs Fine-tune sectorization to divide the sectors
-The device has at least two orientation devices p (p () of the axis at the input for reference axis rotation of at least two Cartesian coordinates of the input parameters (and the module determining the centroid for the input parameters) Of at least one STN module, including the centroid BarZi by the association of the univariate linear module processing the first parameter in system 1 and of the second two lines processing coordinates Determine the first space Zi of the module, the second related second center of gravity determined within the first space Zi, the event of BarZi + 1, the first space divided into distinct corner sectors Second, uniformly distributed Zri0, Zri1, Zri2 ... (each sector being processed by the two-line module of the sector receiving Zi) The center of gravity BarZi + 1 BarZi, and also the signal,
The two-line module of the sector corresponding to the second center of gravity BarZi + 1 presence is uniformly Zra, Zrb, Zrc .. in relation to the set of two-line modules of the sector of the later rank. Sectorization and determining at least one more place to make it possible to progressively divide the sector that owns the second center of gravity BarZi + 1 into a sector of distinct corners delivered Fine-adjust one angle (substantially between the center of gravity s BarZi and BarZi + 1 next to the straight line and the reference axis perpendicular to the direction of the module that merges the center of gravity s BarZi and BarZi + 1 Distance of j,
-Calculate the value (j, ((C ()
-The distance between the reference points that the object comprises at different distances and the system from which it can be observed, further with at least one size invariant device (the size invariant device received at least one at the input) The logarithm of LD and, on the other hand, the value at least one point of the object and module (the distance between the two centroids s BarZi corresponding to the substantially constant projection and the angle and BarZi + 1 ( At least one value C (said device determining) j '(of the rotation in relation (and LD,
-The size invariant device includes means to control the distance allowing the use of an external measuring distance or the internal determination of the distance at will,

-The object is observed by physical replacement from different distances or by a soaring effect,
-The object can be seen to match different angles, and the system allows to correct the value of the rotation corrector angle, including at least one rotation corrector (a set of Reference axis ((angle and value of previously determined module
-The rotation correction device makes it possible to further determine the angle of rotation ((
-The value (LD, ((C ()) is calculated,
-Approval system including a set of histogram calculation and processing STN module of the object includes at least one device for deforming the parameter reference system by rotating the angle ((at least above the two dimensions of the parameter A reference) rotation for a two-dimensional reference for the polar coordinate parameters of pixel X (Y) corresponding to the following matrix behavior:
-Angle (or the corresponding axis ((p (i has a variation ratio that decreases after rotation, so select one of the least parameters),
-The parameters are 2 in total and selected in couple X (Y of pixel or LogD coordinates) (pixel distance with respect to reference point and logarithm of angle with respect to said point) and reference line In
-1 determines over time that several centroids s of the zone appearing with respect to the first zone and its own first centroid store centroidal coordinates in the form of approval data with respect to the label,
-1 performs an analysis tree concatenating different centroids s with respect to their order of events,
-The device includes means for determining whether the observed new object matches or the label stored by the determination of the approval data of the new object previously and the comparison with the approval data of the previously stored label Does not correspond to objects with
-The angle of rotation is further related to the label approval data ((by means to authorize to determine the angle of rotation and the average distance LD ') (and the average distance LD ,
-Further determining the color associated with the dominant color C by means of empowering the label approval data;
-Approval data for the first label is associated with those of at least one-half label to form a new label corresponding to a higher level of approval,
The labels are analyzed by a processing module that can determine and store a histogram calculation and a set of categorization data of said labels.
Invented as a method or apparatus according to 1, or some of the previous features. And it can be combined and operates preferably on image video data.
The space of data in which the object must be localized is preferably, ie, one that is a space and a realm of space that evolves over time, and on the other hand, the object parameter is 1 (space More dimensions if can be represented or more at a given moment on 3 or 2 (surface) or 3 (quantity or light or other tone + saturation +).

_ The present invention also provides a histogram calculation carried in the form of a sequence Aij ... t (A'ij ... t) by means of a digital signal DATA (A) input with a device for the function of analyzing parameters in the method A.prime worried about operating and processing multi-class. ij ... t, ... the binary number associated with the synchronization signal, j ... to allow the number to define an immediate t of space and position i A method wherein the histogram is intrinsic to generate a set of histogram feature extraction results comprising a couple of vertices and at least the maximum amplitude histogram corresponding to the maximum vertex of the histogram and a couple of maximum amplitude data of the location Whereas it can consist of
According to the invention, 1 computes a histogram of the parameters during the computation cycle with respect to the validation signal (VALIDATION), and the presence of a couple of amplitudes RMAXi and position POSRMAXi, couple RMAXi and POSRMAXi reduce the vertex During the calculation cycle, the person decides during the calculation cycle, which is stored in the function memory having the machine hardware sort of the multi-class function device.
The present invention can also be implemented for the following features. And it can be combined according to all technically feasible possibilities:
-For k (one of the classification signals corresponding to the class associated with the vertices of the histogram) where the set of results in the form of k classification is 1 or more, it is further generated at the signaling (Cl1 ... Clk) output.
-Each of the classes of classification signals is generated for a set of parameters whose amplitude of the histogram is greater than a threshold criterion;
-The threshold criterion is a function of the value of the amplitude of the considered vertex of the histogram,
-Threshold criteria are selectable,
-1 implement multi-class function device including:
Application programming interface (API), memory M0 storing sequentially arranged amplitude RMAXi vertices, address memory M1, memory M0 aligned to vertex amplitude POSRMAXi, and a single amplitude and position memory couple M1 (RMAX0 / POSRMAX0; one table of functions)
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2. ..RMAXi / POSRMAXi ...) Further multi-class features including histogram vertex memory M2 in reducing the order of amplitudes (table of functions performing class machine hardware sorts during the calculation cycle) Allows the storage of an ordinal number of device classes (class number class and register RC threshold memory M3).

-Enable to run through API:
-Initialization cycle resetting memory M0, M1, M2, M3 and registers,
-Calculation cycle to determine and automatically classify a couple of values in a table of functions
-Cycles for updating classes, updating cycles include:
(B) (Class search step):
Tag the class in memory M2,
And
Storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C)-the step of verifying the class of the memory M2 by comparing the value of the memory M0 at the address of the considered class of M2 with the threshold of M3 corresponding to the considered class
-API is a programmable sequencer of memory based microprocessor or microcontroller type,

-For functionals, table 1 sorts and stores device B0 ... Bn with memory registers, comparators and associated logic that allows automated classification, histogram amplitude and position Perform a set of remembering vertices,
-Determine and keep in preparation for each class in memory M4 (number of histogram points NBPTSi corresponding to said class)
-Determine and keep in preparation for each class in memory M5 (average position POMOYi of the parameters of the class)
-The output centroid signal further has the first binary state when the parameter matches the average position POSMOYi and for the second binary state in the opposite case for the class considered Execute a center-of-gravity device whose target is to generate
-One tool that the means of selection to authorize to select threshold criteria for at least one of the following for a given class:
-The amplitude value of the vertex of the histogram,
-The threshold value THRESHOLD supplied to the device,
-NBPTS, lots of points in the histogram

Each class of classification signal generated for a set of parameters whose amplitude in the histogram is greater than the threshold criterion, because given class 1 is a criterion in RMAX / 2, THRESHOLD, NBPTS / THRESHOLD Select
-The parameter analyzed by the device of the multi-class function is a complex and it is at least two basic parameters, each of the input signal DATA (obtained by a binary combination of A1A2A3 ... Ap) each basic parameter A1, Support for complex parameters (A1A2A3 ... Ap) including P fields corresponding to A2, A3, ... Ap,
-1 implements a device of multi-class function of the histogram calculation and processing module, the so-called STN, the output signal forming the feedback sent to the feedback bus and the validation signal (VALIDATION) containing the feedback,
-The device with multi-class function receives at least one parameter (validation signal (VALIDATION)) when the linear combination of the feedback signal signals (INIT, CALCUL, END, CLOCK) in the case of the module STN and the arrangement,
-Multi-class device sends each to a class on the feedback bus (Cl1 ... Clk), at least one set behind the matching signal in the case of module STN,

-Object recognition system including a set of histogram calculation and processing modules, one tool with multi-class device, at least two STN modules, operating in this world TD, determining at least one class The first module replenished for at least one half module of said class operating in the area SD of the space,
-Modules operating in this world TD receive speed parameter MVT or color parameter L / T / S,
-A multi-class device, an object recognition system including a set of histogram calculation and processing STN modules by sectorization with a zone determining module and at least one module with a center of gravity s, and itself Which sector in one tool sector that divides the determined zone into several angular sectors that are centered on the corresponding centroid of the zone and its one search, looks new and itself, Divide the sector into several sub-sectors, but the method can continue to fine-tune sectorization for further uplifting

-1 performs two sectorization levels, the first level dividing the first zone into sectors and the second level dividing one of the sectors with new centroids into sub-sectors,
-The module, both given at the end of sectorization, which determines the angle given with respect to the straight line applied to both centroids s and at least one angle and the module corresponding to the straight lateral distance draws a line , Center of gravity s,
-Sectorization is performed in four sequences,
-At least one orientation device p (p (at least one of the axes at the input for the reference axis rotation of at least two Cartesian coordinates of the input parameters (and the module determining the centroid for the input parameters) The second two lines containing the barycenter BarZi by executing the method in one STN module and relating the univariate linear module processing the first parameter in system 1 and processing the coordinates Determine the first space Zi of the module, the second related second center of gravity determined within the first space Zi, the event of BarZi + 1, the first space divided into distinct corner sectors Are distributed uniformly (Zri0, Zri1, Zri2 ...) (each cell being processed by a two-line module of the sector receiving Zi. Tha) BarZi, moreover, the second centroid BarZi + 1 of the signal,
BarZi + 1 presence allowed to divide a sector that owns the second center of gravity BarZi + 1 into a sharp corner sector in relation to a set of two-line modules in a later rank sector The two-line module of the sector corresponding to the second centroid placed to be fine-tuned to uniformly determine the sectorization for ascending and at least one more angle (Zra, Zrb, Zrc ...) (center of gravity s of the reference axis perpendicular to the direction of the straight line and the module substantially merging BarZi and BarZi + 1) Reference to distance j, angle between BarZi and BarZi + 1 (authorize to get translation invariance

-Calculate (j, ((C ()
-The object is a hand, the logarithm of the distance LD between the reference points and the system on at least one point of the object, and at least one size invariant device (at least the size invariant received at the input) The device can be observed from different distances and one tool, and the module (distance between two centroids s BarZi of a substantially constant projection and corresponding to the angle s BarZi + 1 (at least J ”(the rotation of the relevant ones (and LD)
-, Calculate (LD, ((C () ",
-One tool of size invariant device means to control the distance, allowing the use of the external measuring distance freely or the internal determination of the distance,
-The object is observed by physical replacement from different distances or by a soaring effect,
-Allows the object to correct the rotation corrector angle values that can be observed according to different angles and one tool of the system and at least one rotation corrector (of the reference axis for a set of values ( (Angle and of previously determined module

-The rotation correction device makes it possible to further determine the angle of rotation ((
-Perform a method with at least one device for deforming the parameter reference system by rotating the angle ((reference at least above the two dimensions of the parameter) pixel X corresponding to the following matrix operation ( Y) For 2D reference rotation for polar coordinate parameters:
-Angle ((or () so choose one of the parameters on the smallest corresponding axis, projection ((() (p () has a variation ratio that decreases after rotation,
-The parameters are 2 in total and are chosen within a couple X (Y in pixel or LogD coordinates) (pixel distance with respect to reference point and logarithmic and straight of angle with respect to said point) On the reference line
-1 determines that over time, several centroids s of the zone appearing with respect to the first zone and the first centroid of 1 store centroidal coordinates in the form of approval data with respect to the label,

-1 performs an analysis tree concatenating different centroids s with respect to their order of events,
-To determine if the observed new object matches or does not correspond to an object with a previously remembered label, 1 is the new object and 1 approval data is stored previously To be compared with the approval data of the label
-Further, the approval data of the label is related to the angle of rotation ((by means to authorize to determine the angle of rotation and the average distance LD "(and (the said average distance LD",
-1 is further associated with the dominant color C by means of authorizing the label approval data to determine the color;
-1 associated with the first label approval data to form a new label corresponding to a higher level of approval to those of at least one-half label,
-1 Analyzing processing modules that can determine and store label approval data by histogram calculation and a set of categorization data of the label.
Some of the features of the previous function relating to a method or a method relating to a device which is a multi-class function device whose object is to operate according to 1 in the same way.

The present invention is a histogram calculation and processing multiply carried in the form of a sequence Aij ... t (A'ij ... t) by means of a digital signal DATA (A) input with a device for the function of analyzing parameters. A.prime is therefore involved in the class. ij ... t, ... the binary number associated with the synchronization signal, j ... to allow the number to define an immediate t of space and position i A method wherein the histogram is intrinsic to generate a set of histogram feature extraction results comprising a couple of vertices and at least the maximum amplitude histogram corresponding to the maximum vertex of the histogram and a couple of maximum amplitude data of the location Whereas it can consist of
According to the device of the present invention, a histogram of parameters relating to the validation signal (VALIDATION) during the calculation cycle that the device includes means is calculated, and a pair of amplitude RMAXi and position POSRMAXi, couple RMAXi and POSRMAXi are present. It is possible to automatically classify according to the command of decreasing the vertex and determine during the calculation cycle, during the calculation cycle stored in the memory of the function having the machine hardware sort of the device of the multi-class function To do.
The apparatus can also be implemented for the following features: And it can be combined according to all technically feasible possibilities:
-Outputs a signal (Cl1 ... Clk) for a k (each classification signal corresponding to the class associated with the histogram vertex) where the set of results in the form of k classification further enables the means Occur in
-Each of the classes of classification signals is generated for a set of parameters whose amplitude of the histogram is greater than a threshold criterion;
-The threshold criterion is a function of the value of the amplitude of the considered vertex of the histogram,
-Threshold criteria are selectable,

-The device includes:
Application programming interface (API), memory M0 storing amplitude RMAXi arranged in vertex order, memory M1, memory M0 of addresses arranged in vertex amplitude POSRMAXi, and amplitude and position memory couple M1 (RMAX0 / POSRMAX0; stored in a single function table;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2. ..RMAXi / POSRMAXi ...) Further multi-class features including histogram vertex memory M2 in reducing the order of amplitudes (table of functions performing class machine hardware sorts during the calculation cycle) Allows to store the number of device classes (class number of classes and memory number of register RC threshold memory M3)
-Enable to run through API:
-Initialization cycle resetting memory M0, M1, M2, M3 and registers,
-Calculation cycle to determine and automatically classify a couple of values in a table of functions
-Cycles for updating classes, updating cycles include:
(B) (Class search step):
Tag the class in memory M2,

And
Storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C)-the step of verifying the class of the memory M2 by comparing the value of the memory M0 at the address of the considered class of M2 with the threshold of M3 corresponding to the considered class
-API is a programmable sequencer of memory based microprocessor or microcontroller type,
-For functionals, table 1 sorts and stores device B0 ... Bn with memory registers, comparators and associated logic that allows automated classification, histogram amplitude and position Perform a set of remembering vertices,
-Allows to determine by device and further stores for each class of memory M4 (number of histogram points NBPTSi corresponding to said class)
-The device determines and stores the average for each class of memory M5, and further allows to place the POSMOYi of the parameters of the class,
-When the parameter matches the average position POSMOYi and for the second binary state in the opposite case for the class considered, the device further outputs an output centroid signal with the first binary state A centroidal device intended to generate

-The device includes:
A means of selection to authorize to select threshold criteria for at least one of the following values for a given class:
-The amplitude value of the vertex of the histogram,
-The threshold value THRESHOLD supplied to the device,
-NBPTS, lots of points in the histogram
Each class of classification signal generated for a set of parameters whose amplitude of the histogram is greater than the threshold criterion, because the given class, criterion is selected among RMAX / 2, THRESHOLD, NBPTS / THRESHOLD The
-The parameter to be analyzed is a complex and itself at least two basic parameters, each of the input signals DATA (the latter by the binary combination of A1A2A3 ... Ap), each of the basic parameters A1, A2, A3 Support for complex parameters (A1A2A3 ... Ap), including P fields corresponding to, ...
-There is a device in the histogram calculation and processing module (so-called STN), where the output signal forming the feedback presence sent to the validation signal (VALIDATION) containing the feedback bus and feedback,
-The device with multi-class function receives at least one parameter (validation signal (VALIDATION)) when the linear combination of the feedback signal signals (INIT, CALCUL, END, CLOCK) in the case of the module STN and the arrangement,

-Multi-class device sends each to a class on the feedback bus (Cl1 ... Clk), at least one set behind the matching signal in the case of module STN,
-A device comprising a system comprising a set of histogram calculations and a processing module comprising at least two STN modules, said device having a multi-class function (determining at least one class and operating in the region SD of space at least Included in the object approval, which is replenished for one-half module and attached the first module operating in this world TD.
-Modules operating in this world TD receive speed parameter MVT or color parameter L / T / S,
-A multi-class that allows the device to determine the zone and centroid s by system and functional devices and modules including a set of histogram calculations and to divide the determined zone into several corner sectors Processing by the sectorization with at least one module attached to the corresponding centroid of the zone and which sector among the sectors included in the object approval focused on the search, new and visible, and the method In contrast to being able to do so, the sectorization will continue to be fine-tuned during further upscaling to divide the sector into several sub-sectors
-1 performs two sectorization levels, the first level dividing the first zone into sectors and the second level dividing one of the sectors with new centroids into sub-sectors,
-The module, both given at the end of sectorization, which determines the angle given with respect to the straight line applied to both centroids s and at least one angle and the module corresponding to the straight lateral distance draws a line , Center of gravity s,
-Sectorization is performed in four sequences,

-The device has at least two orientation devices p (p () of the axis at the input for reference axis rotation of at least two Cartesian coordinates of the input parameters (and the module determining the centroid for the input parameters) Included in a system that includes at least one STN module, includes a centroid BarZi associated with a univariate linear module processing the first parameter in system 1, and a second second processing coordinate Determining the first space Zi of the line module, the second related second center of gravity determined within the first space Zi BarZi + 1 events, said divided into distinct corner sectors The first space is processed by the two-line module of the sector receiving Zi (Zri0, Zri1, Zri2 ...) (Zi received) S sector) BarZi, moreover, the second centroid BarZi + 1 of the signal,
BarZi + 1 presence allowed to divide a sector that owns the second center of gravity BarZi + 1 into a sharp corner sector in relation to a set of two-line modules in a later rank sector The two-line module of the sector corresponding to the second centroid placed to be fine-tuned to uniformly determine the sectorization for ascending and at least one more angle (Zra, Zrb, Zrc ...) (center of gravity s of the reference axis perpendicular to the direction of the straight line and the module substantially merging BarZi and BarZi + 1) Reference to distance j, angle between BarZi and BarZi + 1 (authorize to get translation invariance

-Calculate the value (j, ((C ()
-One hand that the object can be observed from different distances and systems, the logarithm value of the distance LD between the reference points and at least one size invariant device on at least one point of the object ( On the other hand, including at least the size invariant device receiving at the input, and a module (distance between two centroids s BarZi of a substantially constant projection and corresponding to an angle and BarZi + 1 (at least one of J ”(the rotation of the relevant ones (and LD,
-The value (LD, ((C ()) is calculated,
-The size invariant device includes means to control the distance allowing the use of an external measuring distance or the internal determination of the distance at will,
-The object is observed by physical replacement from different distances or by a soaring effect,
-The object can be seen to match different angles, and the system allows to correct the value of the rotation corrector angle, including at least one rotation corrector (a set of Reference axis ((angle and value of previously determined module

-The rotation correction device makes it possible to further determine the angle of rotation ((
-The device is associated with at least one device for deforming the parameter reference system by rotation of the angle ((reference at least above the two dimensions of the parameter) pixels X (Y ) 2D reference case for polar coordinate parameters: rotation:
-Angle ((or () so choose one of the parameters on the smallest corresponding axis, projection ((() (p () has a variation ratio that decreases after rotation,
-The parameters are 2 in total and are chosen within a couple X (Y in pixel or LogD coordinates) (pixel distance with respect to reference point and logarithmic and straight of angle with respect to said point) On the reference line
-A system that allows several centroids s of the zones that are appearing with respect to the initial centroid of the first zone to be determined over time by means and to store the centroidal coordinates in the form of approval data for the label of,
-1 performs an analysis tree concatenating different centroids s with respect to their order of events,
-The new object's approval data is determined and stored previously by a system that determines whether the observed new object matches or does not correspond to an object with a previously stored label. Enables comparison with approval data for labels
-Further rotation associated with the angle in the approval data of the label ((By means of authorizing to determine the angle of rotation and the average distance LD "((and said average distance LD",
-1 is further associated with the dominant color C by means of authorizing the label approval data to determine the color;

-1 associated with the first label approval data to form a new label corresponding to a higher level of approval to those of at least one-half label,
-Allows the system to analyze the label approval data by a processing module that can determine and store a histogram calculation and a set of categorization data for the label.
Invented as a method or apparatus according to 1, or some of the previous features. And it can be combined and operates preferably on image video data.
The space of data in which an object must be localized is preferably an optional space and this world region (i.e. that of one hand on which it evolves over time and the parameter of the object is 1 (point of space ) Or other hand that can be represented at a given moment on 2 (surface) or 3 (amount or more on light or other tone + saturation +)).
_ The present invention is finally supplemented with histogram calculations, and a module for cognition and a space-powered method of processing the recognition of objects in this world, along with sequence data and time Protested by the parameters carried by the unfolding results.

According to the invention, the module is related for at least one parameter and according to an application programming interface (API) zone and the relevant representation of the centroid s and at least one division of the zone In some corner sectors centered on the corresponding centroid of the zone and its one search that the new centroid has determined in which sector within the sector, and the method is further uplifted While it can last for a while, it itself decides under control to divide the sector into several sub-sectors, fine-tuning the sectorization.
(A relevant expression may be that any dimension that can correspond to a plane, quantity, or more than four dimensions of space).
The current replenishing method of the present invention also involves features that are found in the following description and that must be considered individually or consistent with all their technically possible combinations:

-1 performs two sectorization levels, the first level dividing the first zone into sectors and the second level dividing one of the sectors with new centroids into sub-sectors,
-1 limit tweaking sector (sector containing a second centroid done without new sectorization) (in fact, 1 is also repeated at the first level and / or each level of sectorization) Can be increased by increasing the number of sectors used), simply the first zone to be divided into one level of sectorization and itself
-At the end of sectorization, between the module, the angle given with respect to the straight line joining both centroids s and the at least one angle and the one corresponding to the straight lateral distance draw the line Both of the above centroids s (angle and module determinations allow object representation in invariant searches in)
-Follow the method:

-In the first sequence determined by at least one first module for at least one predetermined parameter, the first zone and centroid cooperated related expressions,
-In the second sequence one colleague with the first module, on the other hand, at least a half module was intended to determine at least a half centroid of the first zone, and on the other hand The set n of the sectorization module was going to divide the first zone into n corner sectors (the sectorization module receiving the first zone and its centroid)
-In the third sequence sent to a set of sectorized modules (the second module whose object is to determine the centroid of at least one half of which the second centroid has been determined); And 1 determine the confirmed sectorization module, where the sector consists of the second centroid,

-In a new set of fourth sequence, sectorized modules other than those freed and confirmed and sub-sectorized modules associated with the confirmed sectored modules, 1 is the second centroid, angle And a confirmed sub-sectorization module including a module that has been determined with respect to the identified sub-sectorization module (the dynamic association method of the module is a dynamic replenishment method)
-The method of failing discovery to the second centroid is stopped and the module is freed,

-1 determined that the zone and center of gravity were associated with at least two modules, the first module operating in this world region TD, the replenishing at least half module operating in space region SD To
-Relevant expressions are Cartesian,
-Relevant expressions are for example X, Y or LTS,
-The centroid corresponding to the couple, the relevant representation is polar (each of the modules of the distance between the centroids s and the logarithm of the distance between the object and the sensor (LD)
-The method can be used for any other type of related expression, eg for a surface:

Colored, separated remotely or for sound:
Harmonic intensity (frequency),
The method comprises at least two orientation devices p (p () of the axis at the input for reference axis rotation of at least two orthogonal coordinates of the input parameter (and the module determining the centroid for the input parameter) For at least one STN module, and including the centroid BarZi by the association of the univariate linear module processing the first parameter, and of the second two-line module processing coordinates Determining the first space Zi is performed within the first space Zi, the second related second centroid determined within the first space Zi, the event of BarZi + 1, the first being divided into distinct corner sectors Space uniformly distributed Zri0, Zri1, Zri2 ... (each section being processed by the two-line module of the sector receiving Zi. Over) a second centroid BarZi + 1 of BarZi, further signals,

The two-line module of the sector corresponding to the second center of gravity BarZi + 1 presence is uniformly Zra, Zrb, Zrc .. in relation to the set of two-line modules of the later rank sector. Sectorization, and at least one more that puts it in order to progressively divide the sector that owns the second center of gravity BarZi + 1 into a clear-angle sector delivered Fine-tuning to determine the angle (substantially the center of gravity s BarZi and BarZi + next to the straight line and the reference axis perpendicular to the direction of the module that merges the center of gravity s BarZi and BarZi + 1 Distance j between 1,
-The method described the features in addition to the angle in it (and the angle determining the value of the projection (C of the line, merging the module's (j, centroid s BarZi and BarZi + 1 on the reference axis) (
-The object has one hand, the logarithm value of the distance LD between the reference points and the different distance on at least one point of the object and its one tool plus at least one size invariant device (received at least at the input And the module (substantially constant projection and distance between two centroids s BarZi corresponding to the angle and BarZi + 1 (at least one) J ”of the two values C (said device determining) (and LD of the relevant ones (and LD,

Calculate the value (LD, ((C () ”,
-One tool of size invariant device means to control the distance, allowing the use of the external measuring distance freely or the internal determination of the distance,
-The object is observed by physical replacement from different distances or by a soaring effect,
-Allows the object to correct the value of said rotation corrector angle which can be observed according to different angles and its one tool and at least one rotation corrector (of the reference axis (( Angle and of previously determined module
-The rotation correction device makes it possible to further determine the angle of rotation ((
-Perform at least one device for transforming the parameter reference system by rotating the angle ((reference at least above the two dimensions of the parameter) of the pixel X (Y) corresponding to the following matrix operation Rotate for 2D reference for polar coordinate parameters:
-Angle (or (or so projection of one of the corresponding on-axis parameters ((p (has a variation ratio that decreases after rotation,
-The parameters are 2 in total and are chosen within a couple X (Y in pixel or LogD coordinates) (pixel distance with respect to reference point and logarithmic and straight of angle with respect to said point) On the reference line

-1 determines over time that several centroids s of the zone appearing with respect to the first zone and its own first centroid store centroidal coordinates in the form of approval data with respect to the label,
-1 performs an analysis tree concatenating different centroids s with respect to their order of events,
-Determining whether an observed new object matches or is stored and does not correspond to an object previously classified as 1 means that the new object and 1 approval data Decide to compare with the approval data of the stored label,
-Further, the approval data of the label is related to the angle of rotation ((by means to authorize to determine the angle of rotation and the average distance LD "(and (the said average distance LD",
-1 is further associated with the dominant color C by means of authorizing the label approval data to determine the color;
-1 associated with the first label approval data to form a new label corresponding to a higher level of approval to those of at least one-half label,
-1 Determine the label and set of data classes by histogram calculation and analyze the authorization data of the processing module that can be stored.
The replenishing method of the present invention also involves a device that is intended to operate according to 1, or some of the features of the previous function related to entering the method.

The present invention is therefore supplemented with histogram calculations and is associated with modules and spaces for cognition-powered devices that process the recognition of objects in the world, with sequence data and time Protested by the parameters carried by the results.
The module is therefore divided into several corner sectors where at least one of the determined zones is centered on the corresponding center of gravity of the zone, and a new center of gravity is found in which of the sectors Is to be sought and because the method can continue for further uplift, for at least one parameter when the sector is divided into several sub-sectors, and Tweaking sectorization consistent with the apparatus of the present invention with regard to putting means to determine under the control of the centroid s to be related according to the application programming interface API zone and related expressions.
(A relevant expression is that any dimension that can correspond to a plane, a quantity or a space of more than four dimensions).

The replenishing device can also be implemented for the following features: And it can be combined according to all technically feasible possibilities:
-There are two sectorization levels in the sub-sector, the first level dividing the first zone into sectors and the second level dividing one of the sectors with new centroids,
-Sectors to fine-tune (sectors containing a second centroid completed without new sectorization) (really, 1 is also used in the first level and / or in case of repetition at each level of sectorization) The first zone to (by increasing the number of sectors being played can be increased in resolution) is limited to one level of sectorization and its one division
-Allows at least one angle to be determined by means at the end of sectorization, and one module, the angle given with respect to a straight line participating in both centroids s and the distance beside the straight The corresponding centroid s of the above-mentioned module between lines drawn (angle and module determinations allow object representation in invariant search in)

-Means enable:
-In the first sequence determined by at least one first module for at least one predetermined parameter, the first zone and centroid cooperated related expressions,
-On the other hand, in order to associate with the first module at least one second module, which is aimed at determining the center of gravity of at least one half of the first zone, in the second sequence, and the other , Set n of the sectorization module was going to divide the first zone into n corner sectors (the sectorization module receiving the first zone and its centroid)
-The centroid of said moment, the third sequence to send to the set of sectorized modules (the second module targeted to determine the centroid of at least one half that had determined the second centroid), and Determine the confirmed sectorization module, where the sector consists of the second centroid,

-A fourth sequence, a sectored module other than the ones that have been freed and confirmed, and a confirmed lower sectorized module containing a second centroid, angle and said confirmed lower sectorized module (module's A dynamic set of sub-sectorization modules associated with the identified sectorization module to determine which module has been determined with respect to the dynamic replenishment method).
-The method of failing discovery to the second centroid is stopped and the module is freed,
-Means determine the center of gravity associated with the zone and at least two modules, the first module operating in this world region TD, at least one-half module operating in the space region SD Make it possible to
-Relevant expressions are Cartesian,
-Relevant expressions are for example X, Y or LTS,
-The centroid corresponding to the couple, the relevant representation is polar (each of the modules of the distance between the centroids s and the logarithm of the distance between the object and the sensor (LD)

-The device can be used for any related expression, for example typing on the surface, coloring, remote, or for sound:
Harmonic intensity (frequency),
-Means having at least two orientation devices p (p () of the axis at the input for reference axis rotation of at least two Cartesian coordinates of the input parameter (and the module determining the centroid for the input parameter) And the means including at least one STN module of it includes the barycenter BarZi by the association of the univariate linear module processing the first parameter and the second two processing coordinates The second related second centroid determined within the first space Zi that allows to determine the first space Zi of the line module of BarZi + 1 events, divided into distinct corner sectors The first space that has been uniformly distributed Zri0, Zri1, Zri2 ... (is processed by the two-line module of the sector receiving Zi S sector) second centroid BarZi + 1 of BarZi, further signals,
The two-line module of the sector corresponding to the second center of gravity BarZi + 1 presence is uniformly Zra, Zrb, Zrc .. in relation to the set of two-line modules of the later rank sector. Sectorization, and at least one more that puts it in order to progressively divide the sector that owns the second center of gravity BarZi + 1 into a clear-angle sector delivered Fine-tuning to determine the angle (substantially the center of gravity s BarZi and BarZi + next to the straight line and the reference axis perpendicular to the direction of the module that merges the center of gravity s BarZi and BarZi + 1 Distance j between 1,
-In addition to the angle (and the module's (j, allows to determine the value C () of the straight line projection merging the centroids BarZi and BarZi + 1 on the reference axis according to the angle by means (

-The object can be observed at different distances and then it is one hand, the logarithm value of the distance LD between the reference points and at least one size invariance on at least one point of the object On the other hand, including a device (at least the size-invariant device receiving at the input), and a module (a distance between two centroids s BarZi of a substantially constant projection and corresponding to an angle and BarZi + 1 (at least J ”(the rotation of the relevant ones (and LD)
-The value (LD, (C () is calculated,
-The size invariant device includes means to control the distance allowing the use of an external measuring distance or the internal determination of the distance at will,
-The object is observed by physical replacement from different distances or by a soaring effect,

-The object can be seen to match different angles, and it enables to correct the value of the rotation corrector angle, including at least one rotation corrector (a set of Reference axis ((angle and previously determined module for value
-The rotation correction device makes it possible to further determine the angle of rotation ((
-It includes at least one device for deforming the parameter reference system by rotating the angle ((reference at least above the two dimensions of the parameter) pixel X (Y) corresponding to the following matrix operation Rotate 2D reference for parameters of polar coordinates of:
-Angle (or (or so projection of one of the corresponding on-axis parameters ((p (has a variation ratio that decreases after rotation,
-The parameters are 2 in total and are chosen within a couple X (Y in pixel or LogD coordinates) (pixel distance with respect to reference point and logarithmic and straight of angle with respect to said point) On the reference line
-To determine that with the time including means, the initial centroid of the first zone and several centroids s of the zone appearing with respect to itself store the centroidal coordinates in the form of approval data with respect to the label Make possible
-The parse tree is performed by concatenating different centroids s with respect to their order of events,

-Determining whether the observed new object matches or is stored and does not correspond to an object previously classified as such, determines the approval data for the new object including means, Allows it to be compared with previously stored label approval data
-It includes means enabling the angle of rotation in the label approval data further to the colleague ((and by means of authorizing to determine said angle of rotation and the average distance LD "(( And said average distance LD '',
-1 is further associated with the dominant color C by means of authorizing the label approval data to determine the color;
-1 associated with the first label approval data to form a new label corresponding to a higher level of approval to those of at least one-half label,
The apparatus comprises means for enabling labeling by histogram calculation for analysis and a processing module capable of determining and storing a set of categorization data of said label;
Replenishing method or apparatus of the invention according to one or some of the previous features. And it can be combined and operates preferably on image video data.

The data space in which the object must be localized is preferably, i.e., it is a space and a realm of the world, but the time and the parameters of the object that expand to other hands are Can be represented, given moment on 1 (space point) or 2 (surface), or 3 (quantity or light or other tone + saturation +) or more dimensions depending on the case.
1 is a replenishing method of the present invention that is essentially using a new STN module that uses the result of the STN module calculation performed to generate a new result during the previous sequence. Understand that it is a general method and that the parameters analyzed by the supplemented module may be any parameters, subtracting only the calculation object and the scene object to be analyzed To do.
The parameters may be, for example, speed, color, coordinates (particularly submitted for rotation to change the reference axis), speed and color, for example to allow localization of objects of constant color motion .

The replenishing method is a dynamic method and, for example, if you are handed over a number of new zones where the criteria are determined, preferably stop replenishing and submit less than For resolution variations where details gradually appear, a second criterion can be implemented) on the threshold.
After a replenishing method of modules completed, possible collection of calculation results, the API allows one to reuse the module in question for other work.
Thus, during the extinction of the element (criteria for quitting the replenishing method), the release of the module containing the included method allows the reused said module.

By way of non-limiting examples, embodiments of the device according to the invention and of the units constituting the device, which can implement the method of the invention, will be described.
Although the present invention will be described with reference to a preferred application for processing video signals, the present invention is not limited to signals of that type. The invention is particularly applicable to signals emitted by synthetic aperture radar.

A first embodiment of the present invention for a device for recognizing objects in the surrounding environment comprises a first set of systems for recognition acquisition and for active image recognition, as illustrated in FIG. The video image sensor 2 is formed. This video image sensor 2 is a part of a camera, camcorder or webcam for observing an object OB substantially placed in a plane 6 and generating a digital signal 7 corresponding to the object OB. CCD type or CMOS type (especially "retina sensor" type in which the pixel density is high at the center and the pixel density decreases as it goes away from the center, and the resolution decreases for a certain period (detailed below) ). A video signal is composed of a continuous sequence (images or frames of a video signal) representing a continuous scene of an object in the surrounding environment, related to the time domain. Each of the sequences is composed of continuous subsequences (video signal lines (scanning lines)). Each of the above subsequences represents a continuous pixel position (forming an ordered subunit) of the video signal, and the continuous pixel position is also related to the time domain. A set of positions (pixels) arranged in a plurality of lines or rows and a plurality of columns forms a matrix, for example a rectangular matrix.

By providing the objective lens 5 for the video image sensor 2, it is possible to change the focal length under the control of the control unit 1, and thus focus on the surface 6 of the object being observed. be able to. Three examples of combination sets of the control unit 1 and the objective lens 5 are illustrated in FIGS. 4a, 4b and 4c described below. The change to the focal length of the objective lens 5 can be controlled by the control unit 1 to change the spatial resolution of the video signal of the object OB to be observed. Effectively, according to the curve of FIG. 3 with the resolution r on the vertical axis and the time t on the horizontal axis, after the preliminary period T0, the resolution is set to its maximum value (at the preliminary period T0) or in the first period T1. optimum base value r _max preferably abruptly reduced to a minimum value or a decrease value r _min from the second period T2 of a single phase t0, maintaining the resolution thereof reduced value r _min, the third period At T3, the resolution is increased stepwise from the reduced value r _min to the optimum basic value r _max . The resolution increase of the third period T3 includes three or more phases t1, t2, t3, t4, t5 of equal duration, and is Gaussian or substantially Gaussian (phases t1, t2, t3 , T4, t5 each of which preferably equals the duration of the frame, but may equal the duration of an integer number of frames). Finally, in the fourth period T4, the increasing process to the maximum resolution r _max is completed. The fourth period T4 includes three or more phases t6, t7, t8, t9, t10 having a duration equal to the duration of phases t1, t2, t3, t4, t5.

From the video signal 7, after the Gaussian transformation (FIG. 1), the Gaussian coordinates x and y (or polar coordinates of the circular matrix of pixels defined by the “retinal sensor”), and As will be described, the synchronization signal Sync (frame (vertical) and line (horizontal)) of the video signal used is extracted.
Further, when the video signal 7 is color, the unit 9 extracts the luminance L sent to the unit 3. If the video signal is black and white, the video signal 7 comprises a luminance signal L.
Unit 3 determines the Gaussian difference between two consecutive sequences of video signal 7. This preferred embodiment of unit 3 preferably performs spatial and temporal smoothing as illustrated in FIG. 9 and described below with reference to FIG.

As will be described below, this unit 3 finally outputs two digital signals, that is, an excess signal DP and an excess quantization signal CO. The continuous overquantization amount signal CO is a function capable of adjusting the time constant of a Gaussian difference, and is a signal representing a recognized object. The signal DP and the value CO can be viewed on a monitor M (for example of a television receiver or computer) and are further represented, for example (without limitation) by reference numeral 8 in FIGS. Processing can be performed by the processing device M ′ shown in FIG. 19 and later and described with reference to FIG. As for the signal DP, it is a binary signal, which takes one of two values 0 and 1, and can inhibit the signal CO, as will be described later.
In the embodiment of FIG. 2, a plane 6 on which the observation object OB is substantially arranged and a unit for extracting the orthogonal coordinates x and y (or polar coordinates) from the video sensor 2, the pixel and the synchronization signal Sync are shown. Has been. For each frame, a synchronization signal at the beginning of the frame and a line synchronization signal are extracted. (If the signal 7 is a color video signal, the unit 9 further extracts the luminance component L from the signal 7 '.) Further in the embodiment of FIG. 3 is provided. The output signal CO can be displayed on the monitor M and can be processed by the processing device M ′. The monitor and processing device are the same as those in FIG. However, the signal CO is not subject to any suppression by the signal DP.

On the other hand, the video image sensor of FIG. 2 is combined with an objective lens having a variable focusing function, that is, a variable focal length function that can be controlled by the control unit 1 to change the focusing, that is, the resolution, The sensor 2 of FIG. 1 is provided with an objective lens 5 ′ having a fixed focal length during the recognition operation and the visual recognition operation. In the latter case, the initial focus is on the object OB that is substantially in the plane 6. Therefore, the signal 7 'is generated with a constant resolution. The resolution can be changed after taking out the coordinates x and y in unit 9 and the luminance L '(if the video signal is a color signal). The unit 9 acts on the (black and white) signal 7 ′ or the luminance L ′ by means of an electronic filter 4 controlled by the control unit 1 ′, and the control unit 1 ′ applies a parameter w forming the filter order. Thus, the resolution is effectively changed according to FIG. The filter 4 generates a signal 7 or L, similar to the signal 7 or L in FIG. The signal 7 or L is processed in a spatial and temporal processing unit 3 similar to that of FIG.

In fact, in the embodiment of FIG. 1, the period of FIG. 3 (after zooming out period T1 of resolution reduction from r _max to r _{min in} FIG. 3 and maintaining constant resolution at value r _min in period T2) Refocusing lens 3 at T3 increases the resolution from r _min to r _max in a Gaussian or pseudo-Gaussian manner, while in the example of FIG. 2 (zoom out in period T1 and maintain resolution in period T2 And after the equal processing, the electronic filter 4 electronically executes the refocusing and equal processing of the lens 5 in the period T3 to increase the resolution.
With reference to FIGS. 4a, 4b, and 4c, three examples of changing the optical resolution are illustrated by various examples of focusing of the lens 5 of FIG.
FIG. 4a shows the surface 6, the object OB, the video image sensor 2, the variable focus lens 5a, and the control unit 1a for adjusting the focal length of the lens of FIG. The control unit 1a acts on the position of at least one of the plurality of lenses of the lens 5a (or the position of the single lens in the case of a single lens), and is schematically shown by a double-pointed arrow in the drawing. As shown, the lens position is changed by displacing the lens in both directions, and the focus of the lens with respect to the surface 6 is changed (the resolution rapidly decreases to r _{min in} period T1 and in period T2. Gradually increase the resolution substantially Gaussian for the period T3 to increase the resolution from r _min to r _max (after maintaining the resolution r _min of

FIG. 4b shows the surface 6, the object OB, the video image sensor 2, the variable focus lens 5b, and a control unit 1b for adjusting the focal length of the lens of FIG. The control unit 1b acts on the focal length of the lens 5b by means of an electronic control current and can adjust the focal length of the lens 5b in order to change the resolution according to FIG. The assemblies 1b to 5b can be of the type described in the following patent document, for example.
French Patent Application No. 2.769.375 filed on October 8, 1997 as French Patent Application 97 12781

Instead of a changeable means to change the resolution of the lens 5b in FIG.
As illustrated in FIG. 4c, a variable focus hollow lens 5c arranged in front of the objective lens 5 ′ and a fixed focus objective lens 5 ′ are combined. The hollow lens 5c is controlled by a control unit 1c that adjusts the supply amount of the transparent liquid inside the lens 5a between the two thin films 51 and 52. The thin films 51 and 52 are substantially parallel to each other in the resting state, and bend and expand when receiving the above-described liquid supply to change the focal length of the lens 5c.
Three means for changing the resolution according to the focal lengths shown in FIGS. 4a, 4b and 4c are to change the resolution of the optical system according to the optical law applied to the lens 5 (type 5a or 5b) or the lens 5c. Or at least pseudo-Gaussian. As a result, during the period T3, the signal L or the signal 7 constituting the Gaussian conversion of the video signal is generated.

FIG. 5 shows an embodiment of the present invention for processing an output signal 7R (similar to the output signal 7 or L in FIGS. 1, 4a, 4b, and 4c) of a radar transceiver 2 'having a lobe and a variable aperture. An application example is shown. The narrowest lobe corresponds to the maximum resolution r _max and the widest lobe corresponds to the reduced resolution r _min (as shown in FIG. 3). The control unit 1d controls the unit 5d with the synthetic lobe of the transmission lobe to adjust the width of the emitted lobes, while normally scanning a rectangular matrix corresponding to this type of video pixel matrix. Adjust. As a result, the signal 7R emitted by the transceiver 2 'inspecting the object OB in the plane 6 is in fact similar to the signal 7 or rather similar to the signal L. 1 can be processed in the same way as signal 7 or rather as signal L. Unlike the signal 7 of FIGS. 4a, 4b, and 4c, the signal 7R of FIG. 5 includes only a single digital component. In the case of the color video sensor 2, the signal 7 in FIGS. 4a, 4b, and 4c includes three color components of color. The signal 7R is rather similar to the luminance signal L of FIG. 1, which is a single component signal. As a result, the unit 9 ′ (corresponding to the unit 6 in FIGS. 1 and 2) extracts only the synchronization signal Sync from the radar signal 7R (in addition to the signals of coordinates x and y not specified in FIG. 5). .

Furthermore, the signal 7R obtained from the output signal from the synthetic aperture radar is actually in the form of a Gaussian transformation or perhaps a pseudo Gaussian transformation.
Similarly, Gaussian processing or pseudo-Gaussian processing is similar to the realization of optical focus adjustment Gaussian or pseudo-Gaussian modification in the case of FIG. 1 (as described below). In the case of 2, the filter 4 performs the electronic filtering operation.
Reference is made to FIGS. 6, 7 and 8a to 8e for the embodiment of FIG. 2 in which the Gaussian or pseudo-Gaussian filter 4 changes the resolution.
The electronic filter assembly 4 forming the filter of FIG. 2 that processes the input signal 7 ′ during the period T3 (FIG. 3) of increasing resolution is from two successive filter units with respect to coordinates x and y. Composed. Each of these units is in Gaussian or pseudo-Gaussian form. In such a case, each unit can be composed of, for example, a canon filter described in the following document.
Didier DEMIGNY and Tawfik KAMLEH “A discrete expression of Canny's criteria for step edge detection performances evaluation”, IEEE Pattern Analysis and Machine Intelligence, volume 19, N-11, pp. 1199-1211, November 1997,

More preferably, for example, the filter can be constituted by a PAOG (Polynomial Approximation of Gaussian) type filter designed by Didier DEMIGNY, JuIien PONS, Nassima BOUDOUANI and Lounis KESSAL. In either case, it is a finite pulse response filter.
The electronic filter assembly 4 performs substantially Gaussian filtering or just Gaussian filtering according to the column coordinate x and the row coordinate y of the pixel matrix. Effectively, this filtering is performed in two stages, preferably according to y in unit 20 and then according to x in unit 21 as illustrated in the block of filter 4 in FIG. .
The filtering unit 20 is preferably of the latter type of those described above and is illustrated in FIG. The filtering unit of FIG. 6 comprises as a building block a single-stage register Re which gives a unit-wise, ie sub-sequence of pixels, ie a delay of one row (line), shown as a rectangular block with a longer vertical side. ing. The filtering unit further comprises two registers A and B, illustrated in detail as "ab", while illustrated as a generally square block. These registers provide a delay of w lines. (W represents the order of the filter and is gradually changed as will be described later.) The filtering unit further includes a subtractor so and a multiplication represented by ordinary scientific symbols “−”, “×”, and “+”, respectively. And an adder ad. The input signals a ₀ , a ₁ , a ₂ of the three adders mu of the subunit 20a are calculated from w based on the equations a ₀ = w, a ₁ = w + 3, a ₂ = 2w + 3. On the other hand, the input signal Cp of the multiplier of the subunit 20c is

Cp = 5 / {2w (w + 1) (w + 2) (w + 3) (2w + 3)}

The signals a ₀ , a ₁ , a ₂ and Cp (normalization coefficients) are functions of only the filter order w and are calculated in the registers A and / or B with the above four equations. Registers A and / or B, as illustrated in detail in “ab” in FIG. 6, have w subsequences or rows (lines) of pixels for a signal w (a variable as described below) received at the input. ) Give a delay.
The set of building blocks described above, i.e. the register and arithmetic operator of the filtering unit 20, comprises three subunits: a non-recursive subunit 20a, a recursive subunit 20b, and a smoothing subunit 20c. Yes. In the subunit 20b, each pair of the register Re and the adder ad placed before the register Re constitutes a loop and constitutes an integrator In. In the subunit 20c, a pair of the register Re and the adder ad placed thereafter constitutes a loop to constitute a smoothing unit Ls.

The filter 20 in FIG. Finally, pseudo Gaussian transformation is performed from the input signal L ′ (or 7 ′) to the coordinate y. This input signal L ′ (or 7 ′) is supplied to the input of the entire filter 4 (FIG. 2).
The unit 21 of the filter 4 that performs the pseudo Gaussian transformation process at the coordinate x is the same as the unit 20 shown in FIG. However, the input is not the signal L ′ (or 7 ′) but the output sy of the coordinate y of the unit 20. The output of unit 21 is a transformation at coordinate y (generated by unit 20) that has undergone a pseudo Gaussian transformation at coordinate x. Furthermore, the unit unit offset of the register does not correspond to one subsequence or row (line) of pixels, but corresponds to the pixel position in the register Re. On the other hand, the offset is the w position of the pixel in the registers A and B. The output of the unit 21 and thus the output of the entire filter 4 in FIG. Or 7).
The pulse response of the filtering assembly 4 (cascaded units 20 and 21) is as follows.

h (w, k) = Cp (w + 2− | k |) (w + 1− | k |) {− 3k ² + (2w + 3) | k | + w (w + 3)}

Cp is given by the above formula according to the y-axis and x-axis of the image.
W constituting the order of filter 4 is

Relational expression σ = 0,3217w + 0481

(If σ is a relatively high value, w is substantially equal to 3σ).
Where σ is the standard deviation of the Gaussian distribution, the number of coefficients of the pulse response of the filter related to resolution, equal to 2w + 1, and the value chosen for Cp is the sum of the above coefficients Must be equal to
In FIG. 7, in the case of a signal 7 ′, ie an output signal from the video sensor 2 of FIG. 2 and a color video signal, the luminance signal L ′ for a particular line (row) of a given frame of the video signal. Probably converted and plotted in the drawing on the abscissa x representing successive pixels of the line. The abscissa x indicates a column number in a rectangular matrix that is a set of pixels of frames arranged in a plurality of columns and rows. On the other hand, on the vertical axis, the signal ni ′, particularly the level ni of the luminance L ′ is plotted. This level has a maximum resolution r _max or base resolution prior to the Gaussian transformation in the filtering assembly of FIGS.

Conversely, in FIGS. 8a, 8b, 8c, 8d and 8e having the same horizontal axis x and vertical axis ni as in FIG. 7, the filter orders at different orders, ie, the filter order, are stepwise. w = 20, w = 16, w = 12, w = 8, w = 4, and therefore increased the resolution (as can be seen with reference to FIG. 3) and after filtering in the filtering assembly 4 Represents the signal 7 or L. In the drawing set composed of the drawings of FIGS. 8a and 8b to FIG. 8e, the order is changed by the law of lowering the order of the filter from one drawing to another.
If FIG. 7 is compared with FIGS. 8b to 8e, before filtering (period T0), the signal 7 ′ of σ = 0 and r _max represents the object observed by the video sensor 2, in particular The level ni of the luminance signal includes all variations (with a very short period). On the other hand, when the resolution is drastically reduced from its maximum value r _max to the minimum value r _min reached at the end of the period T1, the signal 7a in the period T2 when w = 20 (FIG. 8a) Only certain zones are reproduced vaguely. As the resolution gradually increases from the minimum value r _min to the maximum value r _max , the filter order value is changed from w = 20 to w = 16, further to w = 12, w = 8, and finally w = After switching to 4, details gradually appear on the curves 7b to 7e. FIG. 8e is relatively close to FIG.

Conversely, FIG. 7 represents the signal 7 ′ (in particular L ′) at the maximum resolution before filtering, and FIGS. 8a, 8b, 8c, 8d and 8e represent the filtering of FIGS. It represents the signal 7 (particularly L) obtained from the filtering of the signal 7 ′ (particularly L ′) by the assembly 4. Therefore, the output signal that effectively represented the details of the object OB before the resolution change will also cause a sharp drop in resolution (equivalent to unfocusing as expected in the case of FIG. 1). receive. Thereafter (from FIG. 8ak to FIG. 8e) the resolution is increased step by step (as illustrated in FIG. 3) to reach the curve 7e shown in FIG. 8e. Curve 7e shown in FIG. 8e reproduces the first curve 7 ′ of FIG. 7 after some smoothing process. In addition, if FIG. 7 and the set of FIGS. 8a to 8e are compared with FIG. 3, FIG. 7 corresponds to the period T0 in FIG. 3, and FIGS. 8a to 8e correspond to the period T2 in FIG. Corresponding to the phase (or stage) t0 to t4 of T3, until w becomes 0 (zero) corresponding to r _max in phase t5 of period T3, ie corresponding to the resolution before filtering or defocusing. , Gradually decline.

Finally, the output of the filtering assembly 4 denoted by a simplification L or 7 is shown during phases t0 (period T2), phases t1, t2, t3, t4 (period T3) of FIG. 3 (FIGS. 8a, 8b, It consists of a series of signals 7a, 7b, 7c, 7d and 7e (in FIGS. 8c, 8d and 8e respectively).
The signal L (or 7) or 7R, ie the Gaussian or pseudo-Gaussian transformation of either FIG. 1 or FIG. 2 or FIG. 5, is processed by the spatial and temporal smoothing unit 3 schematically illustrated in FIG. The This type of unit 3 is the above-mentioned Patent Document 1 (WO 98/05002) (or its priority document) by the same inventor as the inventor of the present application, in particular page 14, line 24 to page 21, line 19. Reference is made to and described in FIG.
In FIG. 9, the unit 3 includes two subunits 10 and 11. The subunit 10 is a memory, and the subunit 11 corresponds to the subunit 15 shown in FIG. 3 of the above-mentioned document, and executes a spatial and temporal smoothing process. The parameters LO and CO are subjected to a loopback process in the memory 10 after receiving a delay equal to an image or frame in the case of a sequence or video signal, and then re-input from the output of the subunit 11 to the input of the subunit 11. By means of a memory 10 that stores the values of LO and CO during a sequence, unit 11 compares the previous values of LO and CO of the sequence with the current value for each pixel. Where the letters t, t−1,
x and y are time t, time t−1, pixel abscissa x and ordinate y, respectively.
Pointing. t−1 refers to the time before the sequence duration t, so t refers to the current time and t−1 corresponds to the previous sequence.

The recirculation calculation function of unit 3 in FIG. 7 corresponds to the following equation.
LO _{t, x, y} = LO _{t-1, x, y} + (Pix _{t, x, y} _ LO _{t-1, x, y} ) / 2 ^{COt, x, y}
| Pix _{t, x, y} _ LO _{t-1, x, y} |
If DP = 1 and 0 <CO ≤ p, then CO _{t, x, y} = CO _{t-1, x, y} -1
Otherwise,
If DP = 0 and 0 ≤ CO <p, then CO _{t, x, y} = CO _{t-1, x, y} +1

LO _(t-1) represents a substantially Gaussian image before being smoothed in subunit 11 and delayed one sequence in memory 10, and Pix _t is the input signal L (or 7) or the current Gaussian image represented by 7R, thus forming a spatial-temporal Gaussian difference (referred to as DOG). LO represents a continuous smoothing value. Pix represents the value of the radar signal at different positions or in the pixel of the signal L or 7 or in the distribution matrix of pixels. CO and DP are the output signals of the spatial and temporal smoothing unit 3.
In the aforementioned international patent application publication, the input signal is formed by a conventional video signal generated by a video sensor, whereas the signal L (or 7) in FIGS. 1 and 2 or the signal in FIG. 7R is the input signal Pix of the unit 3 in FIG. 9, and the subunit 11 of the unit 3 calculates a Gaussian difference. Unit 3 is for the characteristic parameter of the video signal 7 at the same position of the pixel, for example for the luminance L of the pixel, the value in the frame immediately before the video signal and the value in the current frame. Output signals CO and DP. The adaptation time constant tends to reduce the difference indicated by the above-described formula group. The same applies to radar electromagnetic signals.

The excess signal DP is a binary signal, and the two values of the binary signal are between two consecutive sequences (images or frames), for example at times t = 0 and t = 1. Represents the difference between the absolute values of the pixels in the same position between the images Iα and Iβ in the case of exceeding (exceeding) and not exceeding (not exceeding) the sensitivity threshold. , Indicated by reference numeral 12a in the drawing and larger than the variation of the input signal L (or 7) or 7R due to background noise). On the other hand, the digital signal CO is composed of a small number of multiple bits (unary 2 ^CO , where CO = n, n is the number of bits), and is a signal processed with pixels at the same position. It represents the adaptive time constant that is re-entered into the process to reduce the variation between two consecutive sequences. When DP indicates a value representing an over-threshold value, the value of CO represents a Gaussian difference, and hence an increase in resolution, in successive sequences, as illustrated by image Iγ.
A subunit 14 composed of a module STN (1) of the type described later with reference to FIG. The module STN (1) is calculated in the smoothing subunit 11 (indicated by Dif described later with reference to FIGS. 19, 20, 22, and 23) Pix _{t, x, y} and A histogram of absolute value differences from Lo _{t-1, x, y} is formed. The smoothing subunit 11 is used as described below with reference to FIGS. 10, 11 and 12. FIG.

FIG. 9 shows the apparatus of FIG. 2 with an electronic filter, in the case of the filter orders w = 16 and w = 20, the spatial and temporal characteristics at times t and t? 1, i.e. t = 1 and t = 0. Illustrated are coplanar somewhat blurred images Iα and Iβ converted by different input signals to the smoothing subunit 11. During the period T3 when the order w decreases from 20 to 16 (changes from image Iα to image Iβ), the resolution increases (as illustrated in FIG. 3). When DP = 1 (a value selected to represent an excess threshold), if CO, i.e., n is a slightly higher value in monomial 2 ^CO = 2 ⁿ (where n is a small digital value), it will be described below. A slightly blurred image can be reconstructed. The output image Iγ of the spatio-temporal smoothing subunit 11 at w = 16 and time t = 1 can be displayed on the monitors of FIGS. 1 and 2 and is derived from the value of CO, ie n time. It corresponds to the above sensitivity threshold in absolute value, when CO _{t, x, y} = CO _{t-1, x, y} −1 and DP = 1 | Pix _{t, x, y} − LO _{t-1 , x, y} |
In the case of the apparatus of FIG. 1 in which the resolution is changed by changing the focus, and in the apparatus of FIG. 5 in which the change of resolution is changed by changing the radar lobe, refocusing or opening of the radar lobe. Increase the resolution step by step. The stepwise increase in resolution can be replaced by a stepwise decrease in the value of the order w of the filter 4 in FIG.

The spatial and temporal smoothing unit 3 is accompanied by a control unit 1 (FIG. 1) or 1 ′ shown in detail in FIG. A one-dimensional module STN (1) having a register Reg is provided. The module STN (1) has, for example, a format described with reference to FIG. 20, and the signal Dif = | Pix _{t, x, y} −L0 _t−1 is input data (DATA A) _. Receive _{x, y} | The signal Dif is expressed as | Img _t ? Img _t-1 | (according to FIG. 10), which is the absolute value of the difference between two successive images (Img) at times t and t? 1. Here, the unit value 1 for time corresponds to the duration of one sequence of the signal L, 7 or 7R. The | Img _t ? Img _t-1 | is the difference between two successive images (Img) at times t and t? 1. Here, the unit value 1 for time corresponds to the duration of one sequence of the signal L, 7 or 7R.
The processing of signal Dif in module 14 provides a histogram of this signal as illustrated in FIG.
The control unit 1 or 1 ′ further comprises a control box 15 for the module STN (1) 14. The activation of the control box 15 is controlled by an initialization signal Dep, which is for the module STN (1) 14 under the control of a synchronization signal common to the assembly of the device of FIG. 3 is generated, ie, period selection signals representing periods T1, T2, T3, and T4 are generated.

The control box 15 is controlled by the initialization signal Dep and continuously generates phases t1, t2, t3, t4, t5, t6, t7, and the like. These signals are the continuous values of w in the embodiment of FIGS. 2 and 6 (t1 to t5) or the continuous resolution of the lens 5 of FIG. 1 (in particular the lenses 5a, 5b of FIGS. 4a, 4b, 4c, 5c), or the combined resolution of the radar determines the continuous resolution of the unit 5d which determines the continuous aperture of the radar lobe Lob (FIG. 5). These signals for periods T1, T2, T3, T4 with phases t1, t2, etc. control the operation of module STN (1) 14, and module STN (1) 14 The histogram of the parameter Dif received from the single subunit 11 is determined.

FIG. 10 shows the difference Dif = | Pix _{w0, x, y} ? Pix _{wi, x, y} | (formed in the module STN (1) 14 of FIG. 9) or simply Dif = | Img _w0 ? Img _w i | Here, Img is an abbreviated notation of the image, and w0 and wi indicate the initial value and the current value of w. In FIG. 10, the number of pixels is represented by N plotted in the ordinate, and the absolute value of the difference is plotted in the abscissa. From the curve of FIG. 10, the limit value Li of the difference Dif is for example 90% of this difference point by selecting w = 0 and w = 20, ie the value of w corresponding to r _max and r _min (at least A value equal to 75% can be selected in any case). The difference corresponding to the limit value LI represents the error between the same points of the two images and is not filtered at rank 0 and is the first original image delayed in subunit 10 -Di represents the sequence T0. The second image is filtered with a rank corresponding to Li in FIG. This can finish work on the most stable elements of the image.

The limit Li is determined from the histogram of the distribution of the difference Dif (FIG. 10), which is the maximum produced by filtering with a number of 1 intervals and w _max , such as {Li / threshold> 1}. It is considered that this is a fluctuation. Here, the threshold value in question is the sensitivity threshold value indicated by 12a in the unit 11 of FIG. This threshold is greater than the background noise so that two consecutive stages can be distinguished. The value of p is forced to 1 (number of intervals), ie 5 in the particular example shown.
To illustrate (in the example shown), if LI = 27 (90% of the difference) and the threshold is equal to 5, then 27/5> 5, so I = 5 and p = I = 5. When the fraction of Li / threshold value does not become an integer, it is rounded down to an integer. As a result, in the selected example, there are five stages from phase t0 (period T2), that is, five stages of phases t0, t1, t2, t3, and t4 in FIG. They correspond to the curves 7a, 7b, 7c, 7d, 7e of FIGS. 8a, 8b, 8c, 8d, 8e, respectively. The maximum value of w is, in the specific example, 20 (phase t0), from which the difference in w between two successive stages, Δw = 20/5 = 4, is derived. Therefore, if the maximum value of w is equal to 20, the following parameters of successive frames are used.
t0 w = 20 n = 5
t1 w = 16 n = 4
t2 w = 12 n = 3
t3 w = 8 n = 2
t4 w = 4 n = 1
t5 w = 0 n = 0
n = p Minimum resolution r _min
2 ^CO = 2 ⁿ

Returning to FIG. 9, the Gaussian difference unit 3 has a subunit 16 for calculating the number of steps. The integer value is determined by the default of Li / threshold. LI is supplied from module STN (1) 14 and is derived from the histogram of FIG. 10 determined by module STN (1). On the other hand, the (sensitivity) threshold value is supplied from the location 12a of the subunit 11 described above.
The value of p determined by the subunit 16 acts on the command 17 of the unit 4 (FIG. 2) to determine the order w of the filter 4 and hence the spatial resolution determined by the filter (or FIG. 1). The spatial resolution of the lens 5 is changed, or in the case of FIG.
FIG. 9 illustrates the API that controls subunit 14 of subunits 16 and 17.
In fact, after a sudden decrease in resolution (period T1) switching from r _max to r _min (FIG. 3), during phase t0 (period T2), the input signal to subunit 11 (FIG. 9) is: It is generated from an image of resolution r _min sent by signal 7 (or L) or 7R. It is then entered as a Pix (video signal) or electromagnetic radar type similar signal and by the resolution image r _max before reentering as LO. The calculation of the absolute value | Pix _{w0, x, y} Pix _{wi, x ,, y} | makes it possible to obtain the histogram of FIG. 10 and also Li and LI / threshold and finally the number of steps. That is, it is used at the end of the sequence to determine p (LI / threshold> p).

It is essential to store the image of r _min as a reference during the increase in resolution by forcing p = 0 through the multiplexer 13 during this phase t0. . L0 _{t, x, y} = Pix _{t, x, y} . Here, L0 _{t, x, y} is an image filtered by a filter having a rank corresponding to w = 20, and is an image Iα in the next phase t1.
In the next phase t1 (beginning of period T3) corresponding to the first stage of increasing the resolution from r _min , the multiplexer 13 forces the value p = 5 (the particular example chosen) as the input signal CO. The input signal Pix corresponds to the first stage, and the signal LO corresponds to the previous resolution signal r _min . In that case, as shown in FIG. 9, the output signal of the smoothing subunit consisting of DP = 1 and CO = p? 1 = 4 is obtained.

Thus, the calculation of p is performed in subunit 16, and the value of p so determined is written to 12b and during phase t1, by multiplexer 13 as C0 _t-1 or C0 _to The data is forcibly input to the subunit 11 and transmitted to the subunit 17 on the other hand. In response, subunit 17 provides a sequence of executions of successive steps that increase the resolution from r _min to r _{max in} five steps (FIG. 3). It can be either for the lens 5 to change the focal length (Fig. 1) or for the filter 4 at y for a gradual decrease of w (Fig. 2) or for the radar lobe. This is given to the synthetic aperture radar unit 5d for changing the aperture of Lob (FIG. 5).
FIGS. 11 and 12 show a histogram of the difference Dif values for the corresponding image, one illustrating a still life (FIG. 11) and the other illustrating a human face (FIG. 12)). ing. In both figures, the horizontal axis (horizontal axis) of the histogram represents w (and therefore the level of resolution decreases in the reverse direction), and the ordinate (vertical axis) is the absolute value Dif of the difference (horizontal in Figure 10). As you can see from the axis).

The combination of FIG. 13a and FIG. 13b combined up and down (as specified in FIG. 13) illustrates the sequence of processing steps. The first image at the top left of FIG. 13a (subtitled IMG) corresponds to the signal 7. 'at full resolution r _max before resolution reduction. -Di. It is a complete image that is very easy to read. The topmost image on the right side of FIG. 13a corresponds to the image with maximum defocus (ie resolution r _min ) during phase t0 after applying filter 4 with order w = 20. . This image (which is image Iα at phase t0 or Iβ at phase t1 in FIG. 9) removes most of the details of the previous image IMG.
After a sharp decrease in resolution during the period T1 (FIG. 3) changing to the minimum value r _min of the resolution, details reappear, so at w = 16, the image Iγ of FIG. 9 at phase t1. An image subtitled Iw (16) corresponding to is obtained. In that case, n = CO = 4 (see table above) and DP is equal to 1 in the following images of FIGS. 13a and 13b. (Really, if DP = 0, there is no excess and the signal CO is inhibited, as shown symbolically on the right side of FIGS. 1 and 2.
As w gradually decreases, it switches from 16 to 12, and further switches to 8, 4 and finally to 0 for the image Iw (0) 0, and more and more precise details appear.

Subsequent images Iw (0) 1, Iw (0) 2, Iw (0) 3 and Iw (0) 4 (at the bottom of FIG. 13a and FIG. 13b) are represented by phases t6, t7, t8 and This corresponds to t9, that is, the period T4 in FIG. Since the value of CO is different from zero, the difference between two consecutive images cannot be corrected immediately. After t5, the image gradually fades out from Iw (0) 1 to Iw (0) 4 throughout period T4. During that period T4, the information gradually decreases. This period makes it possible to reversely integrate the previous processing of unit 3.
The following table illustrates the deployment of the system in the sequence of recursive smoothing, ie in the order of frames in the case of video signals.

A, b, c, d, and e on the table correspond to the number of pixels when DP = 1 and n <p. The value a represents a very low spatial frequency filtering process. The values b, c, d, and e correspond to filtering processing that gradually increases the spatial frequency. When t5 (ie, t0 + p) is completed, the maximum information of all stages in the same frame (as shown in FIGS. 15 and 16 and described later) can be obtained.
FIG. 14 shows additional information relating to the above table, in particular the display of the period. In the first period T1, the previous value of n or CO is X, and the numerical value representing the cumulative variation of the pixel in the frame for a, b, c, d, e is from the lower side on the left side to the upper side on the right side. It is increasing.
In FIG. 15, continuous values of a, b, c, d, and e in FIG. 14 are illustrated by curves a ′, b ′, c ′, d ′, and e ′. The horizontal axis represents time t, and the vertical axis represents a number N equal to the cumulative variation of pixels including CO when DP = 1.
The sum of the surfaces under each of the curves a ′, b ′, c ′, d ′, e ′ gives the complete information of the image. The surface under both of the first curves a ′ and b ′ represents approximately half of the total information, so it is possible to maintain only these two curves.

FIG. 16 illustrates an image corresponding to the values of a, b, c, d, and e at time t = p. a, b, c, d, e are listed at the bottom of the corresponding image. Every time we switch from a to b, from b to c, from c to d, and finally from d to e, the details increase in importance, subtitled as shown in the upper left, t = p The image corresponds to the superposition of images a, b, c, d and e, which is similar to the image named Iw (0) 0 in FIG. 13a.
As described above, by appropriately processing the signals CO and DP of FIG. 2, the spatial resolution is increased stepwise after the rapid decrease of the spatial resolution according to FIG. You can get fine details to do.
An example resolution change is realized in the case of FIG. 2 by the filtering assembly 4 (detailed in FIG. 6), commanded by the parameter w forming the filter order. The Gaussian transformation or substantially Gaussian transformation at the filtering assembly 4 is equivalent to the optical defocusing realized in the embodiment of FIG. Accordingly, the luminance L signal 7 and the digital signals CO and DP are essentially the same as in FIG. 1 and FIG. 2, and therefore substantially the spatial and temporal smoothing of FIG. 1 and FIG. The same image as in FIGS. 8 to 16 can be obtained at the output of the unit 3. Therefore, those descriptions also apply to changing the spatial resolution by defocusing optics and then stepwise focusing (rather than electronic filtering) (eg, in five steps). The same is true when the optical sensor of FIGS. 4a to 4c is replaced with a radar type electromagnetic sensor according to FIG.

When optical refocusing is performed following defocusing according to FIGS. 1 and 4a, 4b, and 4c, it is very difficult to perform refocusing in a clear step shape according to FIG. In order to improve this disadvantage, it is effective to combine a known electronic shutter (not shown) with the lens 2. Stepwise refocusing similar to the refocusing obtained by electronic filtering according to FIGS. 2 and 6 by operating regularly but keeping the refocusing substantially constant for a period of time Can be realized. Remove the STN module calculation cycle and open the shutter. This will be described later with reference to FIG.

Thus, FIG. 1, FIG. 2, FIG. 4a, FIG. 11 with reference to FIG. 3, FIG. 7, FIG. 8a to FIG.8e, FIG. 10, FIG. 11 or FIG. 12, FIG. The operations performed in the devices of the embodiments of 4b, 4c, 5, 6, and 9 can be combined.
Beginning with the start signal Dep (shown in FIG. 3 at the beginning of period T1 and shown in FIG. 9 at the input of module 15).

In the first period T1 (for example, lasting for one sequence (one image or one frame)), the signal L, 7 or 7R is reduced very rapidly in resolution from r _max to r _min . On the other hand, for example, w switches from 0 to 20 (FIG. 3) and the spatio-temporal smoothing unit 3 (FIG. 1 or FIG. 2 and FIG. 9) stores an unfiltered image of the period T0. .
The following period T2 consists of a single phase t0 having a duration equal to, for example, one sequence, and so on for the phases t1 to t9 of the periods T3 and T4 (FIG. 3). During this phase t0, the resolution of the signal L, 7 or 7R is kept constant at the value r _min . For example, w is equal to 20, and the value p = 0 is forcibly input to the subunit 11. The subunit memory 10 of FIG. 9 stores a signal of minimum resolution and generates a signal L0to therein. For a variety of different values at pixel matrix locations x, y (similar locations in the case of scanning radar type electromagnetic signals), represented by rank 0 unfiltered images, Accordingly, the signal L, 7 or 7R (image L (7) _{20 in} FIG. 13a) is input to the unit 11. Unit 11 generates a signal Dif for STN module 14, which creates the histogram of FIG. 10 and determines the value Li sent to subunit 16 at the end of phase t0. To do. The subunit 16 has an integer (default) p = int (Li / threshold) obtained by dividing the value of p, ie, Li, by the threshold value from the threshold value << threshold >> (where the symbol int is shown in FIG. 9). Deducting the integer value of the quotient described in (). For example, p = 5, and this value is input to the subunit 11 during the following phase t1.

At the beginning of the following period T3, ie at the beginning of phase t1, the resolution of the signal L, 7 or 7R is increased by one level and the normal routine operation of the spatiotemporal smoothing unit 3 is Beginning, the subunit 11 calculates and generates LO, CO, DP and Dif. During this phase t1, the incoming image of this subunit is image Iβ (FIG. 9). This image is calculated from the input image of the previous phase corresponding to the image with the lowest resolution (ie image Iα (FIG. 9). Conversely, CO (DP = 1 The output image represented by) is image Iγ (FIG. 9), or image Iw (16) (FIG. 13a), and contains few details other than the most typical details. During the following phases t2, t3, t4 and t5, processing continues at w = 12, 8, 4 and 0, and on monitor M, the images Iw (12), Iw (8) of FIG. , Iw (4), and Iw (0) 0, respectively, and during phase t5, w = 0 and the resolution reaches its maximum value r _max .
During the subsequent phases t6, t7, t8, t9) of the subsequent period T4, the resolution is maintained at its value r _max , while in the embodiment of FIG. Maintained equal (FIG. 3) or the focal length of the lens (in particular 5a, 5c, 5b) is refocused on the surface 6 of the object OB, or the control unit 5d To the minimum aperture corresponding to refocusing or w = 0. Since the processing of unit 3 continues and its CO only decreases gradually towards that unit value, the index n of 2 ⁿ gradually increases from 4 to 3, then to 2, then to 1, and finally Decreases to 0, which outputs the images Iw (0) 1, Iw (0) 2, Iw (0) 3 and Iw (0) 4 of FIG. 13a (last row) and FIG. 13b. .

At the end of the period T4, i.e. phase t9, the filter 4 or lens 5 or Rob Lob returns to a stable state with a maximum resolution r _max and a new cycle starts again with a new signal Dep according to FIG.
The simplified embodiment of FIGS. 1 and 2 of the apparatus of the present invention has been described above, but hereinafter, with reference to the subsequent figures, in particular FIGS. 17 and 18, downstream of the temporal and spatial smoothing unit 3. A specific embodiment is described which implements the histogram calculation STN module of FIG. 14 located (added to module STN (1) 14 of FIG. 9). FIG. 17 is based on FIG. 1, and a unit 8 is added. FIG. 18 is based on FIG. 2, and similarly, a unit 8 is added.
17 and FIG. 1, and FIG. 18 and FIG. 2 are different from each other in that the signals CO and DP from the temporal and spatial smoothing unit 3 described above use STN type modules described later. To be processed in an assembly 8 that performs histogram formation and classification.
In the case of the embodiment of FIGS. 1 and 2, the temporal and spatial smoothing unit 3 illustrated in detail in FIG. 9 comprises STN type units, whereas the embodiment of FIGS. Dif is processed to obtain the histogram of FIG. 10 rather than the signals CD and DP of subunit 11.

Needless to say, the embodiment of FIGS. 4a, 4b and 4c is also applied to the apparatus of FIG. 17 similarly to the configuration of FIG. 1, and the embodiment of FIG. 6 is also applied to the apparatus of FIG. 18 similarly to the configuration of FIG. Applied.
The other parameters processed in each of the units 8 are two position parameters x and y, and three parameters relating to color, namely luminance L, tone T and saturation S. These five parameters are retrieved by unit 9 'in the same way as unit 9 of FIGS. 1 and 3 (in fact, only parameters x and y are retrieved as described above). Obviously, instead of parameters L, T and S, a set of other three parameters, for example, parameters representing the basic colors red R, green V and blue B, or luminance L Along with preserving, along with T and S, color can be represented. It can also represent a color with parameters CR corresponding to LR and parameter CB corresponding to LB.

Preferably, when parameters L, T, and S are selected, even if the lighting conditions change, parameters T and S do not actually change, and only the luminance L changes directly depending on the lighting conditions. There is. This means that only one parameter is modified. And as will be described later, it is advantageous with respect to the other two parameters. When using three parameters of red, green and blue format or L, LR and LB formats, the three parameters are modified when the lighting conditions change. In fact, essentially only using L in the first processing phase in assembly 8 while processing other parameters of color, ie both T and S obviously have simultaneous x and y) Used to maintain the << fixed >> image obtained
Additional parameters at the input of the assembly 8 arise from an external device 25 (not shown) that facilitates object recognition by the STN module described below by size invariance calculations. Equivalent to measuring distance between object OB and sensor 2).
The output of assembly 8 (consisting of an STN type module) identifies the nature of the object (signal WHAT or LABEL) and position of the object (signal WHERE or Z0) that generates two groups of signals enable.

Assembly 8 it belongs to Fig. 17 or Fig. 18 (that is by optical means (Figs. 1 and 17) or electronic filtering (Figs. 2 and 18), both with the same kind of resolution increase Whether it was done by Gaussian or substantially Gaussian for the period T3, the following explained from the resolution variation that it was the same as described above) For the signal 7, essentially the same type of luminance L (which is spatially and temporarily processed in this device) and delivering the goods of FIGS. 17 and 18 The temporal and spatial smoothing unit 3 signals the same natural CO and DP in the case of FIGS. 17 and 1 and in the case of FIGS. 18 and 2:
The signal CO (which is the adaptation time constant of the monomial 2 ^CO ) is actually a digital signal n formed of a small integer number, but DP is by definition an over-threshold accumulated in the value signal CO It is a binary signal illustrating (sensitivity threshold).
The assembly 8 of FIGS. 17 and 18 is a single-line or one-dimensional type, a two-variable linear or two-dimensional type, and possibly a three-line or three-dimensional type, a unique parameter. A single-line module processing CO, a bivariate linear module processing two parameters, eg pixels following a matrix column and lines x and y are pixels Or the coordinates entered in the table of radar signals and, finally, the STN of a three-variable linear module processing three parameters (eg three parameters defining color) The module is formed, and in the latter case, these three parameters are conveniently bright L, tone T and saturation S.

Examples of one-variable one-dimensional modules are described and illustrated in the publication WO-01 / 63557, which is the same inventor, and examples of two-variable linear and three-variable linear modules are described. French patent application entitled "Method and apparatus for locally identifying an object by its shape, its dimensions and / or its orientation", filed February 23, 2001 and published as FR-28221459 01-02539 is described and illustrated on page 6, lines 18-23 for bivariate linear modules and on page 6, lines 1-7 for trivariate linear modules. A bivariate linear module or a trivariate linear module, according to the present patent application comprising two or three univariate linear modules, provides an output signal with an AND gate. And the output constitutes the output of a pseudo bivariate linear module and a pseudo trivariate linear module, respectively.
It is possible to use this kind of unilinear, bivariate linear or trivariate linear module to form the device 8 of FIGS.
However, it is preferred to implement the preferred embodiment described below with respect to the univariate linear module with respect to FIGS. 20 and 23a-23b respectively (symbolized 19 in the figure, emphasizing STN (1)). ), And a two-variable linear module (emphasized STN (2), symbolized 22 in the figure), a two-variable linear module, by adding a few, a three-variable primary The formal module (STN (3) highlighted and no example given) comes from very easily (which explained below).

Referring to FIG. 20, for the case of a single parameter, the unilinear or one-dimensional module STN (1) essentially consists of three devices (ie the histogram calculator CH, the classifier CL and the It can be known that it consists of a back-up device RA).
Histogram calculator CH is the number of all digital analysis memories 100, for example several subunits of DRAM or SDRAM with many address equivalents and the first input parameter DATA (A), i.e.CO and number of pixels. For the value of the width of the word representative, there will be a number of possible levels chosen by the frame (or image) (ie 18 bits for an image of eg 256.000 pixels).
This memory 100 includes three inputs: input 100a DATA IN for data, input 100b WR writing, and input 100c ADRESS for address.
The input of the data 100a is 106a when the initialization signal INIT applied to the input 106c of the multiplexer 106 is equal to << 1 >>, while the output signal 107s of the adder or the increment confirmation increment S 107, and On the other hand, it is supplied from the multiplexer MUX 106 of the data input received by 106b (signal << 0 >>).
The input 100b that the memory 100 writes is the order in which the gate OR 113 and WRITE write, both inputs each receiving the initialization signal INIT (and this signal acting as the aforementioned on the multiplexer 106). The output is received from the signal.

Furthermore, the adder / subtractor 108 is controlled with respect to its input 108c, the signal (DIRECTION) by the signal << 0 >> or the signal << 1 >>, i.e. commanding addition, for example << 0 >> and << 1 The subtraction is instructed according to the direction configured by either of >>.
Finally, the adder S 107 mentioned at its first input 107a that it received the signal 100s from the output OUT 100d of the memory 100, and its second input 107b (referred to as -On the confirmation signal 102s) from the druck device RA.
The output 107c from the adder S 107 generates a signal 107s equal to the output of the analysis memory 100 (signal 100s) when the verification signal 102s is equal to << 0 >>, but the verification signal 102s is << 1 >>. The equality to the output increased by 1 o'clock.
The output 107c of the adder 107 is the P (ie P4) of the multiplexer 106 and the comparator 140 (which compares this signal 107s with the output signal 1044s of the register RMAX 1044) as its input, as already defined. 106a to the input received at Q allows POSRMAX (RMAX position) to be inserted into register 1045 by the signal 140s gate connected to the input inferred from this comparison (if P> Q) 141 under the control section of AND signal WRITE.
The histogram calculator CH of module STN (1) (it was just described) finally generates a signal (ie 100s) by the memories 100, 105s and by the multiplexers 105 and 107s. The adder 107 generates the three outputs generated towards the classifier CL and receives the signal 1045s therefrom from the register 1045 of the device.

Furthermore, the device CH receives the signal 102s from the feedback device RA.
For classifier CL (which constitutes a passive classifier), the former consists essentially of operational subunit 101 (a classification update subunit 103 and a storage subunit 104 containing several registers).
The operating subunit 101, which is part of the classifier, consists of two comparators 110 and 111 that receive at their input P (P1 (P2)) the signal CO that constitutes the input data. , Their inputs Q (Q1 (Q2)) from the respective registers 1042 and 1043 (the values of the CO classification limit A and the classification CO limit B, respectively) are described below.
Each of comparators 110 and 111 compares received value Q with received value P.
If the received value P1 or P2 is greater than it was received at Q1 or Q2 (or perhaps equal for P1 and Q1, respectively), each comparator signals the input or the gate's b (after inversion for the latter) AND 112, FIGS. 17 and 18, the signal DP specifying the threshold exceeded by its value 1 or its value of this temporal and spatial smoothing unit 3 It also receives at its input 112c from the spatiotemporal smoothing unit 3 smoothing this time of a non-exceeded threshold by zero.
Gate output signal 101s AND 112,
At the same time as sub-unit 101 and therefore only at the same time, DP = 1 (being better than the others in device 114 of FIG. 9) (P1> Q1 and P2 <Q2) is bus 114 (some STN modules -In addition to the common).
Classifier CL's classification update subunit 103 consists of:
The selectors 1152 and 3 input b and c receive outputs from the registers 1044, 1041 and 1046 of the subunit 104 (ie, RMAX, NBPTS and THRESHOLD), respectively. And that is mentioned below.

The signal at the output 1152s of the selector 1152 is the application Q (designated Q3) of the comparator 1151 (which therefore receives either RMAX or NBPTS), or THRESHOLD, for selection, the signal is input at 1152 Received.
The other input of the comparator 1151 (referred to as P3) is connected to the output 100d of the memory 100 for receiving the signal 100s therefrom.
Comparator 1151 checks whether the value of the signal on input P3 is greater than or equal to the value of the signal on input Q3 (P> Q).
In such a case, the output signal 1151s is applied to the inputs and also their inputs c (after inversion for the second) (signal DIRECTION (it is also the adder / subtractor 108 on its input 108c) 2) logic automaton 1063 and 1064 received on their input b (signal END) at the end of the operation.
The fourth input d of each logical automaton 1063 and 1064 performs the same function as the function represented by the signal 1151s received at that input that can receive the reset signal RESET. The truth of enabling (P> Q) (ie P3> Q3).
The output e of the logical automaton 1063 and 1064 confirms (enables) that it generates the signals 1063s and 1064s, respectively, for the limits Limit A and Limit B, respectively, as described below in register 1042 And 1043.

The third subunit 104 of the classifier CL identifies several registers, first of all for the maximum value RMAX, the register 1044, the position associated with this maximum value POSRMAX Register 1045 consisting of register 1045, the number of points NBPTS and finally register 1041 for THRESHOLD register 1046, registers 1044, 1041 and 1046 generates its output signal 1045s on the input 108b of the adder / subtractor 108 of the histogram calculator CH In contrast, as mentioned above, multiplexer 1152 is generated. At that time, 109 of the gate AND histogram calculation device CH receives the signal RESET. Subunit 104 also consists of the two other registers mentioned above: 1042 for limit A and 1043 for limit B.
Each of both of these registers receives that input generated by En (enables or confirms or authenticates) (confirmation signal 1063s) is output by 1064s logical automaton 1063, 1064, respectively. And the output signal 105s of the multiplexer 105 of the histogram calculator CH on its input.
The feedback that the device containing the set 102 essentially registers 102r registers connected to the bus 114 whose input is common to several devices of type STN highlights 1, 2, 3 ... (M), it receives the value in1 (in2) that in3... Inm of the output signal 101s of the subunit 101 from different devices STN is connected to the bus 114.
For each register 102r, the device RA performs a comparison of input values such as m (eg, m having the value contained in the corresponding register) to determine the Boolean expression 102.

の適用によって、逆のケ−スの0に等しい信号102s以外のヒストグラム計算装置CH（1にレジスタ102rに含まれる値が対応する入力信号に等しい各時等しい確証信号102s）の加算器107の入力107bに送信するレジスタ102rの前の内容を示している登録。
好ましくは、図20の一次元装置STN(1)は、また、インプットされたデ−タCOの平均の値代表の位置を格納するためにレジスタ1047および1048から成ることができる。この平均の位置POSMOYは、このデ−タ項目COの値の重心への道において一致する。入力、示す、レジスタ1047の、と、POSMOYOが注意した、CHがもう一方がインプットしたのに対して、レジスタ1047の、bと称した装置が受信するヒストグラム計算のマルチプレクサ105によって、分割の後の反転（入力Q5が信号NBPTS（入力で受信される信号の中の信号選択によって選ぶ、b、そして、マルチプレクサ1152のc）を受信するコンパレ−タ1153の出力信号1153s）の後、2時までに発生される信号105sを受信する。

Is applied to the input of the adder 107 of the histogram calculation device CH other than the signal 102s equal to 0 in the opposite case (the confirmation signal 102s equal to 1 each time the value contained in the register 102r is equal to the corresponding input signal). Registration showing previous contents of register 102r to be sent to 107b.
Preferably, the one-dimensional device STN (1) of FIG. 20 can also consist of registers 1047 and 1048 for storing the position of the average value representative of the input data CO. This average position POSMOY coincides in the way to the center of gravity of the value of this data item CO. Input, indicating, register 1047, and POSMOYO noted that the other channel was input by CH, while register 1047, the histogram calculation multiplexer 105 received by the device named b, By 2 o'clock after inversion (input Q5 is selected by the signal selection in the signal received at the input, b, and the output signal 1153s of the comparator 1153 receiving the multiplexer 1152 c) The generated signal 105s is received.

The other input P5 of the comparator 1153 receives the output signal 1155s of the register 1155 performing the accumulation from the acceptance of the initialization signal INIT until the acceptance of the signal at the end of the output calculation END at the adder 1156c. The output signal 100s of the memory 100 that is given an input, and the input b receives the digital signal 1155s of the output of the register 1155 (and hence the running sum of the signal 100s).
When P> Q (actually P5> Q5), the first register POSMOYo 1047 in which the comparator 1153 generates an output signal 1153s to be applied after being inverted to the input. An additional second register POSMOY1 1048 until the arrival of the signal END on the other input b of this register 1048, the output signal 1047s of the register POMOYo 1047 is received at the input of the data. Register POSMOYO 1047 determines that the current average position (POSMOY) of NBPTS / 2 and NBPTS resulting from the comparison of the comparator 1153 of the value accumulated in signal 1155s also showed the center of gravity. , Register POSMOY1 1048 stores the previous average position (or centroid) energized by the signal 1047s generated by register POSMOYo 1047.

It must be emphasized last that registers 1047, 1048, 1042 and 1043 contain each additional input supplied by the signal CLOCK-PIXELS. All these registers therefore operate in synchronous mode.
Expose the operation of the mono linear module STN (1) of FIG.
With respect to the operation of the single line block STN (1) illustrated in FIG. 20, it is illustrated in FIGS. 23a-23b for each successive sequence, three consecutive phases or cycles, ie The operation of the bivariate linear module STN (2), as well as that of the three-line type module not illustrated, it updates the initialization cycle, the calculation cycle and the registers storing the results. -Consists of a cycle for
Currently, the one-variable linear STN (1) in FIG. 20 was input while limiting to the module case, ie consisting of one variable. And it is in a specific case CO, 3 cycles mentioned above is as follows:

1) Initialization cycle At the beginning of this cycle, the module STN (1) receives an external sequencer signal INIT which is not represented.
This registers registers POSRMAX 1045, RMAX 1044, NBPTS 1041, and registration 1155 to signal INIT to 0.
b) Start a counter COUNTER that scrolls from 0 to M (ie the maximum size of the memory 100) with successive values.
c) Multiplexer MUX 106 (which receives << 0 >> on its input 106b and INIT on its input 106c) (input 100a DATA IN of memory 100) to force << 0 >> .
d) The memory 100 for writing the mode is set by the 100b WR input to the memory 100 by the gate OR 113.
And
e) Move the output signal 105s to the address of the memory 100, depending on its input 100c (the consecutive value of COUNTER (not illustrated) by the reverse multiplexer MUX 105) of the given multiplexer MUX 105.
2) The calculation cycle signal INIT has stopped (ie at least after the end scrolling the duration of the counter COUNTER (not exemplified from 0 to M)). Once so, the signal WRITE initiates the computational cycle it contains for each pixel cycle (CLOCK-PIXEL):

a) Each time the confirmation signal 102s is true, the memory 100, the latter being transmitted, as described in relation to the structure of the module, communicates to the device 102 (memory 100 (on its input 100a) by the multiplexer 106). The sum of the values of the CO of STN (1), with the output OUT of the CO of the adder 107 being stacked).
b) A comparison of the value RMAX of the comparator 1044 with the value of the signal 107s coming from the adder 107 of the comparator 140, if this signal 102 is true.
The output of the comparator 140 is true (the value of the signal 107s), it constitutes the new value of RMAX (since it has been demoed by comparison of the comparator 140 before it is entered into the register 1044) Big) register RMAX 1044 engraving AND 141 under the effect of command from the gate, while receiving this output of the comparator 140, and on the other hand (signal WRITE), it is the signal The value of 107s, the value of signal 105s that executes the address of register RMAX 1044 (ie the location or address corresponding to this new value of RMAX) is entered in register POSRMAX 1045 (and also connected to this register 1044) In contrast, AND 141 is input to the register POSRMAX to the output of the gate.
At the same time, the value entered in the register NBPTS 1041 is incremented by +1 by << 1 >>.

c) Calculation of the classification of the value CO, which is output 101s each time this value CO is made between the limits A and B of the device 101, in the true value.
3) Result register update cycle that signal WRITE is equal to zero (equal to 0), and signal END is in operation to multiplexer MUX 105, while signal RESET is equal to << 0 >> Later cycle start.
The register update cycle consists of two consecutive half cycles.
3.1) First half cycle. And it thinks POSMOY includes the following sequential actions:
a) The value of POSMOY0 in register 1047 is moved to register POSMOY1 1048.
b) The command Choice is added to the selector 1052 and sets the register output d to the value NBPTS / 2.
c) The value of the counter COUNTER is transmitted by the multiplexer 105 due to its input 100c to this value and also for the command END and address arriving at this multiplexer (and hence the memory 100).

d) The scan by signal COUNTER is registered in register 100, thanks to its initial value made equivalent to 0 and by the adder 1156 receiving in addition to signal 100s (output 1155s of register 1155). Extract all stored values 100s accumulated in registration 1155.
e) 1155s in register 1155 is less than or equal to NBPTS / 2, comparator 1153 (which compares both these values) is as long as the value of the output signal identified by its output 1153s, register POSMOY0 1047 Write the value of signal 105s.
3.2) The second half cycle calculates the limits of classification (ie Limit A and Limit B) from the maximum position POSRMAX.
This second half cycle performs the following operations:
a) Counter COUNTER is reset to zero.
b) The signal RESET is acknowledged, thus forming the output signal 1045s of the register 1045, and the value POSRMAX is conveyed to the adder / subtractor 108 by the gate AND 109.
c) Adder / Subtractor 108, subtraction of POSRMAX-COUNTER (the first value arriving at 108b and second N 108a), and finally the signal DIRECTION resulting from this subtraction forming the output signal 105s of multiplexer 105 command.

d) The counter increments the device signal COUNTER by +1, and the output at 100d of memory 100 is the comparator 1152 (output d) (the comparator that generates the output signal 1151s to the logical automaton 1063 and 1064). This signal is valid as long as it generates a signal 100s that is greater at 1151 than the output signal of the comparison being performed.
e) Confirmation of writing the value of output signal 105s of register Limit A 1042 that the logical function of confirmation 1063 is true and arrives at the input.
f) As soon as output signal 1151s is no longer valid, method of operation c, d no longer mentioned above, and the value of Limit A is fixed to the last value entered in register 1042 It remains.
g) Counter generating signal that COUNTER is reset to 0 (not represented).
h) A signal that DIRECTION forces the adder's POSRMAX + COUNTER adder / subtractor 108 (the result of this addition, which finally forms the output signal 105s).
i) Signal that RESET is confirmed.
j) While the output 100d of the memory 100 generates a signal 100s larger than the output signal (output d) of the multiplexer 1152, the counter increments the signal COUNTER for incrementing the input ADRESS 100c of the memory 100 by one unit. . The comparison is performed in the comparator 1151.
k) The logical function of validation generated by device 1064 (ie, signal 1064s) is true. Then, it confirms the writing of the value of the signal 105s of the register Limit B 1043.

And
l) k is interrupted as soon as the output signal 1151s of the comparator 1151 is valid, method of operation i, no longer in j, and the value of Limit B is set to the last value entered in register 1043 It remains fixed.
Once these three cycles have been accomplished, an external sequencer (not represented) can resume a new set of the aforementioned three cycles.
A histogram of the values of the input parameter CO (which is the output parameter of the smoothing assembly 3 in the space of FIGS. 1 and 2 and this time and detailed in FIG. 9) is 1 Or it can consist of several peaks. With respect to FIG. 21a, while representing a histogram of the value of the parameter CO having a single peak P, FIG. 21b represents a histogram having two peaks P1 and P2.
With respect to FIG. 21a, if the histogram of CO values has only a single peak P, then the histogram (and then the horizontal of these values in the ordinate, the first of all RMAX, i. For the value of CO on the axis and frequency Q, plotted from the position of RMAX), you can define RMAX / 2 (ie half RMAX) and finally that of POSRMAX. And that is, the horizontal axis of the peak is P.

On the other hand, the values of RMAX and, on the other hand, POSRMAX, are stored in corresponding registers (ie, 1044 and 1045) as previously described.
RMAX / 2 defines the limits where A and B are representative in FIG. 21a (both of these limits stored in registers 1042 and 1043, respectively).
These limits are also added to the comparators 110 and 111 for comparison to the input parameter CO.
The comparators 110 and 111 of the aforementioned assembly 101 and the gate output signals 101s AND 112 can be shown as illustrated in FIG. B. Including (positive) notches between.
With respect to FIG. 21b, the case was represented when the histogram of CO values included two peaks P1 and P2 (instead of one peak P in FIG. 21a).
At the highest point, as in the case of FIG. 21a (RMAX) on one hand and on the other hand POSRMAX, P2 matches. Furthermore, from RMAX, RMAX / 2 is defined as the case of FIG. It differs from FIG. 21a in that it defines a maximum value that is also the RMAX referenced to P1 of the peak.

With respect to FIG. 21b, two pairs of limits Limit A, Limit B and Limit A ′, Limit B ′ are shown at the bottom of this FIG. 21b as (positive) notches in the signal 101s (a). Instead of defining only two limits of RMAX / 2 (ie Limit A and Limit B), the value defines that is,
Since POSRMAX (which corresponds to the highest peak P2) is determined, it is possible to move from the signal 101s (a) within the framework of the present invention and the smaller amplitude peak. In the case of Fig. 21a, the first notch corresponding to P1 and only so far the notch corresponding to P2 and limit A and limit B defined by RMAX / 2 A signal 101s (b), which is depicted in detail, is obtained.
It is finally 101s of that variation b (it is communicated to bus 114 if it is 2 or several peaks), this signal being a signal (it has the same shape as signal 101s in Figure 21a) Defined by limits A and B corresponding to the highest peak P2 RMAX.
After the description of the structure, in operation and in the results of the module STN (1) (FIGS. 19, 20, 21a and 21b), we have been schematicized in FIG. 23 and shown in detail in FIG. For the module STN (2) illustrated in 23b, the following similar explanation is provided.

The STN (2) detailed in FIGS. 23a-23b consists of a module STN (1), essentially three devices, namely a histogram calculator CH, a classifier CL and a feedback device RA, Occurs on bus 114, which is common to several STN modules. In addition, in the case of module STN (1), a module that STN (2) can be added to the register is conveniently provided in the classifier CL and the input data (module In the case of STN (2) 2 input parameters), the average value representative of the devices 1156, 1155 and 1153 will be described later for storing the positions.
With respect to the input 105a of the multiplexer 105, since the input of one input signal DATA (A) (ie, CO) is not provided, the histogram calculator CH of the register STN (2) in FIG. Different from the corresponding device CH of STN (1), but in the two input signals accompanying DATA (A) and DATA (Ax), for example by the device 9 of FIG. 1 or from the input signal respectively by the device 9 of FIG. Cartesian coordinates y and x extracted 7 (after defocus) or 7. '. (Similar to defocusing before processing of filter 4 performing actions). For the sake of explanation, the processing of 2 is the module STN (1) in FIG. 20, while the left shift and other hand canceling the device 130 (multiplication) receiving the signal END (1) ( In the case of the module STN (1) (FIG. 20) of the adder / subtractor 108 to the input signal COUNTER, the signal COUNTER is directly added without offset on the input 108a of the adder / subtractor 108, whereas the signal 130s on the input 108a. In addition to the signal MOYy on the association of both these signals (adding the equivalent of the offset to the maximum number of bits of the signal DATA (Ax)) to produce The parameter was entered instead of the device.

In module STN (2) (FIGS. 23a-23b), classifier CL includes a fixed number of overlaps for corresponding devices in module STN (1) of FIG.
In particular, subunit 104 contains a set of registers 1042 and 1043 for Limit A and Limit B, not 2 for one hand and Limit Ay (related to input parameter y) And a set of registers 1042 (2) and 1043 (2) for Limit By (also related to the input parameter y) (these sets of both the receiving registers) 1042 (1) and 1043 (1)) for the top registers (ie Limit Ax (related to input parameter x) and Limit Bx (also related to parameter x)) , Each paired memory input signal (signal 105s generated by multiplexer 105).

Because of the set of overlapping registers 1042 and 1043 for limits A and B, respectively, both automaton 1063 and 1064 in module STN (1) in FIG. If the shock is absorbed, and with respect to FIG. 23b, this type of automaton (ie, 1064 (1) and 1063 (2) and the input parameters corresponding to 1063 (1) and the input parameter x -Illustrated two sets of 1064 (2)) corresponding to y.
Similarly, set 101b, module STN (2) to subunits 101 (1) and 101 (2), subunit 101, each of these subunits 101 (1) and generating signals. 101 (2), one signal 101 (s) generated by the module STN (1) of FIG. 20, with a portion similar to that of 101 (1) s and 101 (2) s respectively. The motion subunit 101 of FIG. 20 absorbs the shock.
Signals 101 (1) s and 101 (2) s are produced at the input and gate b AND 131, which simply means that signals 101 (1) s and 101 (2) s are simultaneously connected to this gate 131. The output signal 101s is distributed only for cases that arrive at both inputs.
It is this signal 101s that is generated on bus 114. Predicted block 1049 (Fig. 23b) is assembled into its input (output from register POSMOY0 1047 equal to POSMOYxy related to the two input parameters x and y) and POSMOY1 1048 and the expected single line Completes the description of FIGS. 23a-23b (one criticism that the former includes) in addition to FIG. 20 for a simple module STN (1) means (determines Δx and Δy representing the values of registers 1047 and 1048 in the response) .

Registers POSMOY0 and POSMOY1 deliver a calculation module 1049 that calculates the variation of each POSMOY between two consecutive frames, as well as for DATA (Ay) (DATA (Ax)).
The change in POSMOY for parameter y is denoted by Δy, and the change in parameter x is denoted by Δx. Δy and Δx are calculated in the calculation module 1049, but are produced in two subtractors 145 and 146, respectively. Receive these values on their inputs b and receive DATA (Ay) and DATA (Ax) on their inputs. Subtractors 145 and 146 (which calculate DATA (Ay) -to-y and DATA (Ax) -to-x, respectively) and hence the predictions based on the linear variation of DATA (Ay) and DATA (Ax) is generated concomitantly to produce classification signal 101s in device 101b. The output POSMOYxy of the register 1047 is applied to the input b of the comparator 1050 (FIG. 23a) which determines the equivalence () between POSMOYxy and the input parameter DATA (Ax). The comparator 1050 generates an output signal 1050s, BarZi (BarZ, which is an abbreviation for the centroid of the zone) and i representing the current value of a portion of the zone. A description of this additional set of 1049 and 1050 and the nature of the signal 1050s is provided below.

The operation of the module STN (2) of FIGS. 22 and 23a-23b takes the STN (1) of FIGS. 19 and 20 with differences after the module operating mode.
Essentially, the STN (1) of cycle 1) module initialization is thus reserved for module STN (2). Conversely, calculation cycle 2 and the classification phase are separated. Finally, DATA (Ax) and DATA (Ay) are subsequently processed to update the resulting system (from cycle 3).
1) Initialization cycle This cycle effectively consists of the arrival of signal INIT, operations b, c, d and e are detailed above for the operation of module STN (1).
2) The calculation cycle module STN (2) receives the input parameters DATA (Ay) and DATA (Ax) on the input 105a of the multiplexer 105.
As a single entered the data item DATA (A) (ie, CO) of STN (1) in FIG. 20, it processes these data items DATA (Ay) and DATA (Ax) in the same way.
Indeed, in both embodiments (STN module (1) and STN (2)), at the input parameter CO or input, the parameter DATA DATA as its signal 105s to the input 100c ADRESS of the memory 100 (Ay) and DATA (Ax) are sent by the multiplexer 105.

Gate 101 and finally subunit 101 (1) from subunit 101 (2) that produces in the case of simultaneous occurrence of both signals (global signal applied to 101s bus 114) 13 output signals 101 (1) s are combined.
3) Processing the data DATA (Ay) in the cycle replicated block to update the resulting register is a logical automaton 1063 (1) and 1064 (1) and Perform those with subscript (1) actions prepaid (ie, registers 1042 (1) and 1043 (1), Limit Ax and Limit Bx for limits)-device 101 (1).
The input parameter for which the method results in that of the second input data item y (ie DATA (Ay), the latter being processed in the same way) to this conversion from the data item x by the device 108a. Subscript (2) to use in the block that is replicated to complete the method of data x (ie DATA (Ax)) (ie registers 1042 (2) and 1043 (2), logical automaton 1063 (2 ) And 1064 (2) and operational subunit 101 (2)).
Cycle start after signal WRITE becomes equal to zero and signal END equals << 0 >> while signal END (1, 0) becomes active in multiplexer MUX 105.
The update cycle for this register consists of:

3.1 Calculate POSMOY in the first 3 minutes of the cycle using the actions a to e described above.
3.2 Calculate the classification limits Limit A (x) and Limit B (x) from the partial DATA (Ax) in the second third of the cycle, with operations a to l described above.
3.3 The third third of the cycle calculates the classification limits Limit A (y) and Limit B (y) from the partial DATA (Ay). Device 130 is operated in association with commands MOYy and END (1) and operations a through l described above.
FIG. 24 represents two dimensions with x and y, namely DATA (Ax) and DATA (Ay). Figure 21b, which also illustrates the same with two peaks V1 and V2, surface in a single dimension with x (ie CO, curve with two peaks P1 and P2) . With reference to FIG. 24, an example of the results obtained is illustrated for the two-dimensional module STN (2) of the type illustrated in FIGS. 23a-23b. Notice that the 3D Cartesian system follows the coordinate (x) and (y) DATA (Ax) and DATA (Ay) and the number or quantity Q of the point with the coordinate (x) Continue (z) and (y). For the highest peak V1, RMAX (POSRMAXx and POSRMAXy and limits Limit Ax, Limit Ay, Limit Bx, Limit By) were represented. Conversely, for simplified objects, these values for peak V2 were not represented. The execution of POSRMAX / 2 (not represented) in the case of FIG. 21b makes it possible to remove the V2 of the second peak because it only keeps V1 of the main peak.

FIG. 23c is a module STN (2 of FIG. 23a-23b of the device CH of FIG. ), Generating its output signal 120s on the input << 1 >> of the multiplexer 106 and receiving the first input of the memory 100 and the output signal 100s on the second input The coefficient Km illustrating the partial variation of block 120 is conveniently constructed with an index of 2, ie, Km = 2 ^m .
This calculation block 120 acts on the coefficient Km (added to both its inputs) with the signal 100sOUT of the memory 100 to give a transfer function

を計算する。この見地から、バッファ・メモリへのモリ100の内容が転送され、バッファ・メモリへの伝達関数

Calculate From this point of view, the contents of memory 100 are transferred to the buffer memory and transferred to the buffer memory.

To calculate the new s value in memory 100 at the new address corresponding to the first address offset from the expected value calculated by (Δx or Δy) module 1049. The content has been rewritten. Therefore, when rewriting the transfer function result, the contents of memory 100 were moved to buffer memory and not directly overwritten to memory 100, so as not to overwrite the value of memory 100 (possibly useful for later calculations). Rewrite. Memory by adding the transfer function to its contents other than storing it in the buffer memory which is then moved to completion after offsetting the address, as previously determined the result It is also possible to perform calculations directly from 100. Understand that these operating modes running buffer memory can use completely clear memory, or at that time instead address pointer (circular block buffer) Have been replaced. In that case, the expected value and hence the place where you can execute the procedure to rewrite the result when running, without using any buffer memory to not overwrite any result. It must be emphasized lastly that the test on the predictions in the direction of the expectation and itself can increase or decrease the address by +1. In the case of a positive prediction, examine the memory 100 from top to bottom, and if the prediction is negative, in both of these cases, the result will be the result of examining and rewriting the memory 100 from the bottom to the top. Also do not overwrite.

The coefficient is an information hold parameter. The larger the value m, the more amenable to storing the histogram in the memory 100. Conversely, when m = 0, K ⁰ = 2 ₀ = 1 and Km−1 = 0, so that the output of the block 120 is equal to << 0 >>. During the input initialization cycle of multiplexer 106, the case illustrated in FIG. 23a is constantly equal to << 0 >>.
When m = 0, the computations in which each cycle stored in the memory 100 is recalculated are arranged in turn. Conversely, when m is different from << 0 >>, the percentage of memory 100 storage (which is more or less high according to the value of m) is reserved for the ongoing computation cycle.

The apparatus of FIG. 23c shows an improvement over that of FIG. As described above for the first cycle (ie, the initialization cycle (INIT = 1)) of FIG. 23a, it is true if the memory 100 is reset by the multiplexer 106. It then zeros its input 100a for DATA IN.
Conversely, the signal << 0 >> on the input 1 of the multiplexer 106 whose memory effect has the arrival of the output signal 120s of the device according to FIG.

Added to block STN (2) by exchanging corrected arrivals.
However, the transfer function capability of block 120 increases the length of the word stored in memory 100, which oversizes the word length for memory 100 of FIG. 23a.
Indeed, for example, if it is a memory, 100 of Fig. 23a is a ready-made 10-bit word, and the coefficient is Km (applied to block 120 of Fig. 23c), 2 ^m (Fig. 23c 100 of which can change 1 to (the amount of memory that has to be sized to store (10 + m) bit words).
The signal Km, with respect to the importance of the learning method desired for this module, is generated by a register 121 commanded by the API of the module STN (2).
The addition of register 121 illustrated in FIG. 23c for block 120 and for two-dimensional module STN (2) is also applied to mono-dimensional module STN (1) as shown in FIG. It should be noted that it can be done.
Finally, on the mono-linear and bivariate linear STN module, the device 8 of FIGS. 17 and 18 can be used as a three-line STN module, eg, a parameter LTS (or RGB). Can be made up of modules for three parameters representing

The trivariate linear module is similar to the bivariate linear module of FIGS. 23a-23b (possibly with the improvement of FIG. 23c). Besides that,
Instead of giving one offset of the histogram calculator CH of register STN (3) described in this module STN (2), two offsets commanded by block 130, i.e. the second and third It is competent for the item DATA (three DATA are DATAL, DATAT and DATAS, for example).
In the devices indicated by (1) and (2) in FIGS. 23a-23b, the devices 101 (1) and 101 (2), 1063 (1) and 1063 (2), 1064 (1), and 1064 (2) A third device similar to 1041 (1) and 1041 (2), 1042 (1) and 1042 (2) is added.
* The calculation of the value POSMOY and the prediction and classification are tripled instead of doubled.

The operation of the three line STN module is similar to that of the bivariate linear module illustrated in the figure at intervals 23a-23b. One processes three parameters with three assemblies (1), (2) and (3) instead of two assemblies (1) and (2). (Three parameters such as L), T, and S, which are subsequently processed in the same way as the parameters x and y in the bivariate linear module, are subsequently processed. Therefore, obtain Limit AL, Limit BL, Limit AT, Limit BT, Limit AS, and Limit BS. Finally, the global signal at the output of assembly 101 of FIGS. 23a-23b consists of signal 101s, 13 input with AND 2 generated by the output of the gate (applied to bus 114) (101 ( 1) and receiving the output of 101 (2)), the output signal of the three-dimensional module STN (3) consists of the signal generated by the output. An additional device 101 (3) for receiving the output of the 3-input AND gate device 101 (1), 101 (2) and similar to the devices 100 (1) and 101 (2) (illustrated (Not shown).
FIG. 25 illustrates the solution of improving the classification of parameter DATA having a number of bounds with expectations (any in total). With respect to FIG. 25, the largest part of the device STN (2) of FIG. 23a-23b or 23a-23c was implemented.

FIG. 25 is a diagram with the exception of device 101, which is replaced with device 101c configured of memory 118 (equal in number of words to memory 100 other than 1-bit word width and by addition of 2-input multiplexer 144). Identical to 23a-23b.
The multiplexer 144 is commanded by the operation end signal END. When signal END is equal to zero, the results of devices 145 and 146 are output. When the signal END is equal to 1, the signal 105s is output. Multiplexer 144 is a memory 118 where the input DATA IN is commanded as output 105s of memory 100 at the input Q, from the register and at the output of comparator 1151 receiving half RMAX at input P. The output signal 144s on the input ADRESS is generated. This comparator 1151 compares the input signal on its input P and that of its input Q, and if P> Q, the comparator 1151 stores the memory 118 and equivalent to the opposite case zero. Generates a signal on DATA IN (equivalent to 1). For anticipation, it is this signal on the output that the OUT of the memory 118 that then constitutes the output signal 101s has been applied to the bus 114.

The memory 118 of this module 101c is written during a cycle to update the register result (signal END equal to 1). For each address in memory 100 and hence in memory 118, the operation of multiplexer 144 causes a comparison to be made between memory 100 and the threshold value defined by command CHOICE and written to memory 118. Defined by the value of the bit If the value of the memory 100 is the reverse case and is greater than or equal to the threshold 0, it is equal to 1. The command to write for the memory 118 is defined by the signal END.
Classification with prediction is performed by the operation of multiplexer 144 (which sends the result of parameter DATA (which is adjusted as the address of memory 118 by prediction)) during the calculation cycle, the output OUT of this memory is Define the classification values.
Apart from that, the implementation of the improvement of FIG. 25 lead to similar results, despite improvements, is obtained by the implementation of the means, illustrated in FIGS. 23a-23b or 23c-23b .

At the beginning of this sequence, for example, two peak events during the same sequence followed by a sudden increase in resolution in phase t1 (FIG. 3), a one-dimensional histogram (peak In the case of two-dimensional histograms (peaks V1 and V2) in FIG. 21b (already considered) and in FIG. 24 (also already considered). These peaks correspond to two classes: the value of CO (FIG. 21b) or DATA (Ax) and the histogram classification of DATA (Ay) (FIG. 24). The one-dimensional module that allows the dominant class to be extracted by the STN (1) of FIG. 20 or the two-dimensional module STN (2) of FIGS. 23a-23b is determined by the P2 or V1 of the peak. Each protested.
For example, several peaks or classes occur simultaneously. Then, generally speaking, there may be global movement for the motion parameter MVT, or the effect of the lens on the scene in a whole scene, for example travel or zooming, Analyze the observed scenes related to movement or deployment and deployment points (especially in the case of an onboard camcorder of a running vehicle).

If at least two peaks appearing during the same sequence are considered, extract more from a given class during this sequence to save processing time, It is advantageous to preserve the dynamics of these classes of events.
A simplified solution is the third sequence in determining a new dominant class, removing this class, and prohibiting this new dominant class by suppression, During the new dominant class and others, in the second STN module during subsequent sequences in the decision to do the third STN module up to the last class (after said suppression) It is essential to determine the dominant class (peak P2 or V1) of the first STN module during the first sequence.

This type of device processes, for example, motion MVT and ST1 (ST2 and ST3) of a one-dimensional STN (1) and includes three STN modules according to FIG. Illustrated in 12). These modules receive parameter Z on all of their inputs. Whereas t2 is the output of module ST2, the second dominant class representative of MVT1 and the output signal MVT1 of module ST1 (representative of the first dominant class) output signal MVT2 is sent as suppression ST2 established at the time to signal the second and third suppression, the time when tb was input as the suppression signal on the second suppression, ST1 (the first dominant class of MVT Representative) MVT sent output signal inputs ST3 b and c generating MVT third dominant class output signal MVT3 representative.
Unfolded over several phases t0, t1, t2 (FIG. 3), this kind of method unfortunately requires a lot of processing time.
The answer to increasing speed is illustrated in FIG. So, during phase t0, the one-dimensional module STN (ST'0) that extracts the set belongs to the signal MVT during cycle CALCUL, and is different from each other during cycle RESULT. After the quantitative order and rewrite being reduced, extract them and rewrite them to the accompanying STN module. The first dominant class known is transferred to the second one-dimensional module STN (ST'1). The sorted values of the histogram memory 100 are described to create new values NBPTS and resets that are then transferred. This second block ST'1 (which generates a signal MVT1 representing a new dominant class MVT) makes it possible to determine a new dominant class. The transfer operation of the new result of the classification and of the new register 104 of the device ST′0 is resumed at this time by inserting the module STN (ST′2). It then generates MVT and other new dominant class MVT2 until the memory 100 histogram values are exhausted. Thus, with reference to FIG. 27, a third one-dimensional module ST′3 that generates a new dominant class MVT signal MVT3 representative is illustrated.
Another means of accelerating the determination of several classes seen in the same sequence consists in localizing these classes to determine the level of pixel quantity from the maximum value.

This type of method (illustrated in FIG. 24, where the types DATA (Ax) and DATA (Ay) have at least two peaks V1, illustrated in FIG. 24, is an inclined topography with V2 having at least two vertices. As the resolution grows on the stage, it can be compared) (added to the bilinear histogram covered by the water layer). The level then decreases to expose at least two vertices for a certain degree to lower the water level. Indeed, for a given moody water level when two or several vertices appear simultaneously, it must be classified by decreasing the height. This illustrative comparison allows a better understanding of the method of sequential class extraction detailed below.
For example, in FIG. 23a-23b, which is conveniently modified according to FIG. 25 in a bivariate linear module, the classifier 118 is a 1-bit memory (0 or 1) such as A (parameter STN (2) follows the pattern that a histogram containing several peaks is stored in the memory 100, whereas a value 1) corresponding to an overvalued threshold by value is constructed Determined by LIMIT Ax, LIMIT Bx, LIMIT Ay, LIMIT By.

When the parameter value A or B exceeds the provided threshold at the output of the selector 1152 (FIG. 20, 23b or 25), such as RMAX / 2, the classifier 118 is << It can be known that 1 >> occurs.
In order to extract several classes during the sequence, it is sufficient to place RMAX by decreasing the value of register 1044 to store the threshold A / 2, and B / class 2 Provide a register for the number of classes in which the threshold is stored.
Therefore, the following pattern is executed:

The processing software consisting of three consecutive cycles (ie initialization) reduces the RMAX and sorts and arranges the decreasing value of the RMAX in the memory 100 contents for storing the address to be placed. To do that is to perform a computation that is feasible and update the resulting register.
1. The initialization cycle consists in resetting all memories M0, M1, M2, M3 and register Rc.
2. The calculation cycle consists essentially in determining and storing the memory M0 whose histogram values are determined by the module STN.
3. The resulting register update cycle whose flow chart is illustrated in FIG. 28 includes three consecutive subcycles A, B and C.
That is, during the resulting register update cycle (illustrated in FIG. 28), the classes are the same in some classes (eg, one or several frames or video images). -When it appears during
The first subcycle A is first specified by (M1) = Sorting (M0) on Figure 28 (memory M1 receives the result of sorting in memory M0) of these decreasing values Sorting the method is performed by decreasing the value of the histogram M0 and the RMAX contained in the address memory in the memory M1 at consecutive addresses as described.

The second subcycle B searches for and defines different classes of histograms contained in the memory M0. It includes several stages, ie continuously
i. On the other hand, an initial initialization step 500 that resets the pointer p that scrolls by successive addresses from 0 to n included in memory M1 (running through i and j), and, on the other hand, the number of classes Register Rc, where the numbered class was found previously (p = 0 and Rc = 0).
ii. For each of these values (three consecutive actions in FIG. 28, ie, block 501) for each of these values, the loop continues to display all the values of p, up to n included to operate. Second step to
a) Position determination At the pt address p (501a, FIG. 28), a pointer equal to the content to sort the memory M1, ie pt = (M1) p (that is, the content of the memory M1 at the height of the pointer p).
b) Judgment of a value A equal to the content of the memory M0 of the histogram at the address · pt (501b FIG. 28) (ie A = (M0) pt)

And
c) A collection of values labeled a, b, c (or a, b...) adjacent to a value selected in the memory of class M2 at level or address pt. That is, in the case of a univariate linear histogram (2 positions), pt + 1 (a, b), in the case of a bivariate linear histogram (8 positions), pt _x + 1 (a, b...), Pt _{y +1 (a, b..} .), and the case of a three-variable linear equation histogram (26 positions), in short the case of 501c on FIG. _{28, pt x +1 (a,} b...), pt y +1 (a, b...), pt _z +1 (a, b...). Indeed, in the case of the one-dimensional method, there are two locations pt + on both sides of pt. For the 2D method, there are 8 locations pt +. In other words, it forms the periphery of a large square of 3x3 centered on pt. For the 3D method, there are 26 locations pt +. That is, it forms the periphery of a 3 × 3 × 3 cube centered at pt.
iii. The third testing step 502 is missing (<< YES >>) or has adjacent values (the absence of these values corresponds to the case where a = b = 0) Decide.
iv. Fourth step including two possible ways:

First example:
Test 502 revealed that there are no adjacent values of << YES >>) such as a, b... In this case, the automaton 504 operates in the following sequence of operations.
a) The number of registers Rc is incremented by +1 by << 1 >>. It is conventional Rc ++ (operation 504a (FIG. 28)).
b) The accompanying threshold A / 2 is an address defined by the value of the register Rc (operation 504b (FIG. 28)) and is stored in the memory M3 (FIG. 28).
c) The class defined in the register Rc (504c, FIG. 28) is written at the address defined by the pointer position pt in the memory of the class M2.

Second example:
Test 502 revealed that there was at least one neighbor value.
In this case, the automaton 503 operates in the following sequence.
a) Logic operations 503a, 503b and 503c choose the cheapest price (not zero) of the class among the different values of the adjacent class (a or b).
b) This value is entered in the class memory M2 at the address defined by the value pt (501a, FIG. 28). On the other hand, writing of << a >> different from zero is distinguished from << b >> other than zero (operation 503d or 503e).
At the end of subcycle B, the pointer parameter p is incremented by +1, regardless of the result of tests 502, 503a, 503b, 503c, by the unit (p ++ of operation 510).
The third sub-cycle C updates the STN modules required for the discovery of at least two classes known during the same sequence. And that is illustrated in Figure 28 (this subcycle with several steps).

i. The first steps include:
Action 505 that imposes the value of the register of the current class CI-Actu on << 1 >>.
(Following steps included, each other fitted into two loops, ie 508 and 509, later existence made the former).
ii. The second step consists of the initial operation of loop 508 from 1 (ie expanding all known classes Cl-Actu) to the final class defined by the class number register Rc.
iii. The third step is performed by the automaton 507.
Assign a module (sequence number equal to the value of the current class Cl-Actu) to the new STN.
A determination threshold value corresponding to the contents of the memory M3 is determined at an address defined by the current class Cl-Actu.
Then, the pointer p is reset to-<< 0 >>.
iiii. To determine if the first action is now in memory M2 and the content copied at address p of the pointer (actions 503d, 503e, 504c, FIG. 28) is equal to the value of Cl-Actu (action 509a) The fourth step is performed by a loop 509 that is intrinsic to performing the test.

If this content is different from the value of Cl-Actu, the automaton resets the content of the memory 118 of the STN that the module previously selected (operation 509b).
If the content is equal to the value of Cl-Actu, test 509c determines whether the content of reading of the histogram M0 at address p is greater than the threshold value, and in this case one bit of the STN module allocated The contents of the memory 118 are adjusted to << 1 >> at the address p.
In the opposite case, this content is adjusted to << 0 >> (operations 509d and 509b, respectively).
iv. The final step increments the address of pointer p (operation 509e) << 1 >>. When this address exceeds n (operation 509f), Cl-Actu (operation 511) is increased by << 1 >>. Then the loop 508 can be implemented around.
Finally, instead of the previous value of Cl-Actu, a new value of Cl-Actu is entered at the input of loop 508, ie towards operator 506.
It allows the execution of the software of the three subcycles A, B and C, which is the device of FIG. In fact, this software runs a minimum number of processes equal to n. [Log (n) + Rc + 1]. If n is high, the number of methods can be very large and therefore relatively long. In this case, it is preferable to extract one class.

To overcome this difficulty, an electronic device can be implemented that allows the sorting of classes during an update calculation cycle. In this case, memories M0 and M1 are ready at the end of this calculation phase, and the resulting display cycle software then consists of subcycles B and C and only needs a short duration. is there. And it is very advantageous.
The minimum number of methods is equal to:
n (Rc + 1).
FIG. 28a implements the flow chart of FIG. 28, which is added as an integrated API with the object of maximum speed of execution optimization.
Subcycle A is integrated into the calculation phase and will be described in a later stage.
Included in apparatus 600 is a flowchart start at the beginning of a direct subcycle B cycle for searching and defining different classes of histogram M0 and updating the register results. Obtained by the previous calculation phase, it was associated with sorting M1. This subcycle B is

Figure 28 includes the same steps defined except for the following details:
* In initialization step 500, the flag BUSY is checked,
* Operations a) and b) of block 501 in FIG. 28 are replaced with the following operations.
* The address routing command Sel..och.p is disabled, allowing access to device 600 by decoding its address 601 of rank p (514, FIG. 28a)
* A pointer position pt equal to the contents of the sort memory M1 is determined as an address p, that is, (POSRMAX) p (501a ', FIG. 28a).
* Determine the value A equal to the contents of the memory M0 of the histogram at address pt, ie (RMAX) p (501b ′, FIG. 28a).
And
* Enable address rerouting Sel..μp to access different memories with the same address field (p equivalent to DATA (A)).
Serving as a classifier for the calculation cycle, the third sub-cycle C updates the memory M2. This sub-cycle C in FIG. 28a includes the same steps defined in FIG. 28 except that step 509b is canceled, and steps 509c and 509d in FIG. 28 are replaced with steps 509c ′ and 509d ′.

Test 509c 'determines whether reading the contents RMAX of histogram 600 at address p by multiplexer 105 is greater than a threshold value, and in this case M2 is done through multiplexer 105, the flag on which Val is imposed and all The memory reading operation (509d ′, FIG. 28a) is rewritten at the same address.
The routing command Sel.μp is disabled (512 forms 28a) at the end of subcycle C (class Rc to be copied), and the flag BUSY (513 forms 28a) is canceled The
In this case where the memory M4 is dedicated to each listed class value NBPTS, it may also be advantageous to know the number of pixels belonging to each class known.

Performing and completing the flow chart of FIG. 28a, adding both next steps in subcycle C, and updating the value NBPTS corresponding to the location memory M4 at address CI-Actu Is done.
* Initialization of memory M4 at address CI-Actu in step 507 (Figure 28a)
* Total execution of this position memory M4 valid value RMAX at function YES for output 509'c of test:
(M4) _CI-Actu = (M4) _CI-Actu + (RMAX) _p

The numbering of the elements to be classified (NBPTS) means that the addition of sub-cycle D causes the center of gravity (POSMOY) of each class to be placed on the flowchart of Figure 28a at the end of sub-cycle C. Taking. Allows computing and completing in this subcycle D linked to each class previously defined in the two additional memories used, temporary calculation memory MT and the result Is a memory M5 that keeps POSMOY.
-Subcycle D start by resetting memory MT and MS and pointer p.
-A loop that calculates the value POSMOY for all values in n from p to 0 follows.
The first action of this loop is to read the value of the class Cl in the memory M2 at the address p and when checking whether the associated flag Val is present.
[Val, Cl = (M2) p]
If this flag is present, the automaton checks whether the contents of the temporary memory MT at the address Cl are less than the contents divided by two of the memories M4 at the address.
[(MT) _Cl <(M4) _Cl / 2]
If the previous test is validated, it does the running sum of the value histogram RMAX at the address p of the memory MT.
[(MT) _Cl = (MT) _Cl + (RMAX) _p ]
Then, the pointer p is stored at the address Cl in the memory POSMOY M5.
[(M5) _Cl = p]
-The final step is to increase the value n by +1 by the pointer p and the address when resuming the loop as long as this address << 1 >> or equivalent.

In the case of multivariate linear histogram calculation, the pointer p represents a different field of the data item DATA. The previous calculation loop defines a part of the POSMOY field by the part of the leftmost field of the value p. It is necessary to restart the loop method, replacing the field with value p as many times as the number of fields defining data item DATA (A).
For example, for a bivariate linear data item DATA (Ay, Ax)) it scrolls the loop twice, initially scrolling p from 0 to n, and the result is Corresponding to POSMOY (y), the second time replaces both fields of p that scroll from 0 to n, and the result then corresponds to the part x POSMOY (x). The combination of both these fields corresponds to the center of gravity POSMOY (x, y).
We have just described the method, and the associated devices enabling several classes and their centroids in the extract associated from multi-line parameters.
Examples that do not sort classes during a calculation cycle according to the above-described method on Figures 29a and 29b (Figure 29a, which summarizes Figures 23a-23b), An example (FIG. 29b) relating to

The modification of FIG. 29b with an integrated sort corresponds to the assembly of a single device 600 of memories M0 and M1 in one device 600. This device M0-1 includes on the opposite side of RMAX and POSRMAX with respect to the values 0 to p of the pointer. This is a decreasing order of RMAX from position 0 to position n.
Therefore, POSRMAX and API 602 are also directly connected to device 601 (maximum value of histogram (RMAXn)) (A = (RMAX) p operation 501b ′ (FIG. 28a)) and its actual position (POSRMAXn) (pt = (POSRMAX) p command the algorithm to read at address p (between 0 and n) decoded by operation 501a '(FIG. 28a)).
The content RMAX can also be accessed by its location POSRMAX using the multiplexer 105 (instructed by the signal Sel..och.p). And it directs the value of the address ADRES-p of the device 600, the output signal OUT of the device is then input to the device 602 by the multiplexer 654, which is commanded from the device 602 by the signal Sel..och.p issue date. Aim for -p direction. This operation is performed at position p with a preset threshold value during the test of value RMAX (operation 509c ′ (FIG. 28a)).

In the calculation cycle, the memory that M2 supplies as many class classifiers, the signal Sel..och.p existence then invalidates the signal 651s with the signal Val (active for the defined class) The multiplexer 105 transmits the signal DATA (A) as an address to the generated memory M2. The signal 651s passes through the demultiplexer 653 that is confirmed from the memory M2 by the signal Val issuance date at the input En. The device 653 generates a signal Cl1 on the membership class Clk representative of the input signal DATA (A).
The assembly 603 of the devices 600 and 601 is illustrated in more detail in FIG.
Although FIG. 29b is represented in FIG. 30 developing this FIG. 29b in more detail, FIGS. 31 and 32 are parts of FIG. .. Illustrates (any of the more detailed trends) of any of Bn owning input RMAX and input POSRMAX). Obviously, the only type of assembly for which it is possible is to use only one set of FIG. 32 (without multiplexer block MUX at the input) for the device B0, including the block memory 600 illustrated in FIG. part.
First of all, it should be noted that B0 receives as input for FIGS. 29b, 30 and 31 for the reference.

* Signal IN 107s (adder 107 output signal). The two inputs receive the validation signal VALIDATION and one of the output signals OUT of the blocks B0, B1, B2... Bn.
* Signal of address ADRES-p from API 602 by automaton 601
* Calculate signal CALCUL,
* Start signal INIT from an external sequencer (not illustrated),
* The value of the signal at the output input IN and (ADR-in) of multiplexer 105, which serves as the signal at address ADR-in for block B0, B1, B2 ... Bn, is the register RMAX in M0 (Figure 31). Stored in 704 and POSRMAX 705.
The input of signal IN 107s, corresponding to the new running sum, is identified by comparator 706 at inputs P and Q, the first and second receiving new signal IN. If only P> Q in register RMAX 704, ie, the value of the new signal, then IN will be greater than RMAX already stored since the new higher value of RMAX was engraved. A new RMAX write is performed by gate AND 707. The gate AND 707 generates a write signal WRITE that is applied to record the input En of the RMAX register 704 during the calculation cycle (signal CALCUL).

At the same time, under the control of signal WR from gate 707 during the write calculation cycle, the address of the new RMAX is written into register POSRMAX 705 by its input En.
When the output signal OUT of the device M0 is the input En of the device 708 including the gate, it is equal to << 0 >>. Check whether this input is confirmed, while it is equal to the contents of register RMAX.
Confirmation is achieved by a signal 709s that forms the output of the comparator 709 that compares the incoming signals.
(a) Address ADR-in, and the contents of register POSRMAX (b), the presence of this confirmation signal 709s emits comparator 709 by the side only = b, that is, ADR-in is effective if it is a signal Represents the address or location of RMAX (ie POSRMAX).
Block B0 in FIG. 31 also has three inputs that receive the calculated signal CALCUL of toggle 711, the inverted comparator 709 output signal 709s, and the comparator 706 output signal 706s after quitting the signal RESET. A gate AND 710 having

Finally, device M0 outputs RMAXout, POSRMAXout, signal ao to subsequent block M1, in addition to its output signal OUT to adder 107 (FIGS. 29b and 30). When signal IN is greater than the contents of register RMAX 704, signal ao is true. Finally, the transmission signal TRo is equivalent to << 0 >> for processing the first pixel of the sequence. When the signal 706s coming from the comparator 706 is equal to 1, ie ao = 1, and the signal ADRES is different from the contents of the register POSRMAX 705. The transmission signal TRo is true when the signal 709s represents a ₌ b.
Signal INIT triggers signal RESET. The signal RESET resets and initializes registers RMAX 704 and POSRMAX 705, and initializes the output of toggle 711.
With regard to the synchronization of the operation of the device M0, the synchronization is ensured by the rising leading edge of a CLOCK (not illustrated) pulse defining a pixel sequence cycle.
The four output signals RMAXout, POSRMAXout, ao and TRo of the device M0 are applied to the input to the next device B1.

With respect to FIG. 32, a device Bi representing one of the devices B1 to Bn.
The device Bi receives the command signals IN, ADR-in, CALCUL and INIT (from which RESET is derived) as the device M0, and further receives (RMAXin) (consisting of RMAXout of the previous device) from the previous device B. , Ai-1, TRi-1, and POSRMAXin (consisting of POSRMAXout of the previous device). M1 receives the output signals RMAXout, ao, TRo and POSRMAXout of M0, and the device Mi receives the corresponding output signal of the device Mi-1.
Device Bi has two multiplexers 712i and 713i, which are either of two inputs for each of registers RMAX 704i and POSRMAX 705i (similar to registers RMAX 704 and POSRMAX 705, respectively, in FIG. 31). Can be selected.
Multiplexer 712i selects either the value of register RMAXin upstream of device B (B0 or more generally Bi−1) or the input IN, and multiplexer 713i selects the value of the contents of register POSRMAXin upstream of device B or the input ADR−in Choose either. Both these selections are determined by the selection signal SelMux for the multiplexer. If signal IN has a value greater than what is contained in register RMAX 704i (comparison being performed in comparator 706i), signal IN and ADR-in (received at input 1 from the multiplexer) are verified by SelMux Is done. And if the comparator 706i from the upstream of device B, equal to the input signal a _i- << 0 >>, is generating a signal that is zero before inversion at the same time ai−1, Gate AND 714i generates signal 714s. Multiplexers 712i and 713 transmit RMAXin and POSRMAXin of device Bi−1 (ie, B0 if BI = B1). Register RMAX 704i, register POSRMAX 705i and comparator 706i, and comparator 709i and gate 707i are similar to the corresponding device of FIG. 31 without subscript i (the role will be described later).

Gate OR 715i and AND 707i enable signal WR for registers RMAX 712i and POSRMAX 713i if SelMux or TRi-valid and signal CALCUL is commanding a write sequence at the same time.
Finally, gate AND 710i (similar to gate AND 710 of FIG. 31) generates signal TRi. At the same time, the output signal of gate 715 is valid, the address signal ADR-in is not equal to the output signal of register POSRMAX 705i, and toggle 711i (similar to toggle 711 in FIG. 31) is Triggered after quitting signal RESET resulting from signal INIT by first signal WR. A device 708i having an AND gate (similar to device 708 in FIG. 31) causes the adder 107 (FIGS. 29b and 30) to register RMAX 704i when comparator 709i detects a = b. A signal OUT consisting of the contents of is generated.
At the end of the period, device Bi + 1 generates, in addition to signal OUT, four signals, namely RMAXout, ai, TRi and PORSMAXout, which are similar to the corresponding signals generated by device Bo in FIG. As further illustrated in FIG. 30, signals RMAXout and POSRMAXout (subscripts 1 through n are simply indicated as RMAX, subscripts 1 through n are simply indicated as POSRMAX) are blocks 617i (ie, 617 ₁ , 617 ₂ ... 617 _n ). These blocks, as blocks 617 ₀ receives B0 output RMAXout (RMAX0) and POSRMAXout (POSRMAX0). ) (These blocks) are activated by the addresses 0 to n of the demultiplexer 618 under the control of the API 602. Conversely, later device Bn generates its signals RMAXn and POSRMAXn only in block 617n. Block 617n is actuated solely by the address signal n of the demultiplexer 618 under the control of the API 602.

It should be noted that it is the suppression of the signal RESET that authorizes the writing of registers RMAX 704i and POSRMAX 705i at the end of the CLOCK INIT cycle. For the signal TRi, if it is an output and the signal WR to the register of block Bi or << 1 >> is valid and it is equal to << 0 >> on the first write, the signal ADR-in is After the comparison performed in the comparator 709i, the contents of the register POSRMAX 705i are different.
Previously given implementations and embodiments of STN modules allow the calculation of the values MIN, MAX, NBPTS, RMAX, POSRMAX (especially) during the calculation phase. In this type of STN module, a large number of logical gates have to be executed because of each of these calculations, especially for consumption, and space requirements that may be relatively important Must be shaved. That is why it is provided instead. Calculations of all or some of these values are performed by a microprocessor type, eg, an automaton, by a more general sequential calculation. From this point of view, during the calculation, Phase 1 essentially updates the memory 100, and at the end of the calculation phase (eg at the very beginning of the phase RESULT) it is itself a calculation option. -To calculate the value by moving the tomato, if calculated, the latter is collected before. This kind of implementation, however, allows a significant reduction in the number of logical gates and power consumption if it can slightly increase the lead time before obtaining results.

With respect to FIG. 33, the implementation of the STN module was illustrated schematically, incorporating the method of class extraction previously described in FIGS. 29-32.
This type of STN module 660 is commanded by and receives a zone signal Z defining a scene analysis zone, during the sequence (motion signal MVT), a bivariate primary This type of signal incorporates scene pixel motion discovery and velocity information.
Demultiplexer 653 generates classes CMVT0, CMVT1, CMVT2 and CMVT3 during the following sequence t1, but extracts during the cycle END for the class with different motion at the end of the sequence and memory M2 Is filled in.
Each classification signal commands a number of class-extracted, so-called STN modules, many classes of STN modules, each receiving a bivariate linear signal at position x / y. Thus, class signal CMVT0 commands STN module 661, signal CMVT2, STN module 662, signal CMVT2, STN module 663, and therefore the number of classes found in the description with respect to acquisition. There are as many STN modules as possible classes that can be entered.
At the end of sequence t1, the signal of class ZiMVTj of the STN module receiving the signal x / y defines all the zones associated with the motion category.
In the first sequence of the two sequences, signal processing of the time domain TD is performed, and the processing cycle of the spatial domain SD from the result before classification from the previous processing in the time domain TD is Executed.

The same principle applies from class ZiMVTj which distributes the command signal defining the scene analysis zone during the sequence t2 and t3. In the STN module of type 664, the color of the three lines of signal L / T / pixel is defined, and S is received separating the different colorimetric classes of each incoming class ZiMVTj The processing cycle is executed in this time domain TD.
The STN module of the module before type 665, 661 and 667 is divided into subzones downstream of the classification signal coming from the bivariate linear position signal x / y in the processing of the spatial region SD, respectively. .
With regard to the speed of execution thanks to the execution of the multi-class extraction means, this method of multiple class extraction shows cognitive performance. At the end of the four sequences t0 to t3, the amount of zone extracted is much higher than the processing and previously defined equipment.
FIGS. 21a, 21b and 24 (it is the position of the observed object OB in the frame work of the use of the single representative signal (ie luminance L) that we currently describe with respect to FIG. 34) The assembly and subs of FIGS. 17, 18, 19, 20, 22, 23a-23b, 23c and 25, with representations of the above results, illustrating the preparation of the necessary information that follows After the description of the assembly, the two objects that can be improved, the execution of the tree representing the relative positions of the centroids of the different zones, the three color components and finally the execution of the storage, and In size and in rotation, the translational invariance determines the relevant location of recognition of the recognized object.

With respect to FIGS. 17-62, the position (also called the parameter or center of gravity BarZi) essentially forms the devices STN (1), STN (2) and the device 8 of FIGS. 17 and 18. Run the average added to the input of the existing STN (3).
On FIG. 34, schematically, a module STN (2) of the type illustrated in FIGS. 23a-23b (possibly with the modifications of FIGS. 23c and / or 25) and two orientation devices pα And pβ are shown. These two orientation devices pα and pβ are for the coordinates x and y, the orientation or angle of rotation, ie α for the device of orientation 150, and for the device of orientation 151 , And rotate the coordinate axis from the initial position determined by the coordinates x and y to the shift position defined by the angles α and β or the inclinations pα and pβ. These orientations and tilts are represented in FIG. 34a described below.
The module STN (2) of FIG. 34 is therefore received. And it rather enters DATA (data other than Ap () and DATA (Ap () (not DATA (Ax) and not DATA (Ay)). This is executed in module STN (2) Processing is controlled by program register 152. Finally, the signal collected on bus 114 (from FIGS. 23a-23b) finally deduces not only Zi, but in particular the slopes pα and pβ, ie the angles α and β. The center of gravity BarZi) is also shown.

FIG. 34a illustrates two straight lines with two specific values of angle α, α1 and α2, and slopes p 畚 och1 and p 秉 och2 with respect to the direction of x, in the apparatus illustrated in FIG. Represents processing. The slopes p 畚 och9 and p 畚 och10 are perpendicular to the slopes p 畚 och1 and p 秉 och2, respectively, and the slopes p 畚 och1 and p 畚 och9 for angles α1 and α9 and for the angle α2 Provides Cartesian coordinates with slope p 畚 och10.
FIG. 34a also represents the limits a and b corresponding to the slope pα2, and the limits c and d corresponding to the slope p 畚 och9. Therefore, the object by its center of gravity BarZ0 (subscript << 0 >> indicates that it is the first center of gravity or the origin center of the tree described later) of the diamond indicated by reference number 160 Define OB. It forms the first zone Zo defined by a straight line of coordinates a, b, c and d describing this object in detail.
As illustrated in FIG. 35, changing α which rotates straight lines Limit A and Limit B, ie, slope pα, shows different orientations of angle α in the frame work or zone Z0. did. It corresponds to the variable Zr, ie Zr0 (α = 0), Zr10 (α = 10 degrees), Zr30 (α = 30 degrees), and Zr170 (α = 170 degrees). FIG. 35 specifically illustrates a slope pα3 having a corresponding Limit A and Limit B defining a and b for a value of 30 degrees, ie, angle α = 30 degrees. In FIG. 35, it should be noted that all stripes BarZr run through the center of gravity BarZ0, as defined in FIG. 34a.

FIG. 36 shows a portion of FIG. 23a, ie, a variation of device 101, comparators 110a and 111a, comparators 110b and 111b, and AND gate 112a (which receives the output directly from comparator 110a at the input). Receiving and returning to comparator 111a) and AND gate 112b (which receives the direct output from comparator 110b and returns it to comparator 111b).
In FIG. 36, the AND gate 131 receives the outputs of the AND gates 112a and 112b. The example of FIG. 36 is new with respect to FIG. 23a. In addition to the AND gate 131, the AND gate 131 has an OR gate 132 that receives the outputs of the AND gates 112a and 112b. That is.
The output of gate AND 131 or 132 is commanded by a select signal 134 (instructed by the module STN API) that selects between the output of AND gate 131 and the output of OR gate 132. It is sent to the multiplexer 133. Then, the signal 101s (consisting of 131s or 132s) output from the multiplexer 133 is sent to the bus 114.
Finally, as a corresponding part of FIG. 23a, FIG. 36 shows a module in a bivariate linear mode, i.e. a module receiving two input signals (i.e. DATA (Ay) and DATA (Ax)). The classification subassembly for Le STN (2) (Fig. 23a on Fig. 36 and 101 'on 101).
In FIG. 23a, four comparators 110a, 110b, 111a and 111b make it possible to output four signals (i.e. ax, bx, ay, by).
In the case of the figure, and beside each. Due to the logical functions implemented by the three gates 112a, 112b and 131, the signal 101s generated by the AND gate 131 is defined by:

上記ブール論理演算が累算的であるので、以下のように変形させることができる。

Since the Boolean logic operation is cumulative, it can be modified as follows.

Referring to FIG. 36, the variation of FIG. 23a is illustrated. It can be seen at the output of gate 131 that the assembly illustrated in FIG. 36 implements the same logic operations and assembly as FIG. 23a. The output signal 131s is equal to the above. However, further considering the gate OR 132 included in the assembly of FIG. 36, the gate 132 whose output signal 132s is (ay.by) + (ax.bx) as compared with the gate 131. The output signal has a wider application range.
The difference between the domain of FIG. 37 corresponding to the arrangement according to FIG. 23a and the domain of FIG. 38 corresponding to the arrangement according to FIG. 36 will be described. 37 and 38 show coordinates x and y and limits Ax and Bx, Ay and By. In FIG. 37, corresponding to the arrangement according to FIG. 23a, the output signal of the AND gate 131 (signal 101s) directly reaches the bus 114 and has a shaded rectangle Zand as a domain due to four limits. This is because it only passes a signal consisting of the time between the AND gate 13, Limit Ay and Limit Bx and between Limit AY and Limit BY. Conversely, in the case of the device according to FIG. 36, the signal 101s is composed of the signal 132s or the signal 131s. In the first case, the region 101s, that is, the region 131 is the example shown in FIG. 37. In the second case, the region 101s, the region 132s, is the shaded region Zor in FIG. . This region is either between Limit Ax and Limit Bx or between Limit Ay and Limit By, but excludes the rectangular blank in FIG. In those rectangular blanks, the signal is less than Limit Ax for Limit y for coordinate x and less than Limit Ay for Limit y for coordinate x or greater than Limit By for Limit y for coordinate x. Therefore, because of the additional gate OR 132 of FIG. 36, the area of application is increased compared to FIG. 23a, which does not include this type of gate.

The multiplexer 133 in FIG. 36 performs the output 131s of the gate AND 131 (shadow area in FIG. 37) and the output 132s of the gate OR 132 based on the control signal 134 (from the API) of the multiplexer according to the desired processing. Selection is possible between (the shadow area in FIG. 38).
In the example of FIG. 35 corresponding to a single gate AND 131, narrow stripes delineated in detail by Limit A and Limit B for various orientations or slopes, eg, pα30 for stripe Zr30. Or many orientations of << Line >> can be used. The centroid BarZ0 of the zone Zr is defined by the intersection of the stripe Zr0 (inclination 0 degree) to the stripe Zr170 (inclination 170 degree).
Instead of multiple narrow stripes, the plane of Z0 can be conveniently covered by a sector as illustrated in FIGS. 43a, 43b and 43c. It corresponds to the use of one gate OR 132 of FIG.
After the first Z0 determination and its center of gravity BarZ0's first sequence, the device goes on to the second sequence according to FIG. So, instead of the shaded part corresponding to the shaded part of FIG. 38 and being perpendicular, the axis is defined with (sharp) angles 畚 och1 and α2, and the limits are Limit Ax, Limit Bx and Limit Ay, Instead of defining Limit By, it is represented by the position of the center of gravity BarZ0 (included in Z0) by the slopes p1och1 and pα2. On this FIG. 39a, the limits Limit Ap 畚 och1 and Limit Bp 畚 och1 for the slope p 畚 och1 and the Limit Limit Ap 秉 och2 for the limit p 畚 och2 and Limit Bp 秉 och2 are represented. With the module STN (2) of FIG. 34, the zone Zr1 is seen as a signal 101s derived from the output signal 132s of the gate OR 132.

An additional signal representing BarZ1 is found in the third sequence illustrated in FIG. 39b. Therefore, the calculation cycle of module STN (2) in FIG. 34 determines the position of BarZ1 above BarZ0, and the resulting updating cycle defines BarZ1, a new limit Limit A'p. Determine 畚 och1 and Limit B'p 畚 och1 and Limit A'p 畚 och2 and Limit B'p 畚 och2.
In the fourth sequence (Figure 39c), during the third sequence, the same by adding module STN (2) to several sectorized subzones Zra, Zrb, Zrc Zrd Thus, the sectorized zone Z21 determined by the subzone operating in the module STN (2) executed according to FIG. 39a is divided and the sectorization is fine-tuned. Thus, the slopes p 畚 och1 and p 畚 och3 are replaced with intermediate slopes detailing the above mentioned subzones.
At the end of this fourth sequence, the subzone STN (2) module searched for BarZ1 (ie, subzone Zrc (FIG. 39c)). The optical axis angle isoline of this particular subzone tilted by Zrc (defined by 2 very close) determines exactly that the tilt p 畚 och3 is illustrated in FIG. 39c.
Consider the slope p 畚 och4 (perpendicular to p 畚 och3). The boundaries Limit A and Limit B can define BarZ0 of zone Z0 for the intersection with a stripe perpendicular to p · that is limited by Limit A and Limit B. In contrast, the limits Limit A. 'and Limit B.' BarZ1 can be defined for the intersection with a stripe perpendicular to p · that is limited by Limit A and Limit B (this stripe includes BarZ1 as well as BarZ0). The distance between the axes of two stripes perpendicular to the stripe BarZ0-BarZ1 and perpendicular to p 畚 och4 running through BarZ0 and BarZ1 is the distance ρ3 between these centroids. The angle 畚 och3 of the inclination p 畚 och3 and the distance ρ3 are polar coordinate positions of this centroid such as BarZ1 when the centroid BarZ0 is the coordinate origin. (For example, parallel to the lower edge of zone Z0 (FIGS. 39a and 39b)).

For the next sequence, all STN (2) modules are freed except for the module STN (2) that received the subzone Zrc. This block records the results ρ3, α3, Cα3 in registers 1070, 1071 and 1072, respectively. These registers belong to the zone register result 104 of the module STN.
Between the center of gravity of the zone Z0, the center of gravity of the zone Z1 included in the zone Z0 is between the center of gravity of the zone Z0 and BarZ0, which is the center of gravity of the zone Z1. In relation 49 of FIG. 40, BarZ0 represents << parent >>, and then comes << child >> BarZ. α3 and ρ3 are the polar coordinates of the child with respect to the parent.
This description of the four sequences shows the process of determining the link ρα between the two centroids from the reconstruction of the STN (2) module between sequences 2-4. The module STN (2) in the fourth sequence (ie, that corresponding to the subzone Zrc), the other selected STN (2) module becomes available. The same goes for STN (2) modules that are not selected in the previous sequence.

FIG. 41 illustrates an apparatus M (0) consisting of a pair of STN modules. The first module 296 consists of a univariate linear module STN (1), and the input parameter DATA indicated by A is processed by the given requested function FoG ′ and is indicated by 296b. To a group of analysis output registers. There, the accumulated value represents the distribution of the statistics of the parameter DATA (A) in the form of a histogram. Accumulate values. The output signal 296c of the module STN (1) 296 is a classification signal CA 350. The second module 297 is a bivariate linear module STN (2) having two input parameters (ie, coordinates x and y) indicated at 297a. These two input parameters are processed by the requested function FoG ″ given the analysis output and fed to a group of analysis output registers 297b which sort the parameters x and y in the histogram. The output signal of module STN (2) 297 consists of zone Z0 and is input as its signal 297s to the auxiliary input 296d of the first module 296. The other output signal of register 297b consists of BarZ0.
Obviously, as done last, modules STN (1) 296 and STN (2) 297 are shown in FIGS. 20 and 23a-23b (probably FIGS. 23c, 25 and / or Having 36 variations), the output signals Z0 and BarZ0 can be obtained.
FIG. 42 illustrates a set of STN modules that can perform the continuous operation described above with reference to FIGS. 39a, 39b and 39c. The apparatus M (0) of FIG. 41 is composed of a pair of blocks STN (1) 296 and STN (2) 297, receives input signals A, x and y, and outputs output signals Z0 and BarZ0.
The first device M (0) consists of two STN modules (1) 296 and STN (2) 297, and determines Z0 and its center of gravity BarZ0.

In succession, a similar device illustrating only the first device, ie M (1), x and y (coordinates), where Z1 and BarZ1 are determined from Z0 and parameter B (and no longer A). The following device with two STNs (not represented) determines Z2 and BarZ2 from Z1 and from parameters C, x and y. Also, the following devices not represented determine Z2, BarZ2, etc.
And
-On the other hand, a set M (2) of STN modules (2) representing the dynamic replenishment sequence 300, 301, 302, 303 ... 307 has a pair of slopes p0 and p1, p1 and p2, p2 and pr, p3 and p4, ... p6 and p7, respectively, as inputs, from Z0 and BarZ0 from device M (0) and from BarZ1 from device M (1) (illustrated in FIG. 43a) Zr0 , Zr1, Zr2, Zr3 ... Zr7 are determined. These three values are received at the input of each of the devices 300-307. Assembly M (2) receives each Z0 and BarZ0 of assembly M (2) and comes with a similar assembly that is not represented, and also ranks following assembly M (2) just illustrated For each of these assembly M (2) STN (2) type M (2) pair, BarZ3 and others for each type M (2) device, not just for that of the next rank BarZ2 receives a set of slopes in STN (2) 300-307, which are similarly exemplified. These devices of the continuous type M (2) are no longer associated with Z0, Z2 etc. but with respect to Z0 at the output of those modules similar to the illustrated modules 300-307. No) Determine the signal defining sectors of type Zr0 to Zr7.

The operation of the assembly illustrated in FIG. 42 is as follows.
During the first sequence (for example, a video image or frame of a video image) (which performs assembly M (0)), it is carefully created in the classification signal 350 (module 296). Arriving at module 297 operating in a two-dimensional mode with coordinates x and y (defining the pixel of the video image) as input parameters (in addition to signal 350) thing).
The result of the calculation of the histogram of the module 297 of the machine classification with the expectation at the arrival of the signal 350, ie during the end-of-sequence phase END, commands the function FoG ″. Register registration calculation and classification allows module 297 297b, Z0 to be determined by depression (as exposed with respect to Figure 34) Limit Ax, Limit Bx, Limit Ay and Limit surrounding Z0 Update By. Over BarZ0 and POSMOYy by POSMOYx. By the first sequence, it is therefore possible to determine Z0 and BarZ0.
STN (2) modules 300-307 (what only a certain module was illustrated for) that the sequence intervenes during a continuous sequence (ie, the second sequence) For the second and fourth programmed to receive a set of 22 degrees 30 minutes at the input, represented by 22 degrees 30 minutes between 0 degrees (slope p0) and 157 degrees 30 minutes (slope p7) Assembly M (2) is replenished.

Recognize BarZ0 within the zone Z0 following each of the degrees of orientation (FIG. 43a), with the inclination of the subscript j representing an angle of 22,5j, respectively. The calculation of the histogram of STN (2) that makes it possible to expose the classification limits Limit A and Limit B by modules of type 300 to 307 with a new end-of-sequence signal END of this second sequence, thus Define the direction zone. As shown, from zone Zr0 on Fig. 43a for zones Zr1, Zr2, Zr3 ... Zr7 determined by blocks STN (2) 300, 301, 302, 303 ... 307, respectively. Search for Zr7. FIG. 43a states that the zone total actually directed Zr formed by modules 300-307 during the second sequence.
The second sequence allows a planar sectorized position around BarZ0. At the end of the second sequence, the features associated with the first classification signal 350, i.e. the features of the subzone (not represented) generating the center of gravity BarZ1, appear. It is used to determine the orientation module ρ and angle α (illustrated in FIGS. 43b and 43c corresponding to FIGS. 39b and 39c).

Still, during the third sequence that follows immediately after the second sequence, the STN (2) module 300-307 device has slopes p0, p1,... P7 in addition to Z0 and BarZ0. And a third signal BarZ1. That is, pi refers to an additional centroid BarZ1 (FIG. 43b) equal to the difference between both classification limits from these inputs, the STN module of M (2), eg to 303 (FIG. 43b), and the sequence Will be the end. The STN module chooses to operate in the subsequent or fourth sequence.

During this fourth sequence, STN (2) modules 300-307 are reconfigured as represented in FIG. 39c by receiving:
* Slopes p3, pi,... Pj, p4 (represents some sectorized subzones of zone Zr3)
And
* Inclination p'3, p'I, ... p'j, p'4 (each perpendicular to the above inclination)
* ρ and α are calculated by the selected module, eg 303. Here, ρ is the distance LimitA′−LimitA (FIGS. 39c and 43c).
A new cycle of four sequences can begin the assembly of the module STN (similar to assembly M (2)). However, free STN modules are used but not represented to determine Z20 and BarZ20 and others.
In the case of approval of the object, the part ρα of BarZ1 can be proposed. Sequence 4 is added directly after sequence 2. Sector-Zr3 is subdivided. The module ρ and its angle α are calculated in the previous sequence.
Reference is now made to FIGS. 44a, 44e, 45 and 46 and FIGS. 47a 47d, 48 and 49. FIG. Two specific examples of arrangements relating to zones Z10, Z20 and Z30 in zone Z0, ie processing of a relationship in which one such as a Russian doll (but two-dimensional) is included in the other) And separate local identification.

44a to 44c and 47a to 47d represent the STN module assembly M, FIGS. 45 and 48 show the corresponding associated arrangement of the zones, and FIGS. 46 and 49 show the center of gravity BarZ0 of these zones. Represents the BarZ30 tree.
44a to 44e show the first of all cases in which the zone according to FIG. 45 is implemented by the assembly M of the STN module for both parameters x and y, with one included in the other. Investigate. Furthermore, we begin by studying the creation of a link (according to FIG. 46) between the centroids BarZ0 and BarZ10 of the zones Z0 and Z10 (according to FIG. 45) which are contained in each other.
During phase t0 (FIG. 44a), parameter Data (A) (A) determines zone Z0 and its barycenter BarZ0 by the action of module M (0).
In the second step, during phase t1, the signal Z0 generated by M (0) of phase Co is converted to a local specific module M of subzone Z10 by means of data item Data (B). (1) Acts upward to determine Z10 and its center of gravity BarZ10. And on the other hand, the module M (2) operates as described above with respect to FIGS. 42, 43a, 43b and 44c, and has different orientations around the center of gravity BarZ0 to place these orientations. Actions p0, p1 ... pm have been reorganized.
In the third step, during phase t2, module M (2) receives BarZ10 (above BarZ0) just determined by M (1) and determines the sector containing BarZ10. Search.

Since the sector is known during the fourth step during phase t3, the module specified by SBI (FIG. 44e) is selected from the STN (2) module of assembly M (2). Le STN (2) is selected. And it determines the value of i corresponding to the angle of the optical axis angle bisector of the sub-sector selected by BarZ10.
Dynamic reorganization of M (2) disappears into the advantages of SBi selected bivariate linear module STN (2). Registers 1070, 1071 and 1072 containing ρ, α and Cα, respectively, are updated.
In FIGS. 44b to 44e, Ia, IIa, IIIa, and Iva indicate structures at times t0, t1, t2, and t3.
The operations described above to move Z10 and BarZ10 relative to Z0 and BarZ0 determine Z20 and BarZ20 from Z10 and BarZ10 for each of the zones that are included in the other. Applicable to determine Z30 and BarZ30 from Z20 and BarZ20, and so on.
The structure of the next phase at times t0, t1, t2 and t3 can determine the zones Z20, Z30 and their centroids.
44c, 44d and 44e, Ib, IIb, IIIb, IVb, then Ic, IIc, IIIc, IVc (not shown), followed by Id, Id, IIId (not shown), IVd (not shown).

Finally, the unbranched tree of FIG. 46 illustrating the parent-child relation BarZ0 → BarZ10 → BarZ20 → BarZ30 of the center of gravity is obtained. The polar coordinates of the downstream center of gravity (child) with respect to the upstream center of gravity (parent) are ρ and α. As an illustration of the parent-child relationship, BarZ20 is a child of BarZ10. The tree of FIG. 46 illustrates the successive inclusion relationships of the corresponding zones Z0, Z10, Z20, Z30 of FIG. Z30 is included in Z20, Z20 is included in Z10, and Z10 is included in Z0. For example, when examining a human face, observe the entire face (Z0), then one eye (Z10), then the pupil (Z20), and finally the iris (Z30). To do.
Since the successive polar coordinates of the parent-child relationship are determined, the tree represents deeper features of the observation object regardless of its rotation.
44a-44e, 45 and 46, after examining the case of a continuous zone (and hence a non-branching linear tree), one included in the other, Examples in which zones Z10, Z20 and Z30 are located in zone Z0 but are separate from each other (FIG. 48) will be described with reference to FIGS. 47a to 47d, 48 and 49. FIG.
In the opposite case of FIG. 45, since Z10 is included in Z0, FIGS. 47a and 47b are also shown in FIGS. Then, in each case to get the first of BarZ0, the zone Z20 of this second example in FIG. 48 is not included in zone Z10, but is clearly separated, but included in zone Z0 Therefore, the processing sequence is different.

At the third time t2, module M (1a), which is similar to module M (1), is C instead of B (since Z20 is not included in Z10 but is included in Z0) ) Determine Z20 and BarZ20 that receive Z10 as in FIG. 44c. However, the STN module M (2a) determines ρ corresponding to the zone Z20.
At the fourth time t3 (FIG. 47d) and subsequent times, its M (9) is no longer the Z0 of the previous phase as in the case of a continuous zone, one included in the other. The process is repeated for new zones that are separate from each other in zone Z0 using Z0 as the input of the module.
Accordingly, Z0, Z10, Z20 and Z30 are subsequently determined, and a branched tree or branch is obtained from BarZ0. As an illustration, examine the entire face (Z0) and observe the mouth (Z10), then the right eye (Z20) and finally the left right eye (Z30). The entire face covers the mouth and eyes. The letters A, B, C correspond to different curves a. '., B.'., C. '. Of FIG.
After examining the case of successive events in one zone that is contained in the other or balanced with a tree that does not have a secondary branch, two or more zones simultaneously (three or more of the video signal Consider the more general situation that appears during the same sequence of frames or images, which corresponds to a tree with multiple branches or secondary branches.
If several peaks or classes appear during the same sequence, a zone of type Z is included in one zone but away from the other zones. Generally speaking, a typical example of a zone (Figs. 44a to 44e, 45 and 46), one included in the other, and a separate zone but included in the first zone Z0 This is an intermediate case between the included typical examples (FIGS. 47a to 47d, 48 and 49).

FIG. 50 illustrates this type of intermediate case. There, at least one zone included in the other at the same time is included in, for example, Z2Z12 (Z12 → Z21) and there are more than two positions such as Z11 and Z22 and Z21 and Z12 Away from the zone. However, all these zones are included in zone Z0, as can be seen in the tree of FIG. 50a.
With respect to FIG. 50, the left to right, the duration of several consecutive sequences or the horizontal axis from the group of sequences t0, t1, t2 and t3, the first during refocusing Processing steps, or more generally increased in resolution, and in ordinate, from top to bottom, processing inheritance performed a sequence or group of sequences for each.
1) First consider all ordinates (ie continuous rows in Figure 50).
From left to right, the first row of FIG. 50 illustrates the zonal events that occur due to the increased resolution associated with the parameters. The difference Dif between the first sequence at the maximum resolution rmax during T0 and the next sequence at the minimum resolution rmin between T2 or t0 (FIG. 3) is expressed as a, b (FIGS. 14 and 15).
The second row in FIG. 50 shows an example in which the first of the consecutive processes to be executed is executed three times in the first zone to be observed (that is, Z0).
The third row of FIG. 50 shows both subzones determined after the first zone Z0 (ie, both subzones Z11 and Z12) with the first of both successive operations being performed. Illustrate.

The fourth row of FIG. 50 illustrates the first of the sequential processes performed on both subzones determined therefrom, namely Z21 and Z22 (Z21 included in Z12).
2) Consider the horizontal axis, i.e. the continuous column of Fig. 50, i.e. the sequence or sequence of groups.
The first column corresponding to the initialization at time t0 is the first zone Z0 following the first increase in resolution (after its sharp decrease), during phase T1 in FIG. It represents the occurrence. Then, the center of gravity BarZ0 of the zone Z0 is determined.
The second column represents both simultaneous processes at time t1 in the next sequence, ie illustrating the continuation of the process for zone Z0.
-Subzones Z11 and Z12 with their respective centroids BarZ11 and BarZ12 occur in zone Z0 (in the first row).
And
Divide zone Z0 into multiple sectors (in the second row) as described above with respect to FIGS. 42 and 43a, 43b, 43c.
The third column represents three simultaneous processes performed in the same sequence (ie time t2). That is,
-Subzones Z21 and Z22 with their respective centroids BarZ21 and BarZ22 occur in the process of increasing resolution (first row)
-In each sector of Z0 (previous column in the same row), determine the center of gravity BarZ11 or BarZ12 and determine the module ρ and the angle α that form the polar coordinates of BarZ11 or Bar12 with respect to BarZ0 (its Second line). The reference radii 50 and 51 are attached to the vector radii of BarZ11 and BarZ12 from BarZ0.
The division of the respective centroids BarZ11 and BarZ12 into sectors of zones Z11 and Z12 (the third row).

The fourth column illustrates the continuation of the same type of processing:
Generation of subzones Z21 and Z22 (contained in Z12) with respective centroids BarZ21 and BarZ22 in the first row;
-Determination of the module ρ and the angle α of the positions of BarZ22 and BarZ21 relative to BarZ0 and BarZ12 in the second and third rows. References 52 and 53 are attached to the existence of the vector radius.
-In the fourth row, the division of zones Z21 and Z22 into sectors.
FIG. 50a represents a tree created from the processing realized in FIG.
In FIG. 50,
-The first parameter Dif can determine the zone Z0 and its centroid BarZ0, which acts as a starting point for the tree (returned).
The second parameter a produces sub-zones Z11 and Z12 (included in zone Z0) with respective centroids BarZ11 and BarZ12. The parent-child relationship with parent BarZ0 and children BarZ11 and BarZ12 is materialized by vector radii 50 and 51 at angle α11 and angle α12 and modules ρ11 and ρ12, respectively.
And

-Generate two new subzones Z21 and Z22 with the third parameter b. The former is contained in (Z12 (and hence Z0), the latter is contained only in Z0. They have centroids BarZ21 and BarZ22, respectively, and have parent-child relationships 53 and 52 with respect to parents BarZ12 and BarZ0. Their α and ρ correspond to α21, α22 and ρ21 and ρ22.
From FIG. 51, in part, from the parameters of the space obtained and this time during refocusing, the application of multiple zone selection developed in FIGS. From the minimum value to the maximum value, the resolution is restored as described above with reference to FIGS.
FIG. 51 corresponds to the periods T2 and T3 in FIG.
During one period T2, ie the sequence t0 uses a signal, in particular a difference signal Dif corresponding to w = 20. This difference signal is also input to the input Dif of the first one-dimensional module STN (1) 90 of the module M (0) comprising: Module M (0) has a second two-dimensional module STN (2) 91 and at the end of period T2 = t0, according to module STN (2) 91 of assembly M (0), Like its center of gravity BarZo, the reference zone Z0 has been localized.
The refocusing, or generally the period T3, of which the resolution recovers from its minimum value to its maximum value includes the sequence t1, t2, t3. Only the sequence t1 (corresponding to w = 16) and t2 (corresponding to w = 12) are illustrated in Fig. 51. During the period t1, t2 ..., the series of spatial in Fig. 15 Temporal signals a, b, c ... appear.

FIG. 51 is similar to FIG. 47d, except for a dynamic replenishment calculation module that can calculate ρ and α for the purification object.
During the first sequence t1 of period T3, a signal w = 16 appears and it is analyzed by an assembly M (1) comprising STN modules (1) 92 and STN (2) 93. At the end of this sequence t1, an analysis of class multiple extraction and the distribution described in each class as described above (especially with reference to FIGS. 31 and 32) are accompanied by a distribution in decreasing order. Dimensional module STN (2), ie the second class on module STN (2) 94, the third class on module STN (2) 95 and STN (2) module 96, STN ( 2) Beside others for 97 others.
During the second sequence t2 of period T3 (sequence immediately following sequence t1) it appears and assembly M (2 with two STN modules (1) 192 and STN (2) 193 ) Is the signal b (w = 16) analyzed by.
During this sequence t2, and in such a case, a distribution is performed in reducing the order of the classes known to the accompanying two-dimensional module STN.
On module STN (2) 194 as module STN (2) 95 (on the second two-dimensional module (not shown) downstream of module STN (2) 194) The second class is downstream with module STN (2) 94 and other, other modules (depicted by arrows vertically positioned below module STN (2) 194. To be placed on).

During the subsequent sequence t3, t4, ... in period T3, resolution, w = 8, w = 4, of blocks similar to blocks M (0), M (1) and M (2) A similar process is performed with an increase of w = 0 in the two-dimensional module STN (2) similar to the modules 94, 95, 96, 97 shown in FIG. And it stops at resolution w = 12.
As mentioned above, the zone Zi where BarZi will be used later and their corresponding centroids are the zones Z0, Process FIG. 51 which illustrates only Z11, Z12, Z13, Z14, Z15, Z21 and Z22.
In the upper part of FIG. 51, M (0), M (1) and M (2) for resolution correspond to the three assemblies defined by w = 20, w = 16 and w = 12. Examples were di and zone.
Some zones are illustrated in the image for resolution defined by w = 16 and w = 12.
Only the first three columns are illustrated on FIG. 51 (the image corresponding to w = 12 representing the fact that there is a column to the right of the last column schematically illustrated in FIG. 51) Arrow on the other side of.

FIG. 52 illustrates the improvements made to the apparatus of FIG. 51 for better locking to correct information shortcomings. The parameters a, b, c ... are temporary.
From this perspective, Figure 52 (STN (2) 97 in Figure 51 processing coordinates, notice module STN (2) 93 x, y and modules, their inputs and their outputs. In an embodiment of STN (1) 92), each subzone is locked.
In the embodiment of FIG. 52, an additional three-dimensional module STN (3), T for processing the color component L, the modification is essentially the addition of S, 93a, 94a, 95a 96a and 97a with inputs connected to the classification output referenced by D of the STN (2) module. It corresponds to 93, 94, 95, 96 and 97.
The dominant color is the input L, T, S, ie each subzone by module STN (3) 93a to STN (3) 97a with three color components and these The zone Zi for each is locked by analyzing the dominant color as a prepaid command.
At the end of the sequence, the histogram of the component L (T) that makes it possible to define the classification by S restricts L (T) and therefore the classification signal 101s corresponding to the dominant color (Fig. Outputs 23a-23b or possibly 23c, 25).

Locking is obtained by connecting this classification signal corresponding to the dominant color to the input of the corresponding STN module that computes the accompanying zone.
In Figure 52, it is shown symbolically on the right. Modules 93-93a, 94-94a, 95-95a, 96-96a and 97-97a receive position inputs (coordinates x, y), color inputs (components L, T, S) and From 51 parameters, namely Z11 and BarZ11, Z12 and BarZ12, Z13 and BarZ13, Z14 and BarZ14, Z15 and Bar15, the outputs C11, C12, C13, C14 and representing the calculated and dominant colors C15) is output.
The last column in FIG. 52 represents a tree with a tree node BarZij (subscript ij represents consecutive 11, 12, 13, 14, 15). A node is characterized by its zone Zij and its associated dominant color (Cij).
Thus, it was possible to recognize the object within the frame work of the present invention.
Referring to FIG. 53, first of all, the process will be described, nevertheless, within the frame work of the present invention, whether the new shape is similar to the shape already determined, in particular In order to investigate, it is possible to delineate the features of the shape of the object to be localized with respect to the recognized features of the object in terms of its shape being realized in two stages.

All the first by recognizing whether the new object already corresponds to the feature of the previous object and then coding information about the feature with respect to that feature Whether it is new or it is new.
Thanks to the assembly illustrated in FIGS. 44a, 44e, 45 and 46 in FIG. 44a all such and on the other hand must first be reminded to the assembly illustrated in FIGS. 47a-47d, 48, 49 . It is possible to obtain different couples consisting of the angle α (or slope p) and the module ρ corresponding to the features of the zone Z0 and other zones.
The example shown in the left column of Fig. 53 is a bivariate linear module m (0), m (1), m (2) ... m (j) ... Corresponds to the bivariate linear module SBi of FIG. The inputs and outputs of module m (0)... Are the same as those of module Sbi. In particular, the output of these modules m (0) ... delivers all couples mentioned above. ((Z0 added to the double bus BB α, ρ recognized by the act as input to the set MM that constitutes an important means of determining the feature representation of the object (subscripts 0, 1, 2,. .. j.) in the zone, the assembly MM consists of several sub-assemblies MM (0), MM (1), MM (2). ) ... (each of these assemblies consisting of two STN modules) The first of the two-dimensional types referred to 801, 802, ... 80i, which is doubled on input (And the second one of the one-dimensional type referred to 810, 811, 812..81i .. and it is awarded a LABEL signal whose role is explained.

Finally, the order (0, 1, 2 ..i) (enclosed in the square) instructions are by each subassembly STN (2) second of all subassemblies by LABEL bus L The input of module STN (1) and the first module to command are defined.
The zone in which all the operations of the device of FIG. 53 are applied to the second feedback input of the assembly MM, ie the first STN (2) module of 800, 801, 802 different ... The software manages the different sequences 0, 1, and 2 depending on the software, though i need to keep in mind the Z0 signal involved ... i ... connected with the function.
After an initial reset of all memories of the STN module of assembly MM and the pointer value of the last object obtained, the apparatus of FIG. 53 can operate. A sequence number from 0 to n (via i) with increments of 1 for each of the assemblies MM (0), MM (1), MM (2) ... MM (i) ... MM (n) Assigned by (not represented).
It should be noted that for the first subassembly MM (0) corresponding to sequence number 0, the classification limit of both STN modules 800 and 810, it is published to the maximum number. It then adds << 0 >> to the maximum value of the input bus.

During processing, the resolution switches from its minimum value to its maximum value:
* For the period T2, reset the single-line histogram of each subassembly MM (0), MM (1), MM (2) ... MM (i) ... And the bilinear histogram calculation is blocked.
* For period T3, an available cycle has begun because each event of the new couple ((test to prove whether the couple is known.
Two cases are possible:
The learning phase when the couples α and ρ do not find a valid classifier other than that defined by the pointer identifying << last acquisition >> in subassemblies MM (1) to MM (n) Is activated, the last acquisition subassembly MM for which the classifier is still valid forces the verification of that LABEL associated with LABEL bus L, and the first module of the last acquisition subassembly MM 820-82i -Histograms (800 to 80i) store couples α and ρ and associated values 820-82i. This kind of operation is repeated for new couples α and ρ during the period T3 of switching from the minimum resolution to the maximum resolution (eg throughout the refocusing phase).

-Conversely, if the couples α and ρ find at least one other classifier defined in the << last acquisition >> pointer in subassemblies MM (1) to MM (n) When this couple is already known, these values α and ρ are stored in the corresponding STN modules.
For example, couples α and ρ are well classified by subassemblies 1 and 2, both of these values 1 and 2 are output on LABEL bus L, and a classifier of single-line module STN (1). (Ie 811 and 812) are validated. The latter is validated by the feedback process. Couples α and ρ are stored in the accompanying two-variable linear STN module, that is, 801 and 802, and labels 1 and 2 are stored in subassembly MM (1) and subassembly MM (2).
At the end of each processing of the previous type, the values assigned by RMAX and POSRMAX of each single-line module STN (1) are updated and all single-line modules already used (STN1) The general and read which allows to retrieve the largest value RMAX by the test above, and therefore its associated register POSRMAX corresponding to most expected contestants evaluates LABEL. And that exposed the above regarding API calculation software. The contents of the accompanying bilinear histogram are then read and act as a supplementary forecast for the decision-making process on duty as a dynamic supplementary officer, the four-step process previously 3 (previously specified with respect to Figure 42 As described) to decrease. And it speeds up convergence for object approval.

Once the maximum resolution has been obtained, i.e., complete refocusing in the matter, calculation phase cancellation and analysis time, the resulting phase begins. The latter uses a histogram of bivariate linear STN (2) and single-line STN (1) modules, a comparison of the values read for the reference threshold, i.e., RMAX / 2 and transcription. Features are rendered by reading memory. The result of the 1-bit memory of the bivariate linear module classifier is extended to receive adjacent couples α and ρ during the next test. The final value of LABEL corresponds to the name of the object to be included in zone Z0. This value LABEL corresponds to so-called episodic memory.
During processing on bus L, several values LABEL can be moved simultaneously. Accumulating these matches by calculating a histogram with a single-line STN module 670 with multiclass extraction represents a new class, CAT-1, CAT-2, CAT-3, and together this These classes associated with different values of LABEL are reflected in the categorization memory. Indeed, several objects can belong to the same category, for example, a surface category that includes a given person's label. That is, this process and associated devices represent both forms of storage, i.e. episodic and semantic forms. All approval of events in the first continuous learning process is categorized (object approval).

One of the features of the present invention is the problem of obvious variations between the object being examined and the observation device (eg camcorder) as the size of the object being examined, eg distance increases or decreases. It is possible to solve. Realization of size invariant resolution is illustrated in FIGS. 54, 55a, 55b, 55c and 55d.
FIG. 54 represents an assembly of the same electronic elements as the assembly illustrated in FIG. 42, but the processing is performed on different input parameters. The basis on which the device was described with respect to FIG. 34 replaces input parameters with two input parameters x and y corresponding to one Cartesian coordinate (issue date from sequencer 9). (And the logarithm LD of the distance was derived from the device 400 of the first distance parameter that Dis (issue date from the sequencer 450) itself originated from the external device 25 (not explained). 54 assemblies use a device 450 corresponding to the sequencer, which generates three potential values LD and ρ throughout the sequence (ie 400, 451 and 440). The device 400 consists of a logarithmic calculator receiving the distance Dis at the input and calculating the LD (the logarithm of Dis composing the output LD) device 45, the binary signal Barρi. (Equivalent to the previously used signal BarZi), which is activated when the signals LD and ρ from the sequencer 450 match the current values LDI and ρi .
Finally, the device 440 corresponds to a dynamic replenishment module M2 defined 42 in the figure. Instead, the following processing is performed.

* By using the variables LD and ρ exiting the sequencer 450, equivalent to the variables x and y in FIG. 42, the sectorization of the surface LD / ρ is transformed by the devices 420-427. Devices 420-427 ensure each two axis transformations following θi and θj, for example, devices 150 and 151 described above.
This device 420 generates two signals 4200s and 4201s corresponding to the definition of sector-Zr0 by transforming the axes of angles θi and θj. The following devices 421 to 427 divide the plane defined by the variables LD and ρ into sectors.
The invariance calculation of the input data Ldi / ρi is performed by using a two-variable linear expression STN (2) module 430 to 437 that receives signals 420is and 420js as input data. It is controlled by a zone validation binary signal Z and a signal Barρi corresponding to the values of the inputs Ldi and ρi.
FIG. 55d illustrates the result after processing the dynamic replenishment STN module during the three consecutive sequence photos of FIGS. 55a, 55b and 55c. Devices 400 and 451 remain the same, only device 440 is replaced with devices 401 and 402.
Device 401 (which corresponds to device 150 in FIG. 34) receives incoming sequences LD and ρ coming from device 450 and receives at its command input θi corresponding to ρi. Finally, the device 402 is a one-dimensional module STN (1), receives an input signal 401 (ρ with respect to _θ , expressed as ρθ), and is controlled with a control signal (Barρi).

The input value Dis is equal to the distance of the object relative to sensor 2 (as camcorder) or sensor 2. ′ (as radar) (FIGS. 1, 2, 4a, 4b, 4c, 5, 17, 18). This input value Dis is measured externally 25 by, for example, a laser attached to the sensor, and converted into a logarithmic function value. Both parameters Dis and ρ are important input parameters of the device of FIG. Instead of using a laser, the distance Dis may be accurately measured, for example by binocular convergence, using the parallax, using the recognition of the actual size of the perceived object. .
As illustrated in FIG. 42, multiple groups of STN modules may be operated continuously with the output of the dynamic replenishment module 440. In that case, the sectorization of FIG. 55a is obtained with the apparatus of FIG. 54, as illustrated in FIG. 43a corresponding to the arrangement of FIG. Therefore, there are zones or sectors such as Zr0, Zr1, Zr2, Zr3... Zr7. Conversely, instead of using the barycenter BarZ0, sectors -Zr0 to Zr7 are arranged around the points ρ0 and LD0 defined by Bar ρ0. Figures 55b and 55c also correspond to Figures 43b and 43c, respectively. The centroids such as BarZ0 and BarZ1 are replaced with points ρ0, LD0, and ρ1, LD1 to become Barρ0 and Barρ1 in zone Zr3, respectively. When comparing FIGS. 43b and 43c with FIGS. 55b and 55c, the coordinate axes x and y are actually replaced with the coordinate axes ρ and LD. At the end of the sequence, FIG. 55c, the calculated values ΔLD, θ, Cρ ′ are written to registers 1070, 1071 and 1072, respectively, and a single single-line module STN representing FIG. 55d. Only (1) remains as it is.

The sectorization in the case of FIGS. 54-55c is therefore made from the first measurement ρ0, LD0 and from the second measurement made from a different distance than the first one, and ρ1 and ρ0LD0 on FIG. Output LD1. Orientation is defined by a straight line ρθ perpendicular to the direction of a pair of points ρ0, LD0 and ρ1, LD1 as illustrated in FIG. 55c. Thus, there is Cρ ′, which is the output of the one-dimensional module STN (1) 402, corresponding to the projection of a straight line running through a couple of points, following a direction perpendicular to ρθ. The value of Cρ ′ corresponds to the protrusion of a straight line passing through a pair of points ρ0, LD0, ρ1, and LD1 on a straight line having a slope Cρ perpendicular to the direction passing through the pair of points ρ0, LD0 and ρ1, and LD1. What is important is that Cρ ′ is independent of the distance Dis, which does not depend on the size of the object to be observed, and therefore results that do not depend on the relative position to the sensor.
FIG. 56 illustrates the measurement results obtained by the apparatus of FIG. 54 that operates as described with respect to FIGS. 55a, 55b and 55c. There are two planes on FIG. 56, a first surface having coordinates ρ and LD, and a second surface having coordinates ρ ′ and D. ′ by rotating an angle θ from the first surface. .

By way of illustration, for successive distances of 50 cm, 75 cm, 100 cm and 150 cm, the points are substantially aligned and have a value of the coordinate Cρ ′ with respect to the axis ρ ′. Regardless of the distance (within tolerance) of the object to be observed, and hence its obvious size, the module ρ ′ is practically invariant.
FIG. 57 illustrates the progress of the apparatus of FIG. 55d. On this Fig. 57, first notice that the device is operated with 55d signals LD and ρ of all the devices 400 in the diagram, and has (its input Dis and its output LDi (ie, the logarithm of distance Dis). LD comprises (parameters of device 410 including the assembly of devices 401 and 402 of FIG. 55d, and further improvements consisted of devices 403, 404, 405, 406, 407 and 408. Subscript i. 55d with devices 401 and 402 having only output C (), the output of assembly 410 consists of the values Cρ and LD ′. Indeed, the device of FIG. This includes processing because it also controls the distance that can be used (this process being looped), a measure outside the distance Dis is available, or an internal calculation of it in the opposite case, from this point of view. The multiplexer 403 receives the LDi from the device 400. Choose one of these input signals on the described (output of adder 406) and on the calculated signal LD ′.
If the logarithmic value of distance (ie, LDi) is valid, the multiplexer 403 transmits this value if it does not transmit the calculated value LD. This output the signal being applied to device 401, similar to device 401, in FIG. 55d.
The selection between both these values is a command signal 403c (from the device 408 that receives the input signal LDi and analyzes its validity and is moved during the learning phase at a minimum) Ordered by.

When the external value LDi is valid (device 407 comparing the incoming value of LDi with that of LD), the value LD 'is calculated before it is calculated.
This device 407 commands K2 405 equal to −tg (/ cos () for the first for the increase of two amplifiers K1 404 and the second and to one.
Adder 406 adds the values −ρi × tg 鐔 ochi and Cρ ′ / cos 鐔 ochi (sum of these values equal to LD) ′ (add the subscript i for the parameter on FIG. 57 ((On one hand and Cρ and LD '', on the other hand). Device 407 corrects the coefficient K1, and the amplifier K2 reduces the difference between the calculated value LD'i and the input In order to evaluate LDi.
If the main interest of the device of FIG. 57 with respect to the device of FIG. The option to make a new object within a known object to a certain size without measuring the distance.
There is therefore a lot of computation, C (as the module to be stored.
Except for the integration of module 410 of FIG. 57 to calculate size invariance, FIG. 58 corresponds to FIG.
All modules M occur (or (the accompanying module 410 generating C (is preceded by 'and LD' in FIG. 47d).
The value LD ′ makes it possible to recognize the quantity side by their associated variation as well as the distance by calculating their average value.

With regard to invariance, one process (using calculation from BarZ0) has been solved so far, by measuring the distance to solve the size problem, the device 410 of FIG. 57 allows invariance. Translational invariance against
For the rotation invariance, the last invariance that is attached is the latter figure where this problem is solved by the device, which is going to determine the object of that distance and its position in translation as well as in rotation. 59.
The Le device of FIG. 59, on the other hand, consists of the first of all bivariate linear modules STN (2) 800 receiving C (') and the other (' and the assembly 900 It is composed of another single-line module STN (1) 901, a counter 902, a multiplexer 903, and an adder 904.
The role of the device of FIG. 59 is to add double correction to the couple. ((
(As previously identified, C (changes to ', but (deforms (by device 900.
During the search for the phase (counter 902 starts for each value (and the signal starting the search phase, which is incremented by 2 (+1 radians), multiplexer 903 causes counter 902 to Command the transfer of values from modulo 2 to 904 of the adder (radians, which calculates the value ('.
Couple C (', ("The classification of the STN module 800 that matches the already known value to be calculated instructs the counter's STN module 901 and the confirmation of the current value to be stored in its histogram. Later module.
At the end of the refocusing, the POSRMAX content corresponding to the average value ('Multiplexer 903 has been replaced, and generally obtained more than maximum resolution (module 90 analysis of the resulting STN histogram)) This is a new calculated value that can be sent to the adder 904 to allow recognition of the object. The value ((it exits the assembly 900 identifying the object's rotation).

Referring to FIGS. 60a and 60b, for the assembly, the portion of FIG. 58 before generating the triple C (represented in FIG. 60a '(the (LD) of the block MM in FIG. 53 is the size and Improved by incorporating the integration of rotational invariance calculations.
Indeed, the sub-module M (0) of FIG. 53 is integrated with the single-line module STN (1) 920 for calculating rotation (and the single-line module STN (1) 921. And It calculates the average distance LD '(module 800 corresponding to the STN module 800 of FIGS. 53 and 59), whereas the module 810 of FIG. 60b is replaced by the STN module 810 of FIG. (Figure 60b)
6 Illustrative diagram, object and sensor 2-5 rotation approval by diagrammatically increasing resolution, or object OB1 and OB2 shape, memory of tissue with angle D1 and equal D2 Thanks to the generation of relevant self-centered references, 2-5 '.
The sensor rotation actually reduces the resolution of all spaces first (period T1 in FIG. 3).
In the event of rotation, within the relatively high value of the signal Dif, the sequence for stopping the rotation is emitted by means (not represented), and the processing of Gaussian increase in the above-mentioned resolution of space Begins (period T3 in FIG. 3).
Also, the rotation angle of the sensor is determined and stored.
This rotation may be realized by the following two axes, and this, in essence, requires the storage of two angles, as well as the order of rotation following both axes.

However, by choosing the axis of rotation of the so-called learning phase sensor, the order of rotation following both axes selected is fair, and the information stored is then just rotated (it is advantageous ) Pulled down to both two angles. -The reason is as follows. Then you can reconstruct a true self-centered reference.
Finally, with reference to FIG. 62, it was explained what sensor 2 perceives (FIGS. 1 and 2) and the results obtained by the method of the present invention:
Each object is as follows
* The label is named LABEL (ie 1, 2, ...),
* Placed at a distance by the value of LD (1,2,3 ...),
* Placed in location by BarZ (1, 2, 3 ...),
* Rotation position (((1, 2, 3)
And
* Characterized by its color C (1, 2, 3).
Obviously, the sensor recognizes part of the environmental scene depending on the location and focus on that above, which means that the concept of <object> also depends on both these parameters .
As well as being able to change the sensor into a frame or a continuous sequence, the entire person or his surface, or his (woman) surface (different each time << object >> ) Part of.
A preferred embodiment is the method and apparatus of the present invention just described, and it is a general tool applicable to other types of data and / or input parameters. At the end of this, we will be able to obtain the results described below. An application that draws a certain number.
A) The main results are as follows:

First
For the method and apparatus described with respect to FIGS. 1-16, ie essentially for the assembly of FIG. 1 or 2, performing digital signals. (In particular, digital video is subject to spatial and spatial results (spatial resolution, substantially Gaussian increase in it, defocused (Figure 1), electronic filtering (Figure 2), and smooth Realized for the same location of the increase (substantially Gaussian) between two consecutive sequences on the other hand (spatial and this time differentiation) with the same effect As an increase in spatial resolution being represented for this time sequence representative of a continuous figure of the object after a decrease in reduction and subsequently, on the other hand, representative of the location on the place to attract the heart) Signals that make it possible to obtain the derived digital signal M. '. Hierarchy that is increasingly specific to objects (Figures 13a-13b and 16)-Representing the organized details and thus the resolution And, in general to increase the resolution of (3), stearyl - thanks to the discount di (generally sudden events), replacing the digital first signal,
Where
Without any hierarchy, all the details were present at the same time, therefore the importance according to the hierarchy that can be materialized by the appropriate method, with a tree with a certain number of parent-child associations, followed by the details Of the object that it represents by means of an extracted signal whose reduced order from the most features is evident in the details of that object's least characteristic, i.e. making a detailed view Difficult to use to describe features.

No. 2
That is, it has essentially a more sophisticated method and apparatus as described with reference to FIG. 17 and the next part. Either having the assembly of FIG. 17 performing optical variations of spatial resolution (by defocusing and refocusing), or having the assembly of FIG. 18 with variations by resolution electronic filtering ,
Extracted signal M. 'above. The format includes one or several STN modules (where the preferred structure is described with respect to FIGS. 20, 23a-23b, 23c and 25), one or several different levels of data Processed by several histogram representatives of the weight (ie, large amount) included in the derived signal M. ′. Thus, the first digital signal (especially the video digital first signal) (which determines the successive zones corresponding to the details appearing by reducing the order of importance of the requested signal. Allows the first to reduce all, and therefore the importance, to calculate the position of these zones with their centroids determined from their weight of their sequence of events Conveniently, provisioning hierarchy-organised the representation in the form of a tree with parent-child relationships (Figures 44c-44e to 49).
For the determination of the center of gravity, the feature s of the observed object is executed for each translation.

Third
Preferably, the sectoring of the corners (forming 34-43c) switching from the Cartesian coordinates x and y to the polar line is coordinated. (And determining (or p) a relatively large corner zone to enable is the first zone appearing (including its most striking features), then relatively large Narrow sub-zone covering the zone, inside the new center of gravity was found, and concentrated on the center of gravity of the first zone, next to the first parent-child relationship These corners are defined to determine the center of gravity of the subzone (absolutely any rotation in each.
This method is repeated for the following successive zones having their centroids.
4th
With the exception of parameter x, the same is then sectorized and then y ((other than LD, no longer (current D. '., ("Determining invariance Yes, each of the objects to the sensor belongs to the object with respect to the distance of the scale.

5th
In the preferred embodiment, the description is performed by four consecutive sequences of objects with their features appearing by reducing the critical order. By changing the x and y coordinates to allow for the calculation in (and (the feature of the direct direction, ie its features, the first of all the most important ones, then the least important ones) By defining the elements that are related to this object.
At the end of this four-step process, the tag that defined the label or object as well as the determined tree with the parent-child relationship was examined.
If there is a new object that is examined to quickly check, it is then possible whether this object corresponds to a substantially previously stored object, and at the beginning of a possible approval, In order to switch directly from the inferior to the fourth sequence with a test of identity between the elements of the already stored objects, the third sequence is suppressed and a faster dynamic method is used. The corresponding element of this fourth sequence of new objects can now be implemented and will now be the third sequence.
It should be noted that for the features whose labels are known, these features of the object, ie each of the objects, are related by reducing the order of importance. Part of the matching label.
* Therefore, during the learning period, i.e. in the first object observation, there is a running sum of its features in the label.
* Approve about it to examine a new object, ie, whether it corresponds to an object that has already been analyzed and labeled. The appearance of an already known label makes it possible to extract the features associated with this label for faster approval.

6th
Following two angles (it performs defocusing for period T0 in Figure 3) is to determine the event of the first important feature of the observed object (thanks to the high value of the signal Dif) Sensor rotation to enable.
Sensors that have stopped in their rotational movement, due to refocusing due to reduction, the details are already important with respect to the important order of information about the details and at the same time these objects have previously been examined in the form of labels It is clear when compared with the stored information. It is likely that skipping will facilitate the approval of new objects with known objects and their comparisons, as well as speed up the testing of new objects.

7th
Finally, it is possible for each of the objects to get complete information of its rotation, in scale, for the translation and for previously exposed raisons.
8th
From the zone Z0 representative of the scene, and its center of gravity BarZ0, to the scene where it can be seen that it is no longer included in the object and is no longer containing some object (the method of the invention) Those who have the ability to determine the localization of different objects in the scene ((relatively associated with BarZ0 and have that object, remember the scene in the form of a label, by will it then Allows you to determine whether a new scene to be observed is composed or not labeled by the first scene.
This kind of method of traveling is particularly useful in navigation where the principle is exposure below in the frame work of the use of the present invention.

B The method and apparatus of the present invention will be described with reference to finding a number of applications referring to some of them via a non-limiting example (surface approval)
Thus, one finds a privileged use of the present invention.
The perception of the surface and its associated label collection allows for a new surface label with the surface label already examined for comparison.
Memorizing surface labels that can be inspected can be implemented in calculators, chip cards, magnetic identity cards or bar codes, etc ...
From this kind of information stock of surface labels, it is possible to find out whether a new surface (which has just been examined by the method and apparatus of the present invention) corresponds to a label already stored.
It is so much to allow individuals for a variety of reasons, especially to localize missing persons, or to allow individuals to access determined zones where access is selective. Is possible.

Other applications within the surface approval framework work are as follows:
At those chips (owner's label), as they join, the traditional type of credit card is finished.
The latter appears in front of the cash enforcing machine and is detected by a camcorder or webcam when inserting a credit card based on his (female) chip that includes a label on his (female) surface Including a camcorder or web cam after performing the method and apparatus of the present invention to determine if the surface image corresponds to the label entered on the credit card chip inserted by the client. Cash dispensers are competent.
This is forbidden if the card is lost or stolen. To its owner, the surface of the user that is not authorized by another person does not correspond to the label stored on the credit card chip.
Person labels can also be inserted in identity cards, driver's licenses, car ownership documents, etc.
Similar applications can be seen in mobile phones that include visual sensors.

With its first use, by a real client, the phone can determine the label of it and probably authorized one or the other.
By comparing the label stored with the user's label, it can then be checked whether the current user is an unauthorized person with his (female) possession. After the call, steal or lose the phone.
In this case, the phone can send the unauthorized person's label to the network operator for the intended object of the mobile phone, possibly locking the phone. it can.
Finally, it can be seen that the method and apparatus of the present invention allows identification of a person for multiple applications.

The present invention can also be applied to photo-library indexing or image libraries containing multiple photos.
In the first phase, one or several labels are associated with each photo or image (for example, a label corresponding to << or cow >> or << sunset >>).
Alphanumeric references stored in << dictionary >> and photo reference numbers facing the matching labels and assembly labels, or stored in the image memory, are stored in the second -Associated with each label.
When a user wants to find a picture or image in a cow or sunset containing them, he / she inserts a keyword identifying the cow or sunset To do.
A photo-library dictionary translates this keyword into an alphanumeric reference. And it makes it possible to find in the memory the number of photos that contain the keyword.
Obviously, when a user wants to find a photo or image that includes some features (eg sunset and cow), he / she simply typed both corresponding keywords. I need.

Another use of the invention involves navigation.
For example, in the case of sea navigation (methods and users of the device of the invention) in the case of what is known in the bearings and having a sensor, the environment to be examined (especially a harbor or a jagged) The beach) and this environment is labeled.
The sensor then moves beside the sea map, and in the case of identity it is a observable location between the zone of the sea map zone environment and the label, which is -The location where the target is located can be determined.
Other applications for navigation, such as searching for when to return to the starting point, are essential in localizing the itinerary.
Thus, as it moves and for the corresponding environment, the device of the present invention allows a continuous position on the label.
During the return journey, if << Navigator >> is an already visited location of the device, then an attempt to search in the environment observed for the sensor will allow a trip away .
The invention is formed below by the claims,
Where
In the above description, the term << imposed >> imposes that only the object is adequately well imaged or scene (especially for navigation applications), possibly a person or person, especially a surface And did not mention to speak, except for some of the animals.

Utilizes optical means for increasing the spatial resolution stepwise and substantially Gaussian for a certain period, a unit for obtaining a Gaussian amount difference by smoothing spatially and temporally, and the difference 1 illustrates a first embodiment of an object recognition or recognition device according to the present invention comprising means. A means by an electronic filter that increases the spatial resolution stepwise and substantially Gaussian for a certain period, a unit that obtains a Gaussian difference by smoothing spatially and temporally, and uses the difference FIG. 2 illustrates a second embodiment of the object recognition or recognition device according to the invention, comprising means for The curve shows the required variation in spatial resolution, including increasing in steps between r _min and r _{max over a} period of time. FIGS. 4a, 4b, and 4c illustrate three embodiments of the optical assembly of FIG. 1 having means for controlling the change in focal length. The example which applied this invention to the synthetic aperture radar system is illustrated. 2 illustrates an example of the electronic filter of FIG. Fig. 7 represents the input signal of the electronic filter of Figs. 8a, 8b, 8c, 8d and 8e show the output signals of the electronic filters of FIGS. 2 and 6 when the filter order w is continuously changed to different values, FIG. From FIG. 8e, the order w decreases while the resolution increases. FIG. 3 illustrates an embodiment of a Gaussian difference unit with spatial and temporal smoothing of FIGS. 1 and 2 with output signals CO and DP. The histogram showing the absolute value of the difference between the image immediately before the increase in spatial resolution and the image gradually increasing the spatial resolution on the horizontal axis is shown by a curve, and the useful part of the histogram is shown by Li. While a still object is illustrated as a recognized object, a change in the absolute value of the difference (shown on the horizontal axis in FIG. 10) is indicated by a curve using the filter order w of FIG. 6 as a parameter. While showing the face of a person as a recognized object, the change in absolute value of the difference (shown on the horizontal axis in FIG. 10) is shown as a curve with the filter order w of FIG. 6 as a parameter. FIG. 13 illustrates an aspect in which the partial views of FIGS. 13a and 13b are combined up and down to form an overall view, and FIGS. 13a and 13b illustrate novel changes in the human face during processing. ing. FIG. 11 is a table of cumulative values calculated for CO in the case of DP = 1 at successive stages t of processing including a spatial resolution increase period by changing the filter order w shown in parentheses in five stages. The stages with different cumulative values in the table of FIG. 14 are illustrated by curves a, b, c, d, and e. The continuous space components a, b, c, d, and e in the period T ₃ -t ₅ (FIG. 3) when r _max and w = 0 are illustrated. An additional set formed from a number of histogram-forming STN-type units has been added to allow for object recognition and positioning, and a detailed object recognition or recognition device compared to FIGS. FIG. 2 illustrates an embodiment and details the embodiment of FIG. 1 with an optical system that gradually increases the spatial resolution. An additional set formed from a number of histogram-forming STN-type units has been added to allow for object recognition and positioning, and a detailed object recognition or recognition device compared to FIGS. FIG. 3 illustrates an embodiment and details the embodiment of FIG. 2 with an electronic filter that increases spatial resolution. FIG. 19 represents a graphical representation of a one-dimensional histogram decision called STN (1), ie a one-variable linear STN unit, of the additional set of FIGS. FIG. 19 shows a detailed view of a one-dimensional histogram decision STN unit called STN (1) in the additional set of FIG. 17 and FIG. 18; 21a and 21b show two one-dimensional histograms with one peak or maximum value and two peaks or maximum value, respectively, as determined by the STN (1) unit shown in FIGS. 19 and 20. An example is shown. 2 represents a graphical representation of a histogram-forming two-dimensional or two-variable linear unit. FIG. 23 illustrates an aspect in which the partial view of FIG. 23a and FIG. 23b are combined with each other along the line Z-Z 'to form the overall view. In combination with FIG. 23b, the additional set of FIGS. 17 and 18 illustrates in detail a histogram decision two-dimensional or two-variable linear STN unit called STN (2). In combination with FIG. 23a, the additional set of FIGS. 17 and 18 illustrates in detail a histogram decision two-dimensional or two-variable linear STN unit called STN (2). The partial modification of the classification subunit of the STN (2) unit shown in FIGS. 23a and 23b is illustrated. Fig. 24 represents a two-dimensional histogram with two peaks determined by the STN (2) unit of Fig. 22 and Fig. 23a and Fig. 23b. FIG. 3 is a schematic diagram depicting important elements of a self-adaptive module STN (2) having a prediction function. Fig. 21 illustrates a set in which classes can be selected sequentially in the case of more than two classes corresponding to more than two peaks according to Figs. 21b and 24; FIG. 26 illustrates an improved set of FIG. 26 to allow fast selection for more than two classes corresponding to more than two peaks according to FIGS. 21b and 24. FIG. FIG. 6 illustrates a flowchart of steps for displaying three or more classes in succession in the integrated API and displaying the results in the framework of the application. FIG. 6 illustrates a flowchart of steps for displaying three or more classes in succession in the integrated API and displaying the results in the framework of the application. For more than two classes, represents an assembly that performs the computation phase without sorting the classes. In the case of more than two classes, the classes are sorted to represent an assembly that performs the calculation phase. Fig. 29b illustrates in detail the memory of Fig. 29b. Part of the set of FIG. 30 is shown in detail. Part of the set of FIG. 30 is shown in detail. Illustrates an assembly that implements an STN module with multiple class extraction capabilities. Illustrates the assembly of the STN (2) unit and the two directional units pα, pβ of the plane generated in the STN (2) unit for determining the average position or barycenter (BarZi) generated in the bus To do. The boundary determination of the object OB by the zone defined from the two axial directions pα2 and pβ3 of the surface and the determination of the center of gravity BarZ0 of the zone are illustrated. A band between limits A and B in different directions, operating at the center of gravity BarZ0, output from the set of two-dimensional units STN (2) of FIG. 34 (generated on the common bus) and determined according to FIG. 34a. Illustrated schematically. Figure 23 illustrates an improved classification device of the assembly 101 of Figures 23a and 23b. A zone bounded by the output 101s of the assembly 101 is illustrated. FIG. 37 illustrates a zone delimited by the apparatus of FIG. 36 instead of assembly 101. FIGS. 39a, 39b and 39c show the relative position of BarZ1 generated according to FIG. 36 relative to the position of BarZ0 generated by unit STN (2) of FIG. 34 and determined according to FIGS. 34a and 35. Figure 3 illustrates the three stages of determining polar coordinates ρ3 and α3. The parent-child relationship between the upstream BarZ0 and the downstream BarZ1 is illustrated using polar coordinates ρ ₃ and α ₃ determined according to FIGS. 39a, 39b, and 39c. 42 shows a set formed of a module STN (1) and a module STN (2) used in the apparatus of FIG. Fig. 4 illustrates an apparatus for determining polar coordinates by performing the method illustrated in Figs. 39a, 39b, 39c. 43a, 43b and 43c illustrate the sequential operations performed in the electronic assembly of FIG. 42 to determine the position of BarZ1 (child) relative to BarZ0 (parent) in Cartesian coordinates. 44a to 44e, 45 and 46 relate to zones Z ₀ to Z ₃₀ included in each other, and FIGS. 44a to 44e include each other (as illustrated in FIG. 45). Mareta, thus illustrating the successive stages to determine BarZ30 from the center of gravity BarZ0 consecutive corresponding to Z ₃₀ from the zone Z ₀ in unbranched form dynamic tree (as illustrated in Figure 46). It is determined by the apparatus of FIG. 42, illustrating the relative positions of BarZ30 from the center of gravity BarZ0 of Z ₃₀ from the zone Z _0. FIG. 45 illustrates the unbranched tree from the center of gravity BarZ0 to BarZ30 in FIG. FIGS. 47a to 47d, 48, and 49 relate to zones Z10 to Z30 that are not included in each other but are all included in zone Z0. FIGS. 47a, 47b, 47c, and 47d Are not included in each other (as illustrated in FIG. 48), and therefore (as illustrated in FIG. 49) are successive centroids BarZ0 corresponding to zones Z ₀ to Z ₃₀ that are in the form of a branching dynamic tree. Illustrates the successive steps to determine BarZ30 from Determined from Figure 47a in the apparatus of FIG. 47d, illustrating the relative positions of BarZ30 from the center of gravity BarZ0 of Z ₃₀ from the zone Z _0. 48 illustrates the tree structure of barycenters BarZ0 to BarZ30 in FIG. Figure 50 illustrates the sequential determination of zones Z0 to Z21 and Z22 in the intermediate case with zones that are included and zones that are not included with each other, and Figure 50a is a tree structure with branches from BarZ0. Illustrate the parent-child relationship between BarZ12 and BarZ21. FIGS. 17 and 18 comprising a one-dimensional unit according to FIGS. 19 and 20 for processing parameters representing an observation object and a two-dimensional unit according to FIGS. 22 and 23a to 23b for processing coordinate parameters x and y. Illustrates part of the additional set. The temporal information determined by the apparatus of FIG. 51 can be maintained, while the values of CO, the line and column coordinates x and y, the color, brightness, tone and saturation (saturation) parameters L, FIG. 17 illustrates another part of the additional set of FIGS. 17 and 18 consisting of one-dimensional, two-dimensional, and three-dimensional units respectively relating to T and S. Represents a device for storing perceived objects only for translation invariance. Fig. 4 illustrates a unit for determining dimension or size invariance. FIGS. 55a, 55b, and 55c are shown in FIGS. 43a, 43b, and 43c, in which the module ρ and the angle α are predetermined according to FIGS. 43a to 43c and the logarithm of the input distance LD is predetermined according to FIG. FIG. 55d illustrates a unit for determining dimensional or size invariance. Figure 54 illustrates the results obtained by the method of Figures 54, 55, 55b and 55c. 56 represents an improved example of the size invariance determination device of FIG. 54 with distance logarithmic tuning capability that allows further operation to receive any input signal representing distance. FIG. 47 illustrates the improved example of FIG. 47d further comprising a size invariance determining device according to FIG. Fig. 2 illustrates a rotation invariance determination device. 60a and 60b are partial views of an assembly of an object recognition device having size invariance, rotation invariance, translation invariance, including the units of FIGS. 58, 53 and 59. Illustrates the angular displacement of a video sensor that switches from an observation object to another object to be observed. Fig. 3 illustrates a screen including three objects identified from the properties of the object and the position of the object.

Claims (197)

There is an essence in deriving the representative input digital signal from which the objects of the environment are composed of a succession of representative sequences of successive diagrams of the objects of the environment, and each representative of the location A method for recognizing and therefore relating to space areas, thus relating to timing each arranged region (each said sequence consisting of succession of results) after each other of the sequence, comprising the following features: Have:
The duration of substantially several sequences (this world variation in the spatial resolution of the object of the input digital signal), including a phase of Gaussian increase in resolution from a decreasing value to an optimal basic value Run while
For the smoothing effect between two consecutive sequences, the difference between the substantially absolute value of performing differentiation, for each same spatial location of the derived signal, exceeds a certain threshold To obtain a derived digital signal representative of the variability of Gaussian differences in both these sequences, Gaussian increases in resolution and
From the derived signal by comparison of successive sequences between the values of this derived signal corresponding to the exact same location of the result (starting with the most characteristic of the object, hierarchy-organized details) Derived from. (From the location where the input digital signal consists of a video sensor, the sequence in which a continuous image of the video signal is formed, the result in which the lines of the video signal are formed, and the location of the pixels in the video signal And the feature of the information transported by the digital signal relates to an object present in the scene observed by the sensor. The method according to claim 1 or 2, characterized in that the Gaussian increase in spatial resolution increases between two consecutive sequences in the stage, without any change in resolution between the sequences characterized therein. It is executed during the steps. Method according to any of the previous claims (characterized in that the spatial resolution variation includes :)
From the base value, the decreasing value before the phase that the increase phase is relatively slow from the gaussian increase substantially and from the decreasing value to the base value. Until the previous phase of the sudden reduction in the resolution of the space. Method according to claim 4 (characterized in that the spatial resolution variation comprises:
The phase of the phase of the sudden increase in phase with a substantially constant resolution and a slow increase in resolution between sudden events. A method consistent with any of the preceding claims, wherein the substantially spatial resolution Gaussian feature characterized therein is realized by electronic filtering of the input video signal generated by the sensor Is done. A method consistent with claim 6 is that the filtering formula h (w, k) = Cp. (W + 2- | k |) (w + 1- | k |) {-3k2 + (2w + 3) For the spatial resolution given by | k | + w (w + 3), Cp = 5 / {2w (w + 1) (w + 2 ) (w + 3) (2w + 3) with w}
((Standard deviation of Gaussian mean) = 0.3217w + 0.481. Consistent with any of claims 1-5 is marking the object that the sensor has a varying focal length and that a resolution variation is implemented by acting on the focal length of the lens A method that characterizes. Claim 8 (Substantially during the pitch of Gaussian increase in resolution (replacement of the lens being performed at a constant speed between two stages) from the object in at least one lens of its one movement) The method according to claim 2) characterized to modify its focal length. 8. Control of electricity (form of at least one lens to change at a constant rate at which its focal length substantially acts during a Gaussian increase pitch in resolution between the two from the object) ) Characterized in that it is modified by). A method characterized in itself to be consistent with any of the previous claims is a substantially Gaussian but increasing recurrent world and space method of performing differentiation ) To implement a substantially Gaussian resolution variation
-
Both values on a single hand (binary signal DP) representing a threshold above each and not in excess of the resolution for each same spatial location of the sequence between two consecutive sequences, Decrease in bits, representing the adaptation time re-entered to reduce the variation between two successive sequences in the method, forming a threshold and the derived digital signal by the difference in absolute value Having a number and a processed signal of difference in resolution, a continuous value of CO, an increase in resolution, a value representative of an over-threshold value (representing) related to a continuous sequence of DP When on the other hand (digital signal CO) for the same location of the sequence. Consistent with claim 11 is that for absolute values less than the threshold, the resolution of the space chosen for the limit L in the histogram (a sequence having a basic resolution and the number of points in the histogram includes: A method characterized in that one form decreases during the previous phase of a sudden event (in a histogram of the absolute values of the differences in the sequence with decreasing resolution later):
For the aforementioned phase of spatial resolution increase, the stage finds at least 75% fractional sensitivity of the total number of histogram points selected by the increase step p, which is not greater than the cheaper integer price due to the L / S defect, and the sensitivity The threshold itself is a decreasing phase at the beginning of the increasing phase of the S resolution given to the digital signal CO and a zero value at the end of the value p. A method consistent with any of the previous claims to perform is characterized in that the derived signal other than the recognition and local identification of the object of the environment and only for recognition At least a variety of locations (which are characteristic details of the object), which are derived from and detailed by forming at least two histograms of the extracted digital signal of the object. The information relating to the digital magnitude of the signal), and at least about what else relates to the local identification of the location of the signal. And it supplies information related to the location of the feature details. A method characterized by itself in conformity with any of claims 11 to 13 is a classification method which is essential to generate a histogram which sorts at least one pair of these values. Submit consecutive values of. The DP then has a value corresponding to the raised threshold (according to the different frequency events of CO and the other frequency due to local identification of CO). 15. Method according to claim 14, characterized in that at least one of the histograms is of a multiline type based on a classification of several parameters 16. The method x and the location y according to claim 15, characterized in that at least one of the histograms is characterized based on a classification of two line type coordinates. 17. A method as claimed in claim 15 or 16 (characterized in that at least one of the histograms is based on three parameters defining three line-type colors). Consistent with any of claims 13 to 17 is the manner in which the features are described in themselves associated with the dominant color in each featured zone, and it is in accordance with its coordinates and its dominant color at one time. Define each successive zone. Starting from the sequence of events in these zones (analysis tree including), each of these zones and their own centroids are defined as well as being consistent with any of claims 13-18 The method characterized in itself determines a continuous zone corresponding to less and less characteristic details of the object:
The normal origin corresponding to the center of gravity of the first zone with the initial one that has increased branching to the base resolution and the point corresponding to the center of gravity s of the zone with it has reduced resolution. The method characterized in itself to be consistent with any of the previous claims is that after rotation according to the matrix formula, the rotation of the axis x, the rotation and the axis rotation on the reference plane between y and Y before axis X Perform the action. A method consistent with claim 19, wherein the new zone is centered on the center of gravity of the zone where the corner sector was previously determined, in which sector of the sectors the new center of gravity is located; Dividing the search into several positioned one search, which sub-sector in which the center of gravity of this new zone is located in the sector, several substitute players-sectors, one After portraying features to divide into searches and to define the polar coordinates of this, the angle portraying the features of the subzone that previously contained the new center of gravity and both center of gravity s Center of gravity with respect to the center of gravity of the zone determined from the distance between. The method characterized by itself in conformity with claim 21 is followed by determining the polar coordinates of several centroids s of the zone appearing with respect to the first zone representing the object at the time of minimum resolution. And store the coordinates of these polar regions in the form of a label representative of the object itself. 23. The method of claim 22, wherein the new object observed therein is matched or stored and characterized to determine if it does not correspond to an object previously classified as 1 Decides to compare the label of this new object with a label in which the label of the new object according to this claim 22 and the method itself is stored. A logarithm LD of this distance D performing sectorization is characterized by subsectorization, characterized by matching any of claims 13 to 23 to determine the distance D of the object to the sensor. And the method of which the coordinates of the following are considered, and (according to claim 21 and itself is that the sub-zone has all the values of the LD dimensioned for the same object and the values to be deformed therein, I think I'm determined to be (ie C ()) is constant. An object of the environment including means for deriving a representative from it inputs a digital signal and consists of the succession of a representative sequence of successive figures of the object of the environment, wherein each representative of a location After that, an apparatus for recognizing and therefore relating to a space area, which is related to timing the arranged area (each said sequence consisting of succession of results), has the following characteristics:
Duration of several time sequences (worldly variations in the spatial resolution of the object of the input digital signal), including a phase of Gaussian increase in resolution from a decreasing value to an optimal basic value Means to perform during-
The means for performing differentiation, the smoothing means performing between two consecutive sequences, is the difference between the substantially absolute values for each same spatial location of the derived signal. In order to obtain a derived digital signal representative of the variability of Gaussian differences in consecutive sequences when a certain threshold is exceeded, the Gaussian resolution is increased and, in comparison, the extracted of consecutive sequences Means for deducting hierarchy-organized details starting from object features between the signal and the value of this derived signal corresponding to the exact same location of the result. 26. An apparatus consistent with claim 25, characterized in that the means for deducing the input digital signal comprises a careful video sensor, from which the input digital signal is formed. A video signal is deduced, presented in a location consisting of a sequence of video signals and a sequence of video signals and pixel locations of the video signal. An apparatus consistent with claim 25 or 26 wherein said means for cashing characterized therein substantially without any change in resolution during the sequence, Gaussian increase in spatial resolution, The stage generates its increase during an increase step between two groups of consecutive sequences. 25-27 (realizing the variation of spatial resolution before said phase that said increase phase is relatively slow substantially from Gaussian increase and from its decreasing value to its underlying value Characterized in that said means to generate a pre-phase of a sudden event reduction in spatial resolution from its basic value to its decreasing value). Consistent with claim 28 is characterized in that the means for realizing spatial resolution variation generates a substantially constant resolution between a sudden event reduction phase and a phase of a slow increase phase of resolution. Equipment. 30. An apparatus consistent with any of claims 25-29, wherein the means for cashing characterized therein is substantially the latter (where the input is for receiving the input video signal therefrom). In order to increase the electronic filter (connected to the output of the video sensor), a Gaussian increase in spatial resolution is included. An apparatus consistent with claim 30, wherein the filter has the formula h (w, k) = Cp. (W + 2- | k |) (w + 1- | k |) {-3k2 + (2w + 3 ) Cp = 5 / (2w (w + 1) (w + 2) (w) given by the formula, describing the feature of performing filtering with the resolution given by | k | + w (w + 3) +3) (2w + 3) with w}
((Standard deviation of Gaussian mean) = 0,3217w + 0481. Coinciding with any of claims 25-29 attaches the object that the sensor has a varying focal length and that the means for realizing spatial resolution variation affects the focal length of this lens. A device that describes the characteristics of Consistent with claim 32 is that it substantially corrects its focal length (replacement of the lens being performed at a constant speed between the two stages) during the pitch of the Gaussian increase in resolution. Characterized in that it comprises means for transferring at least one lens from said object. Consistent with claim 32, characterized in that the apparatus characterized by including electrical means to modify the type of at least one lens to change it from the object by a constant, the speed of its focal length is Raising substantially Gaussian, but increasing the resolution step between the two stages. A device consistent with any one of claims 25-34, wherein a substantially Gaussian increase in said means for realizing the characterized differentiation therein is a means of recurring world and A Gaussian resolution variation, consisting essentially of performing a spatial method)
-The resolution of each sequence for the same spatial location between two consecutive sequences on and not exceeding one hand (binary signal DP), where both values each represent an exceeded threshold , Threshold of absolute value, difference and
Forming a derived digital signal, a constant representing the adaptation time re-entered to reduce variation between two consecutive sequences in the method, having a decreasing number of bits, and resolution Difference processed signal, CO continuous value, resolution increase, value of the above threshold (representing) related to the continuous sequence of DP, of the same location in the result On the other hand (digital signal CO). 36. An apparatus consistent with claim 35, wherein said advance of abrupt event reduction of a sequence having a spatial resolution, a sequence difference absolute value having a base resolution and a subsequent decreasing resolution. The limit L, characterized by including means for generating means for selecting in the histogram during the phase, the number of points in the histogram for absolute values less than the threshold includes:
A fraction S of at least 75% of the total number of points in the histogram (means to choose) for said phase of increase in spatial resolution by stage (increase step p not greater than cheaper integer price due to defects in mass ratio L / S) Presence detection sensitivity threshold,
And a zero value means to disturb the digital signal CO at the beginning of the decreasing phase and the increasing resolution phase at the end of the value p. In order to allow not only recognition but also approval and local identification of objects in the environment, a device consistent with any of claims 25-36 characterized in that further, From the derived signal, the details of the object signaled by forming at least two histograms of the derived digital signal of at least various locations (which provide information related to the characteristics of the object) It includes means for deducing less and fewer features related to the signal, and at least about what relates to local identification of the location of the signal. And it supplies information related to the location of the feature details. Matching any of claims 35-37 corresponds to a threshold (according to different frequency events of CO and the other frequency due to local identification of CO) that exceeded at least one pair of DPs A device characterized by including means for submitting successive values of the signal CO in a classification method that is intrinsic to generating a histogram that sorts these values. 39. Apparatus according to claim 38, characterized in that at least one of the histograms is of a multiline type based on a classification of several parameters. 40. Apparatus x and location y according to claim 39, characterized in that at least one of the histograms is characterized based on a classification of two line type coordinates. 41. Apparatus according to claim 39 or 40 (characterized in that at least one of the histograms is based on three parameters defining three line-type colors). Matching any of claims 37-41 means that it is associated with the dominant color for each characteristic zone and means for defining each successive zone according to its coordinates and its dominant color at one time. A device that characterizes the inclusion. Claims 37-41 as well as the center of gravity of each of these zones (characterized as including features for determining successive zones corresponding to less and less characteristic details of the object) Any of the devices, according to
And
From the sequence of events in these zones, means to define an analysis tree that includes:
The normal origin corresponding to the center of gravity of the first zone with the initial one that has increased branching to the base resolution and the point corresponding to the center of gravity s of the zone with it has reduced resolution. 45-43 (which further comprises means for performing an axial rotation operation of the reference plane between axis x, y before rotation and axis X following the matrix formula (Y after rotation) Any of the devices that are characterized) 43. (To find out which sector of the sector the centroid of the new zone is located in, the new zone is centered on the centroid of the zone where the corner sector was previously determined; Means to divide the sector into a number of sub-sectors, the center of gravity of the new zone being within the zone The angle describing the features of the sub-zone that previously contained the new center of gravity after the means of searching for which sub-zone is located and meant to define the polar coordinates of this And the centroid with respect to the centroid of the zone determined from the distance between both centroids s. Consistent with claim 45 is that in the form of several centroids s of the zones that appear with respect to the first zone that represents the object at the time of minimum resolution and the label representative of the object. Apparatus characterized by including means for performing determination of polar coordinates of means for storing polar coordinates. 47. The apparatus of claim 46, wherein a new object observed therein is matched or stored and characterized to determine if it does not correspond to an object previously classified as it And means for determining means for comparing the label of the new object to the label of the new object and the previously stored label according to the means of claim 46. According to any of claims 37 to 47, characterized in that it comprises a distance D of the object or sensor, means for calculating the logarithm LD of this distance D, means for determining means for performing sectorization But the device followed the dyslexia coordinates by sub-sectorization, and (all claims 45 and all of the LDs where the sub-zone is dimensioned for the same object and the value to deform in it) Is calculated according to the means by which it is determined that it is (ie C ()) is constant. The amplitude RMAX of the maximum vertex of the histogram,
-The position of said maximum number of memory POSRMAX in the histogram,
-A set of classification limits determined with respect to a histogram added by examining the memory and the discovery of said criteria,
A means of processing the histogram to generate a set of limits, while addressing memory (100) addresses from the location of the maximum POSRMAX to generate a first set of limits with respect to the reference Performing increasing scans of addressable memory (100) to reduce the scan, while generating a second set of limits with respect to the reference, from the position of maximum POSRMAX Described, the limit presence was therefore determined from the position of the maximum POSRMAX in the histogram by scanning both of the addressable memory (100). 50. The method of claim 49, wherein the determination occupies status 1 of the following aspect:
-The position of the maximum POSRMAX for counting each step generating the limit in the address, the counting signal being added, and the reduced one scan,
-First, all first limit cycles of the address limit then instantaneously, the counting signal presence is the new counting POSRMAX during the first and then the second is added to the position of the maximum POSRMAX Investigate to generate a limit on two scans subtracted from the maximum number of positions to examine the signal,
-Then the first limit that generates all the second limit limits of the two examining cycles of the address, the counting signal presence increased the position of the maximum POSRMAX during the first scan, and The new counting signal presence reduced the position of maximum POSRMAX during the second scan. The method according to claim 49 or 50 describes the feature of allowing the means of selection (1162) to be performed for choosing a criterion with respect to at least one of the following values:
-Maximum value RMAX,
-The threshold value supplied to the module by THRESHOLD,
-A lot of histogram points NBPTS generated by processing the histogram and stored in the module registers. 52. A method according to claim 51, characterized in that the criterion is selected among RMAX / 2, THRESHOLD, NBPTS / THRESHOLD. Consistent with any of claims 49-52 is a method characterized at least in its one tool of the STN module:
-One adder / direction checking signal DIRECTION coincides with the binary value of the signal COUNTER generating the addressing signal COUNTER is offset to the value POSRMAX where the offset value is the maximum number of positions in the histogram A subtraction (108) that allows adding or subtracting the value of the signaling offset of the addressable memory (100), while or at least equivalent
-Comparator (1161) and, on the other hand, the logic circuit (1163, 1164, 1165, 1166, 1167, 1168) receiving the mentioned data item in the addressable memory (100) and the other hand On the fiance to generate a limit updating signal to allow verification of one of the limits according to the criteria and examining the direction,
-A selection multiplexer (105) arranged for addressing, an addressable memory (100), said multiplexer having three inputs each receiving an input signal carrying parameters, a counting signal COUNTER and The adder / decrement output (108) was input. A method consistent with any of claims 49-53, in that one tool of the STN module means for prediction having a calculation of the average position POSMOY related to the parameters and calculation of the differential, Before the execution of the operational subunit (101, 101 ') characterized (characterized above and of which the parameter is subject to generate an output signal (101s) for feedback with respect to a set of limits The parameter due to the difference between two consecutive means determined to be subtracted from the signed differential (145) (146). When the parameters match the average position POSMOY and for the second binary state in the opposite case, the method characterized in itself is further characterized by the first binary state. A centroidal device is implemented whose object is to generate an output centroid signal for having feedback. 49-53 (characterized by the fact that the parameters analyzed by the STN module are complex and that each itself obtains the latter by a combination of at least two basic parameter input signals DATA (binary numbers of A1A2A3 Of any of the complex parameters (A1A2A3 ... Ap) containing P fields corresponding to the basic parameters A1, A2, A3, ... Support Ap and that of the STN module to generate and store a set of registers P corresponding to the basic parameters from the position of the maximum POSRMAX in the histogram of complex parameters by scanning both of each memory Of one tool means. 57. The method of claim 56 (characterized in performing offsetting device (130)) wherein the device digital is addressed in memory (data corresponding to a particular field of a complex parameter). Allows to insert into counting signal COUNTER, offsetting enabling. A method consistent with claim 56 or 57 is characterized in that one tool means for prediction having a calculation of the average complex position POSMOYA1A2 ... Related to the calculation of a set and differential (A1, (A2, ...
Ap of each p-basic parameter of the complex parameter (A1A2A3 ... Ap) is determined by the difference between two consecutive complex averages and subtracted from each of the base parameters matching Sign differential unit (145, 146) prior to execution of motion subunit (101) targeted to generate an output signal (101s) for feedback related to a set of bounds did. It would be in accord with claim 58 when the complex parameter (A1A2A3 ... Ap) matches the average complex position POSMOYA1A2 ... and for the second binary state in the opposite case. The method described in FIG. 5 further implements a centroidal device that is intended to generate an output centroid signal for feedback having an initial binary state. 60. Operational subunit (101 ′) comprising means, characterized in itself to be consistent with claim 58 or 59, that allow the generation of output signal classification room (133, 134) freely in combination. AND (131) or classification zones, respectively Zand and Zor OR (132) of each basic parameter. Matching any of claims 49-60 is the first corresponding to a histogram zone in the form of two-state binary data in the classification memory (118) whose number of class words corresponds to the size of the histogram Addressable memory (100) of module STN (first state corresponding to reference discovery and second state to reference non-discovery on histogram) with class having a state value between limits The method delineated in that one store and in itself sends the output signal (101s) of the classification memory (118) on the feedback bus (111). Functional calculations that the addressable memory (100) of the STN module during a given calculation cycle that has an initialization value related to the value stored in the addressable memory (100) at the end of the previous calculation cycle 62. A method characterized by means (120) characterized in one of its initializations to be consistent with any of claims 49-61. 62. A method characterized in itself to be consistent with claim 62 performs a function to calculate equal initialization values (Km-1) / Km times stored value zero or more in Km format We get m equal to zero, corresponding to 2m with m, allowing memory effects above some m0 and lack of memory effects. A method characterized by itself in conformity with any of claims 49-63 comprises at least determining a given histogram into which a set of data is classified, the data comprising the vertices and amplitudes of the histogram, and Each class corresponding to the position of the vertex and one of the classes corresponding to the maximum vertex (RMAX (POSRMAX)) of the histogram are stored. 65. At the maximum vertex of a method histogram according to claim 64, characterized in that it determines a set of data classes between several STN modules and distributes the means to authorize them for storage The first module (ST1) that has determined and remembered the first class,
Then, the second class corresponding to the highest second is determined, the lower vertex of the stored second module (ST2) histogram is determined, and one of the following modules stored is stored. Histogram of modules in the larger vertex class prohibiting tuples and vertices in so-called output (ST3 ...). 65. A determination depicting the characteristics of a method consistent with claim 64 enabling a means to determine and store a set of data classes between several STN modules to distribute means. And storing a set of classes, data corresponding to one of the classes (thus determining each class it has between modules and storing modules (ST '1, ST'2, ST'3 ...) allocated class) towards the set of modules received (ST'1, ST'2, ST'3 ...) The first module (ST'0) that is sending the class-by-class classification basis earlier in reducing the order of histogram vertices. 68. A method consistent with claim 64 comprising a set of data classes between the STN module and an application programming interface (API) to distribute means, the memory M0 of histogram values determined by the module Means, memory M2 class depicting features of enabling to determine and store the memory M1 of the address aligned to the amplitude of the vertex of the histogram (class M3 with different number of classes and memory RC3 of the register RC Enables API execution that allows storing order numbers):
-Initialization cycle resetting memory M0, M1, M2, M3 and register RC,
-Calculation cycle for loading on M0, the value of the histogram determined in the module,
-Cycle to update class,
The above updating cycle consists of:
(A) Decreasing the order of the memory M0 and sorting the amplitudes when storing the matches referred to in the memory M1
(B) (Search for classes by :)
Tag the class in memory M2,
And
Storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C) (Verify the class of memory M2 with the address of the considered class of M2 having the threshold of M3 corresponding to the class considered by comparing the values of memory M0). 68. A method consistent with claim 67 corresponds to the addressable memory of memory M0 and MSTN module, and the corresponding data is a single function table of amplitude and position memory couple (600) Describes the features that can be put into (RMAX0 / POSRMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ...) The peak of the histogram in decreasing the order of the amplitude (table of functions (600) performing the class mechanical hardware sort during the calculation step). 69. A method according to claim 67 or 68 (characterized in that it is assembled in a module STN in a device with multi-class functionality)
-Table of one function of amplitude and position memory couple (600) (RMAX0 / POSRMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ...) Amplitude, table of functions performing mechanical hardware sorting of class during the calculation step (600), of addressable memory replaced with a unique table of said functions Histogram vertices when decreasing order
-Memory M2, which allows the order number of classes in the store
-Class threshold memory M3,
-Class number register,
-Application programming interface API,
Said multi-receive receiving at least one simple (DATA (A)) or complex (DATA (A1 ... Ap)) parameter, validation signal (VALIDATION), linear combination of feedback signals and arranging Class function devices signal (INIT, CALCUL, END, CLOCK)-Each multi-class device that is sending back at least 1 outputs each corresponding to a class (Cl1 ... Clk) on the feedback bus The signal got stuck. Matching any of claims 64 to 69 is that it determines at least one class with at least two multi-class modules (the first module (660) (664) operating in this world TD), Acting in a spatial domain SD to be performed in an object approval system including replenishing for at least one half module (661, 662, 663) (665, 666, 667) of the class A method that describes the characteristics of Claims 49-70 (characterized as being performed by a sectoring, zone determining module and centroid in an object recognition system including a set of histogram calculation and processing STN modules) Regardless of which sector of the sector that divides the determined zone into a number of corner sectors that are centered on the corresponding centroid of the zone and its one search, new and visible And, itself divides the sector into several sub-sectors, but the method can continue to tweak sectorization for further upscaling. A method consistent with claim 71 characterized in that it is performed in at least one STN module with at least two orientation devices p (p (p (150, 151) for reference The first of the input parameters (plus the module determining the centroid for the input parameters) of the input coordinates (and the axis rotation of the system 1 includes the centroid BarZi by the association of the univariate linear module with the system 1 axis rotation Determining the space Zi (296) processing the first parameter and determining the second association (298) (299) of the second two line module (297), processing the coordinates Inside the first space Zi second center of gravity BarZi + 1, the first space divided into distinct corner sectors (Zri0, Zri1, Zri2 ...), Zi ( 2 of the center of gravity BarZi + 1 of BarZi, further events each sector being processed by a module of the two lines of the sector receiving the signal) (300, 301 ... 307)
BarZi + 1 presence allowed to divide a sector that owns the second center of gravity BarZi + 1 into a sharp corner sector in relation to a set of two-line modules in a later rank sector The two-line module of the sector corresponding to the second centroid placed to be fine-tuned to uniformly determine the sectorization for ascending and at least one more angle (Zra, Zrb, Zrc ...) (substantially the center of gravity s BarZi and BarZi + 1 combined with the reference axis (1071) perpendicular to the direction of the module Authorize to obtain a translation (translational invariance) according to the (1070) angle of the distance between the centroids BarZi and BarZi + 1. Claim 71 or 72 (characterized in that the object can be observed from different distances of the system and at least one size invariant device (450) (451) (410) and its one tool) The method described at least with one input, the value of the logarithm (400) of the distance LD between reference points and the size invariant device received at least one point of the object, and the module (substantially The distance between the two centroids s BarZi of constant projection and corresponding to the angle s BarZi + 1 (of the device determining at least one value C) (of the rotation with respect to ( And LD. Consistent with claim 73, in that one tool of the size invariant device (410) means for controlling the distance freely allowing the use of an external measuring distance or the internal determination of the distance, The method depicted. Method according to any of claims 70-74 (characterized in that the object can be observed according to different angles of the system and at least one rotation correction device (900) and its one tool) Allows correction of the device angle value (of the reference axis ((angle and module previously determined for a set of values). A method consistent with any of claims 49-75, comprising a set of histogram calculation and processing STN modules having at least one device for transforming a parameter reference system by rotation of an angle 2D for polar coordinate parameters of pixel X (Y) corresponding to the following matrix actions (characterized at least a reference above the two dimensions of the parameter) characterized as implemented in the object approval system Rotation for a simple reference: A method characterized in itself to be consistent with any of claims 71 to 76 is that, over time, several centroids s of the zone appearing with respect to the first zone and its own first centroid It is decided to store the barycentric coordinates in the form of approval data. Matching claim 77 determines the new object and one authorization data to determine if the observed new object matches or does not correspond to an object with a previously stored label A method characterized to do that compares it with previously stored label approval data. Claim 77 or 78 (characterized in that it is further associated with label approval data) and a method of matching the angle of rotation (by means (920) (921) and the average distance LD It allows to determine the angle of rotation (and the average distance LD). A method characterized in itself to be consistent with any of claims 77-79 is provided by a processing module (670) capable of determining and storing a histogram calculation and a set of categorization data for the label. Analyze the label. -Amplitude RMAX of the maximum vertex of the histogram,
-The position of said maximum number of memory POSRMAX in the histogram,
-A set of classification limits determined with respect to a histogram added by examining the memory and the discovery of said criteria,
It processes the histogram, generates a set of limit-equivalent, one hand, and is addressable from the position of maximum POSRMAX to generate the first set of limits with respect to the reference Reduce the scan of the address of the memory (100), while scanning the address of the memory (100) addressable to generate a second set of limits with respect to the reference from the position of maximum POSRMAX Characterized in that it includes means for increasing, the limit existence was therefore determined from the position of the maximum POSRMAX in the histogram by scanning both addressable memories (100). Consistent with claim 81, apparatus characterized by processing the histogram to allow the means for generating a set of limits to make a determination according to one of the following aspects:
-Maximum POSRMAX position for each walking by address, manufacturing of counting signal in turn, and limit manufacturing in one scan of reduced walking
-By the manufacture of the limit in two scans where the first first limit cycle of the address is then instantaneously limited, the counting signal presence is subtracted from the maximum number of positions, POSRMAX between the first is then Examine the new counting signal being added to the position of maximum POSRMAX during the second scan,
-By producing limits in two examining cycles of addresses, first of all, then limiting the first limit, counting signal presence increased position of maximum POSRMAX during the first scan, and The new counting signal presence reduced the position of maximum POSRMAX during the second scan. 84. Apparatus according to claim 81 or 82, characterized in that it comprises:
A means of selection (1162) to authorize the selection of criteria for at least one of the following values:
-Maximum value RMAX,
-The threshold value supplied to the module by THRESHOLD,
-A lot of histogram points NBPTS generated by processing the histogram and stored in the module registers. 84. Apparatus according to claim 83, characterized in that the criterion is selected among RMAX / 2, THRESHOLD, NBPTS / THRESHOLD. The device at least complies with any of the claims that 81-84 described its features in it:
-One adder / direction checking signal DIRECTION coincides with the binary value of the signal COUNTER generating the addressing signal COUNTER is offset to the value POSRMAX where the offset value is the maximum number of positions in the histogram A decrement (108) that makes it possible to add or subtract the value of the offset signal of the addressable memory (100), while or may be equivalent
-Comparator (1161) and, on the other hand, the logic circuit (1163, 1164, 1165, 1166, 1167, 1168) receiving the mentioned data item in the addressable memory (100) and the other hand On the fiance to generate a limit updating signal to allow verification of one of the limits according to the criteria and examining the direction,
-A selection multiplexer (105) arranged for addressing, an addressable memory (100), said multiplexer having three inputs each receiving an input signal carrying parameters, a counting signal COUNTER and The adder / decrement output (108) was input. 85. An apparatus consistent with any of claims 81-85, characterized in that it comprises means for prediction having a calculation of an average position POSMOY that is related to the parameters and calculation of the differential. Before the execution of the operational subunits (101, 101 '), which are subject to generating an output signal (101s) for feedback on said set of parameters and for the set of limits The parameter due to the difference between two successive means determined to be subtracted from the differential (145) (146). According to claim 86, the parameter is the average position POSMOY, and when the second binary state in the opposite case matches the output centroid for feedback with the first binary state Device characterized by including a centroidal device further meant to generate the signal. Claims 81 to 85 (characterized by the fact that the parameters analyzed by the STN module are complex and that each itself is obtained by a combination of at least two basic parameter input signals DATA (binary numbers of A1A2A3 Any device according to any of the following ... Ap) of complex parameters (A1A2A3 ... Ap) containing P fields each corresponding to a basic parameter A1, A2, A3, ... Support Ap, and in that it includes the means of the module to generate and store registers, P from the position of the maximum POSRMAX in the histogram of complex parameters to the basic parameters, each by both scans of memory Pairs of supported limits. 90. Apparatus according to claim 88, characterized in that it includes:
Offsetting device (130) Possible to insert into counting the signal COUNTER offsetting what the device digital allows to address in memory (data corresponding to specific fields of complex parameters) To. An apparatus consistent with claim 88 or 89, characterized in that it comprises means for prediction having a calculation of the average complex position POSMOYA1A2 ... Related to the calculation of a set and differential (A1, (A2, ...
Ap of each p-basic parameter of the complex parameter (A1A2A3 ... Ap) is determined by the difference between two consecutive complex averages and subtracted from each of the base parameters matching Sign differential unit (145, 146) prior to execution of motion subunit (101) targeted to generate an output signal (101s) for feedback related to a set of bounds did. Matching claim 90 is the first time when the complex parameter (A1A2A3 ... Ap) matches the average complex position POSMOYA1A2 ... and for the second binary state in the opposite case An apparatus characterized by including a centroidal device further meant to generate an output centroid signal for feedback having a binary state of: 92. Apparatus according to claim 90 or 91, characterized in that it comprises:
Operational subunits (101 ′) AND (131) or classification zones with means enabling the generation of classification room of the output signal by any combination (133, 134), respectively of Zand and each basic parameter, Zor's OR (132). A device according to any of claims 81-92, characterized in that it comprises:
Classification memory (118) with number of words corresponding to histogram size, two-state binary data, reference to non-discovery of criteria on histogram and first state corresponding to discovery of second state, limit Module STN () in the form of a class having an initial state value corresponding to the histogram zone formed between and the output signal (101s) of the classification memory (118) being sent to the feedback bus (111) Stored class addressable memory (100) in memory (118) storing presence. A device according to any of claims 81 to 93, characterized in that it comprises:
The addressable memory (100) of the STN module during a given calculation cycle having an initialization value related to the functional calculation means (120) value is addressable memory (100) at the end of the previous calculation cycle To the initialise stored in). 95. A device consistent with claim 94, wherein the function used to calculate the initialization value that characterizes the generated (Km-1) / Km stored value times the value stored in Km 2m with m greater than or equal to zero, a memory effect that allows a certain m0 to be exceeded and m equal to zero corresponding to the lack of memory effect. 96. A device, consistent with any of claims 81-95, characterized in that it includes means for determining, and a given histogram into which a set of data classifies, the amplitude and position of the vertex And each class corresponding to at least a vertex of the data, one of the classes corresponding to the maximum vertex (RMAX (POSRMAX)) of the histogram. 99. Device histogram maximum according to claim 96, characterized in that means for determining and authorizing a set of data classes are distributed among several STN modules The first module (ST1) that determines and stores the first class corresponding to the vertex,
Then, the second class corresponding to the highest second is determined, the lower vertex of the stored second module (ST2) histogram is determined, and one of the following modules stored is stored. Histogram of modules in the larger vertex class prohibiting tuples and vertices in so-called output (ST3 ...). 99. An apparatus consistent with claim 96 has determined that the means class characterized therein determines and stores a set of data that is distributed among several modules, and the class A set of data corresponding to one of the classes (thus determining the class it has between each module and storing the modules (ST'1, ST The order of histogram vertices towards the module of the set of modules (ST'1, ST'2, ST'3 ...) received by the distributed class ('2, ST'3 ...) The first module (ST'0) that has sent the class unit classification basis earlier in reducing the. 99. An apparatus consistent with claim 96, wherein said means comprising a histogram value memory M0 in which means classes characterized are distributed within an STN module and an application programming interface (API) module ( The memory M2 class (the number of classes and the memory RC of the threshold value of the register RC) that makes it possible to determine and store a determined set of data by the memory M1) of the addresses arranged in the amplitudes of the vertices of the histogram Enable to execute an API that allows you to store a sequence number:
-Initialization cycle resetting memory M0, M1, M2, M3 and registers,
-Calculation cycle for loading on M0, the value of the histogram determined in the module,
-Cycle to update class,
The above updating cycle consists of:
(A) Decreasing the order of the memory M0 and sorting the amplitudes when storing the matches referred to in the memory M1
(B) (Search for classes by :)
Tag the class in memory M2,
And
Storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C) (Verify the class of memory M2 with the address of the considered class of M2 having the threshold of M3 corresponding to the class considered by comparing the values of memory M0). 100. Apparatus according to claim 99 corresponds to the addressable memory of the memory M0 and MSTN modules, and the corresponding data is a single function table of amplitude and position memory couple (600) Describes the features that can be put into (RMAX0 / POSRMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ...) The peak of the histogram in decreasing the order of the amplitudes (table of functions performing class mechanical hardware sorting during the calculation step). 101. Apparatus according to claim 99 or 100 (characterized in that it is assembled in a module STN in a multi-class function apparatus):
-Table of one function of amplitude and position memory couple (600) (RMAX0 / POSRMAX0;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2 ...) Decrease the order of amplitude, addressable memory being replaced with a table of machine hardware sorting classes during the calculation step, replaced with a unique table of said functions The vertex of the histogram
-Memory M2, which allows the order number of classes in the store
-Class threshold memory M3,
-Class number register,
-Application programming interface API,
Said multi-receive receiving at least one simple (DATA (A)) or complex (DATA (A1 ... Ap)) parameter, validation signal (VALIDATION), linear combination of feedback signals and arranging Class function devices signal (INIT, CALCUL, END, CLOCK)-Each multi-class device that is sending back at least 1 outputs each corresponding to a class (Cl1 ... Clk) on the feedback bus The signal got stuck. A device consistent with any of claims 96-101 determines at least one class, with a minimum of two multi-class modules (first module (660) (664) operating in this world TD) Operating in the spatial region SD to be included in the object approval system including replenishing for at least one half module (661, 662, 663) (665, 666, 667) of the class Delineated the characteristics of doing. Matching any of claims 80 to 102 is characterized in that it is included in the object approval system including a set of histogram calculation and processing STN module by sectorizing, zone determining module and centroid s And a device that makes it possible to divide a determined zone into several corner sectors, which corresponds to the center of gravity of the zone, and which sector of the sector focused on the search, Tweak the sectorization that divides the sector into several sub-sectors for new and visible and upstairs. An apparatus consistent with claim 103 characterized in that it comprises at least one STN module with at least two orientation devices p (p (on the shaft (150, 151) at the input for reference The first space Zi of at least two Cartesian coordinates of the input parameters (and the module determining the centroid for the input parameters), and the system 1 axis rotation includes the centroid BarZi by the association of the univariate linear module (296) processing the first parameter, and then determining the second association (298) (299) of the second two line module (297), processing the coordinates Inside said first space Zi second centroid BarZi + 1, said first space, Zi (second centroid) divided into distinct corner sectors (Zri0, Zri1, Zri2 ...) distributed uniformly BarZ i + 1 BarZi and even signal) the second centroid of each sector BarZi + 1 being processed by the two-line module of the sector (300, 301 ... 307) receiving the signal) The two-line module of the sector corresponding to the BarZi + sector to the sharp-cornered sector uniformly in relation to the set of two-line module of the sector of the later rank sector Delivered to fine-tune the divide (Zra, Zrb, Zrc ...) that owns the second centroid and also allows to split the sector determining at least one angle (1071) of the reference axis (1071) perpendicular to the direction of the module and the direction of the line substantially merging the centroids BarZi and BarZi + 1 (the distance between both centroids BarZi and BarZi +1 j (1070). 103 or 104 (characterized in that the object can be observed from different distances and the system further comprises at least one size invariant device (450) (451) (410) ) On the one hand, the logarithm (400) of the distance LD between the size-invariant device reference points received at least at the input (400) and on the other hand the value (substantially at least one point of the object and module) The distance between the two centroids s BarZi corresponding to an angle and a constant projection and BarZi + 1 (j of at least one value C (the device determining)) Rotating (and LD. Consistent with claim 105 is characterized in that the size invariance device (410) includes means for controlling the distance allowing free use of an external measuring distance or an internal determination of the distance. apparatus. Device according to any of claims 102-106 (characterized in that the object can be observed according to different angles and the system further comprises at least one rotation correction device (900)) Rotation corrector allows the angle value to be corrected (a reference axis for a set of values ((angle and previously determined modules). 108. An apparatus consistent with any of claims 81-107, wherein the object recognition system including a set of histogram calculation and processing STN modules causes at least one apparatus for deforming the parameter reference system by rotating the angle. 2D reference case for polar coordinate parameters of pixel X (Y) corresponding to the following matrix behavior (represented at least above the two dimensions of the parameter) : A device characterized in itself to be consistent with any of claims 102-108 is, over time, several centroids s of the zone appearing with respect to the first zone and its own first centroid with respect to the label. It is decided to store the barycentric coordinates in the form of approval data. Matching claim 109 includes means for determining whether the observed new object matches or has previously determined approval data for the new object and has previously stored label approval data A device characterized as not corresponding to an object having a label stored by comparison. Claim 109 or claim 110, wherein the device characterized by the label approval data and 110 is further associated with an angle (by means (920) (921) and an average Distance LD ”allows the angle of rotation to be determined (and the average distance LD). Matching any of claims 109-111 is characterized in that the label is analyzed by a histogram calculation and a processing module (670) capable of determining and storing a set of categorization data of the label. The depicted device. By comparison of consecutive sequences between values of this derived signal corresponding to exactly the same location of the result (object hierarchy-organized details) by forming at least two histograms of the extracted digital signal From the derived signal, there is an absolute value representative of the variability of Gaussian differences in both of these sequences where the difference is above a certain threshold and for each same spatial location of the extracted signal above 1. The derived digital signal regarding the digital magnitude of the signal to be pulled out, at least the various locations (which provide information related to the characteristic details of the object) is being done in orderto and what about At least already related to the local identification of the location of the signal, which relates to the location of the details Supply information. A processing multiclass carried in the form of a sequence Aij by means of a digital signal DATA (A) input with a method of operating the histogram calculation and a functioning device for analyzing parameters ... t (A'ij ... t ) A.prime. ij ... t, ... a parameter relating to the confirmation of this space in the binary number, j ... associated with the synchronization signal, that allows the number to define an immediate t of space and position i Is essentially generating a set of histogram feature extraction results that include a couple of the maximum amplitude histogram and the maximum amplitude data of the position corresponding to at least the maximum vertex of the histogram. Some can consist of several vertices and methods that characterize, while the calculation cycle and the signal (VALIDATION) during itself determine during the calculation cycle that the amplitude RMAXi and the position POSRMAXi A couple of couples,
The couple RMAXi and POSRMAXi presences automatically belonged by the command to reduce vertices and saved functional memory with machine hardware sort of multi-class functional device during the calculation cycle. A set of results in the form of k classifications greater than or equal to 1 (each classification signal corresponding to the class associated with the histogram vertices), with a signaling (Cl1 ... Clk) output, corresponds to claim 114 This further occurs in a way that characterizes itself. 116. A method according to claim 114 or 115, characterized in that it implements a device with multi-class functionality, including:
Application programming interface (API), memory M0 storing amplitude RMAXi arranged in vertex order, memory M1, memory M0 of addresses arranged in vertex amplitude POSRMAXi, and amplitude and position memory couple M1 (RMAX0 / POSRMAX0; stored in a single function table;
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2. ..RMAXi / POSRMAXi ...) Including histogram vertex memory M2 in reducing the order of amplitude (table of functions (600) performing class machine hardware sort during the calculation cycle) It is possible to store the order number of the device class of the multi-class function, the memory M3 of the class threshold value, and the register RC of the number of classes. Allows execution by a method in accordance with the claim that claim 116 described the features of an API therein:
-Initialization cycle resetting memory M0, M1, M2, M3 and registers,
-Calculation cycle to determine and automatically classify a couple of values in a table of functions
-Cycle to update class,
The above updating cycle consists of:
(B) (Class search step):
Tag the class in memory M2,
And
Storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C) (validation step of the class of the memory M2 by comparing the value of the memory M0 at the address of the considered class of M2 with the threshold of M3 corresponding to the considered class). In accordance with claim 116 or 117, the method characterized in itself determines and in memory M4 (number of histogram points NBPTSi corresponding to said class) one store further belongs for each . Consistent with claim 118 is a method characterized by itself, and in the memory M5 (mean position POMOYi of the parameters of the class) one store further belongs for each. The method characterized by itself is consistent with claim 119 when the parameter matches the average position POSMOYi and for the second binary state in the opposite case for the class considered. Furthermore, a centroidal device is implemented whose object is to generate an output centroid signal having an initial binary state. A method consistent with any of claims 114-120 allows the means of selection (1162) to be performed to choose for a given class for which at least one of the following threshold criteria is evaluated: Delineated the characteristics of:
-The amplitude value of the vertex of the histogram,
-The threshold value THRESHOLD supplied to the device,
-NBPTS, lots of points in the histogram
Each class of classification signal being generated for a set of values of parameters whose histogram amplitude is greater than the threshold criterion. 122. Method according to claim 121, characterized in that it selects a criterion among RMAX / 2, THRESHOLD, NBPTS / THRESHOLD for a given class. 114-122 (parameters analyzed by a device of multi-class function are complex and that each itself obtains the latter by a binary combination of at least two basic parameter input signals DATA (A1A2A3 According to any of the following features ... Ap) Complex parameters (A1A2A3 ... containing P fields, each corresponding to a basic parameter A1, A2, A3, ... Ap) Support Ap. A method characterized in itself to be consistent with any of claims 115 to 123 is based on the feedback sent to the multi-class device of the histogram calculation and processing module, the so-called STN, feedback bus (114). Execute a validation signal (VALIDATION) that includes the forming output signal and feedback. A method consistent with claim 124 comprising a set of histogram calculations, a module, an STN module having at least two tools fitted with a device of multi-class function, the first module of this world region TD (660 ) (664) Processing operating, determining at least one class and replenishing for said class at least one half module (661, 662, 663) (665, 666, 667) In that of the object recognition system, the characteristics of operating in the area SD of the space are depicted. Consistent with claim 125 is a method / T / S characterized in that a module operating in the realm TD receives a speed parameter MVT or a color parameter L. 114-126 (which is performed in an object recognition system including a set of histogram calculations, and a sectorized processing STN module having at least one module determines a device, zone of multi-class function The module and the centroid s are characterized to fit) and the method itself is centered on the corresponding centroid of the zone and its one search Any sector among the sectors that divide the zone into a determined, new and visible, and itself divides the sector into several sub-sectors, but the method is further upgraded Therefore, the fine adjustment of sectorization can be continued. A method consistent with claim 127 characterized in that it is performed in at least one STN module with at least two orientation devices p (p (p (150, 151) for reference The first of the input parameters (plus the module determining the centroid for the input parameters) of the input coordinates (and the axis rotation of the system 1 includes the centroid BarZi by the association of the univariate linear module in the first axis Determining the space Zi (296) processing the first parameter and determining the second association (298) (299) of the second two line module (297), processing the coordinates Inside the first space Zi second center of gravity BarZi + 1, the first space divided into distinct corner sectors (Zri0, Zri1, Zri2 ...), Zi ( 2nd center of gravity BarZi of BarZi + 1, and the second centroid of events in each sector being processed by the two-line module of the sector (300, 301 ... 307) receiving the signal) BarZi + 1 to a straight corner sector with a set of two-line modules in a rank sector after a two-line module in the sector corresponding to the presence of BarZi + 1 for the BarZi + uniformly upstairs Sectorization and further fine-tuning to determine at least one angle delivered (Zra, Zrb, Zrc ...) allowing to split the sector that owns the second centroid (substantially (1071) of the reference axis perpendicular to the direction of the straight line and the module that centroid s BarZi and BarZi + 1 are united (the distance j between the centroid s BarZi and BarZi + 1 ( 1070), reference according to the angle (to get translation invariance To grant permissions to. 128. Described in claim 128 (characterized in that the object can be observed from different distances of the system and at least one size invariant device (450) (451) (410) and its one tool) Method at least with input, one hand, the value of the logarithm (400) of the distance LD between the reference points and the size invariant device received at least one point of the object, and the module (two centroids BarZi The distance between and the BarZi + 1 (j of at least one value C (the device determining) j) (1072) of a substantially constant projection and corresponding to an angle (rotation with respect to (1071) (and LD. Consistent with claim 129, in that one tool of the size invariant device (410) means for controlling the distance freely allowing the use of an external measuring distance or the internal determination of the distance. The method depicted. Method according to any of claims 128 to 130 (characterized in that the object can be observed according to different angles of the system and at least one rotation correction device (900) and its one tool) Allows correction of the device angle value (of the reference axis ((angle and module previously determined for a set of values). A method consistent with any of claims 127 to 131 is characterized in that it is carried out for at least one device for deforming a parameter reference system by rotation of an angle ((parameter A reference that is at least above the two dimensions of the rotation) in the case of a two-dimensional reference for the polar coordinate parameters of pixel X (Y) corresponding to the following matrix behavior: A method characterized in itself to be consistent with any of claims 127 to 132 is that, over time, several centroids s of the zone appearing with respect to the first zone and its first centroid It is decided to store the barycentric coordinates in the form of approval data. Matching claim 133 determines the new object and one approval data to determine if the observed new object matches or does not correspond to an object with a previously stored label A method characterized to do that compares it with previously stored label approval data. 135. The method according to claim 133 or 134 characterized rotation angle features that it further relates to label approval data (by means (920) (921) and average distance LD 'said angle of rotation (And the average distance LD). A method characterized by itself in conformity with any of claims 133-135 is a process capable of determining and storing label approval data and a set of categorization data of said label by histogram calculation Analyze module (670). A processing multiclass carried in the form of a sequence Aij by means of a digital signal DATA (A) input with a device for the function of histogram calculation and parameter analysis ... t (A'ij ... t) A.prime . ij ... t, ... the binary number associated with the synchronization signal that allows the number to define an immediate t of the space and position i, the histogram of this space of j ... at least It is essential to generate a set of histogram feature extraction results that include a couple of the maximum amplitude histogram and position maximum amplitude data corresponding to the maximum vertex of the histogram, allowing it to include means. Calculating a parameter histogram on the VALIDATION during the calculation cycle, and determining during the calculation, whereas it can consist of several vertices and methods depicting the characteristics of Circulates a couple of couples, with amplitude RMAXi and position POSRMAXi
The couple RMAXi and POSRMAXi presences automatically belonged by the command to reduce vertices and saved functional memory with machine hardware sort of multi-class functional device during the calculation cycle. 137. An apparatus consistent with claim 137, wherein the means characterized by further enables a set of results in the form of k classifications to be one or more k (each corresponding to a class associated with a vertex of the histogram. The (classification signal) is generated at the output of the signal (Cl1 ... Clk). 138. Apparatus according to claim 137 or 138, characterized in that it includes:
Application programming interface (API), memory M0 storing amplitude RMAXi arranged in vertex order, memory M1, memory M0 of addresses arranged in vertex amplitude POSRMAXi, and amplitude and position memory couple M1 (RMAX0 / POSRMAX0; stored in single function table (600);
RMAX1 / POSRMAX1;
RMAX2 / POSRMAX2. ..RMAXi / POSRMAXi ...) Including histogram vertex memory M2 in reducing the order of amplitude (table of functions (600) performing class machine hardware sort during the calculation cycle) It is possible to store the order number of the device class of the multi-class function, the memory M3 of the class threshold, and the register RC of the number of classes. Allows execution by a device that conforms to the claim that claim 139 described the features of the API therein:
-Initialization cycle resetting memory M0, M1, M2, M3 and registers,
-Calculation cycle to determine and automatically classify a couple of values in a table of functions
-Cycles for updating classes, updating cycles include:
(B) (Class search step):
Tag the class in memory M2,
And
Storing the corresponding threshold of memory M3 (the number of classes stored in register RC and thus determined)
(C) (validation step of the class of the memory M2 by comparing the value of the memory M0 at the address of the considered class of M2 with the threshold of M3 corresponding to the considered class). An apparatus consistent with claim 139 or 140 allows a determination to be made thereby by characterizing it, and further of memory M4 (number of histogram points NBPTSi corresponding to said class) Remember, for each class. 144. Apparatus in accordance with claim 141 allows to determine thereby delineating features therein, and further each class of memory M5 (mean position POMOYi of said parameter of said class) For, remember. Consistent with claim 142 is that the parameter has an average position POSMOYi, and when the second binary state in the opposite case for the considered class matches the output with the first binary state Apparatus characterized in that it includes a centroid-like device that is further meant to generate a centered centroid signal. A device according to any of claims 137-143, characterized in that it includes:
A means of selection (1162) to authorize to select threshold criteria for at least one of the following values for a given class:
-The amplitude value of the vertex of the histogram,
-The threshold value THRESHOLD supplied to the device,
-NBPTS, lots of points in the histogram
Each class of classification signal being generated for a set of values of parameters whose histogram amplitude is greater than the threshold criterion. 145. Apparatus according to claim 144, characterized in that, for a given class, the criterion is chosen among RMAX / 2, THRESHOLD, NBPTS / THRESHOLD. 137-145 (characterized in that the parameters to be analyzed are complex and that they are each obtained by a combination of at least two basic parameter input signals DATA (A1A2A3 binary numbers each Ap) Support Ap for complex parameters (A1A2A3 ... Ap) containing P fields, each corresponding to a basic parameter A1, A2, A3, .... Matching any of claims 138 to 146 includes an output signal and feedback that are integrated in a histogram calculation to form a feedback, being sent to a module, a so-called STN, feedback bus (114) A device that characterizes the fact that it is processing a VALIDATION signal. The device in accordance with claim 147 is a minimum of two STNs with a set of histogram calculation and multiclass function devices (first module (660) (664) operating in this world domain TD) Object comprising a processing module comprising determining at least one class in the module and replenishing for said class at least one half module (661, 662, 663) (665, 666, 667) Depicted the feature of operating in the spatial domain SD to be integrated in the approval system. Consistent with claim 148 is a device / T / S characterized in that a module operating in the realm TD receives a speed parameter MVT or a color parameter L. An apparatus consistent with any of claims 137-149, characterized in that it is integrated in an object approval system that includes a set of histogram calculations, and a sectorization having at least one module The STN module according to the process of installing the device of the multi-class function and determining the zone and the center of gravity s by the module and allowing the determined zone to be divided into several corner sectors, To divide the sector into several sub-sectors to the corresponding center of gravity and which sector among the sectors focused on the search looks new and the method can Continue to tweak the sectorization for further ascension. A device consistent with claim 150 characterized in that it is included in a system comprising at least one STN module with at least two orientation devices p (p (of the shaft (150, 151) of the reference At least two Cartesian coordinates of the input parameters (and the module determining the centroid for the input parameters) at the input for, and the axis rotation of the system 1 includes the centroid BarZi by the association of the univariate linear module Determining the first space Zi (296) processing the first parameter and processing the coordinates of the second two-line module (297), the second association (298) (299) Determining the interior of the first space Zi second center of gravity BarZi + 1, the first space being evenly distributed (Zri0, Zri1, Zri2 ...) divided into distinct corner sectors , Zi (second centroid BarZi + 1 of BarZi, even signals) two lines events each sector being processed by a module of sectors being received (300, 301 ... 307) the
BarZi + 1 presence allowed to divide a sector that owns the second center of gravity BarZi + 1 into a sharp corner sector in relation to a set of two-line modules in a later rank sector The two-line module of the sector corresponding to the second centroid placed to be fine-tuned to uniformly determine the sectorization for ascending and at least one more angle (Zra, Zrb, Zrc ...) (substantially the center of gravity s BarZi and BarZi + 1 combined with the reference axis (1071) perpendicular to the direction of the module Give the center of gravity s the distance between BarZi and BarZi + 1 j (1070), the reference according to the angle (to give translation invariance. Claim 151 (depicted that the object can be observed from different distances and that the system further comprises at least one size invariant device (450) (451) (410)) At least input of the described device, one hand, the value of the logarithm (400) of the distance LD between the reference points and the size invariant device received at least one point of the object, and the module (two centroids) s The distance between BarZi and BarZi + 1 (j of the device determining at least one value C) (1072) of a substantially constant projection and corresponding to an angle (relevant Rotation on (1071) (and LD. Consistent with claim 152 characterized in that the size invariance device (410) includes means for controlling the distance allowing the use of an external measuring distance or the internal determination of the distance freely. apparatus. 151. Apparatus according to any of claims 151 to 153 (characterized in that the object can be observed according to different angles and the system further comprises at least one rotation correction device (900)) Rotation corrector allows the angle value to be corrected (a reference axis for a set of values ((angle and previously determined modules). Device characterized in accordance with any of claims 151 to 154 characterized in that it is associated with at least one device for deforming the parameter reference system by rotation of the angle ((parameter 2 References that are at least one dimension above) Rotate in the case of a two-dimensional reference for the polar coordinate parameters of pixel X (Y) corresponding to the following matrix behavior: Consistent with any of claims 151 to 155, there is determined several centroids s of the zones that are characterized with respect to the initial centroids of the first zone followed by time, characterized by the system. A device that makes it possible to store barycentric coordinates in the form of approval data for labels. A new object by the system characterized in that conforming to claim 156 does not correspond to an object having a label that is consistent or previously stored to determine whether a new object has been observed A device that makes it possible to determine product approval data and compare it to previously stored label approval data. 156. The device according to claim 156 or 157, wherein the device is characterized in addition to the label approval data, the rotation associated with the angle (by means (920) (921) and the average distance LD It allows to determine the angle of rotation (and the average distance LD). 157. A processing module, wherein an apparatus consistent with any of claims 150-158 can determine and store a histogram calculation and a set of categorization data of the label by a system characterized therein. 670) makes it possible to analyze the label approval data. The substantially Gaussian having a smoothing effect between two successive sequences results in forming at least two histograms of the extracted digital signal in resolution (object hierarchy-organization For each same spatial location of the extracted signal in absolute value from the extracted signal by comparison of successive sequences between values of this extracted signal corresponding to the exact same location of the extracted signal) In order to obtain at least the derived digital signal related to the variability of the Gaussian difference variability of both of these sequences, where the difference is above a certain threshold and above one Increase the digital magnitude of the signal (and provide information related to details) and what And at least already related to the local location of the signal provides information related to the location of the details. Space--represented by a parameterized method carried by the histogram calculation and processing module's supplementing power for recognition and approval of objects in this world, and the resulting data evolving over time and sequence The
The module is related to at least one parameter and according to the application programming interface (API) zone and the relevant representation of the center of gravity s and one division of the zone and therefore the zone and its one search To several corner sectors that are centered on the corresponding center of gravity that the new center of gravity has determined about which sector within the sector, and the method can continue for further ascent On the other hand, it is characterized by a fine-tuning of sectorization that itself determines under control to divide the sector into several sub-sectors. The method characterized in itself to be consistent with claim 161 is one of two sectorization levels, a first level dividing the first zone into sectors, and a sector with a new centroid in a sub-sector. Perform the second level that separates. A method consistent with claim 161 wherein the first zone to a fine-tuning sector characterized by its one limit (a sector containing a second centroid completed without a new sectorization) is Divided into one level of sectorization and itself. 161. A method according to claim 161 (162 or 163), characterized in that the module characterized at the end of sectorization, the angle given with respect to the straight line joining both centroids s and the straight sideways Both of the above, the center of gravity s determine at least one angle and one of the modules corresponding to the distance drawing the line. 164. The method of claim 164, characterized in that:
-In the first sequence determined by at least one first module for at least one predetermined parameter, the first zone and centroid cooperated related expressions,
-In the second sequence one colleague with the first module, on the other hand, at least a half module was intended to determine at least a half centroid of the first zone, and on the other hand The set n of the sectorization module was going to divide the first zone into n corner sectors (the sectorization module receiving the first zone and its centroid)
-In the third sequence sent to a set of sectorized modules (the second module whose object is to determine the centroid of at least one half of which the second centroid has been determined); And 1 determine the confirmed sectorization module, where the sector consists of the second centroid,
-In a new set of fourth sequence, sectorized modules other than those freed and confirmed and sub-sectorized modules associated with the confirmed sectored modules, 1 is the second centroid, angle And a confirmed lower sectorized module including a module determined for the identified lower sectorized module. 165. A method consistent with claim 165 comprising determining a zone and replenishment operating in an area SD of space, wherein the center of gravity is at least two modules, the first module operating in this world region TD Delineated features related to at least one-half module. 170. A method according to any of claims 161 to 166, characterized in that the relevant expression is Cartesian. A method of matching the centroid corresponding to the couple (each of the modules of the distance between the centroids s, respectively) and any of claims 161-167 (characterized in that the relevant expression is polar) Logarithm (LD) of the distance between the object and the sensor. A method consistent with any of claims 161 to 168, characterized in that it is performed for at least one STN module with at least two orientation devices p (p (of the axis (150, 151) At least two Cartesian coordinates of the input parameter (and the module determining the centroid for the input parameter) at the input for reference, and its own axis rotation is the centroid by the association of the univariate linear module Determining the first space Zi containing BarZi (296) Processing the first parameter and then processing the coordinates of the second two-line module (297), the second association (298) (299) Inside said first space Zi second center of gravity BarZi + 1 determined, said first divided into distinct corner sectors (Zri0, Zri1, Zri2 ...) distributed uniformly Of each sector being processed by a two-line module of sectors (300, 301 ... 307) receiving the space, Zi (BarZi of the second center of gravity BarZi + 1, and even signal) Event
BarZi + 1 presence allowed to divide a sector that owns the second center of gravity BarZi + 1 into a sharp corner sector in relation to a set of two-line modules in a later rank sector The two-line module of the sector corresponding to the second centroid placed to be fine-tuned to uniformly determine the sectorization for ascending and at least one more angle (Zra, Zrb, Zrc ...) (substantially the center of gravity s BarZi and BarZi + 1 merged with the reference axis (1071) perpendicular to the direction of the straight line and the module (both centroids s The distance between BarZi and BarZi + 1 j (1070) beside the straight line. A method consistent with claim 169 characterized by an angle at it ((1071)) and a module (j (1070), determining the value C ((1072) centroid on the reference axis according to the angle s A straight line projection that combines BarZi and BarZi + 1 ( 170. (characterized in that the object can be observed from different distances, and which itself further performs at least one size invariant device (450) (451) (410)) The method described in at least one input, the value of the logarithm (400) of the distance LD between the reference points and the size invariant device received at least one point of the object, and the module (two The distance between the centers of gravity s BarZi and BarZi + 1 (j of at least one value C (the device determining)) (1070) corresponds to a substantially constant projection and angle (related (1071) of rotation on things (and LD. Consistent with claim 171, in that one tool of the size invariant device (410) means for controlling the distance freely allowing the use of an external measuring distance or the internal determination of the distance, The method depicted. 169. The method according to any of claims 169-172, characterized in that the object can be observed according to different angles and which is further characterized in that it implements at least one rotation correction device (900) The rotation corrector allows the angle value to be corrected (of the reference axis for a set of values ((angle and previously determined modules). A method consistent with any of claims 161 to 173 characterized in at least one device thereof (at least one of two parameters) for its one tool for deforming a parameter reference system by rotation of an angle. References that are at least above the dimensions) Rotate in the case of a two-dimensional reference for the polar coordinate parameters of pixel X (Y) corresponding to the following matrix behavior: A method characterized in itself to be consistent with any of claims 161 to 174 is that, over time, several centroids s of the zone appearing with respect to the first zone and its first centroid It is decided to store the barycentric coordinates in the form of approval data. Consistent with claim 175 is that the observed new object matches or is memorized to determine whether it does not correspond to an object previously classified as 1 new object and 1 approval A method characterized to determine data compares it with previously stored label approval data. 175 or 176, wherein the method according to claim 175 or 176 characterized the feature of the angle of rotation that it is associated with the approval data of the label (by means (920) (921) and average distance LD 'said angle of rotation (And the average distance LD). 175. Approval of a processing module (670) capable of determining and storing a label and a set of data classes by histogram calculation, characterized by itself in conformity with any of claims 175-177 Analyze the data. Space-Represented by the parameters carried by the replenishing power device of the histogram calculation and processing module for recognition and recognition of objects in this world region, and the resulting data developing over time and sequence The
The module is therefore divided into several angular sectors where at least one of the determined zones is concentrated at the corresponding centroid of the zone and for at least one parameter and application programming interface ( API) to determine under the control part of the center of gravity s and the relevant center according to the zone and the relevant expression so that it includes means, and in which sector among the sectors the new center of gravity is found Is characterized by fine-tuning the sectorization to be sought and that the sector can be divided into several sub-sectors, whereas the method can continue for further upgrades Depicted. Consistent with claim 179 is that there are two sectorization levels, a first level dividing the first zone into sectors, and a second level dividing one of the sectors with a new center of gravity in the sub-sector A device that characterizes the existence. 179. An apparatus consistent with claim 179, wherein only the first zone to a fine-tuning sector (a sector containing a second centroid completed without a new sectorization) characterized therein is sectorized. Limited to one level and one division. 179. An apparatus according to claim 179 (180 or 181), in which it is possible to determine at least one angle by means characterized at the end of sectorization, and one module, both Both the above-mentioned center of gravity s, while the module corresponding to the angle given with respect to the straight line participating in the center of gravity s and the distance along the straight side draws a line. An apparatus according to the claim that claim 182 described the features of the means in it enables:
-In the first sequence determined by at least one first module for at least one predetermined parameter, the first zone and centroid cooperated related expressions,
-On the other hand, in order to associate with the first module at least one second module, which is aimed at determining the center of gravity of at least one half of the first zone, in the second sequence, and the other , Set n of the sectorization module was going to divide the first zone into n corner sectors (the sectorization module receiving the first zone and its centroid)
-The centroid of said moment, the third sequence to send to the set of sectorized modules (the second module targeted to determine the centroid of at least one half that had determined the second centroid), and Determine the confirmed sectorization module, where the sector consists of the second centroid,
-Determined with respect to the fourth sequence, a sectored module other than the one that has been freed and confirmed, and the confirmed lower sectorized module including the second centroid, the angle and the confirmed lower sectorized module A new set of sub-sectorization modules associated with the identified sectorization module to determine which module is. 183. An apparatus consistent with claim 183, by means characterized therein, in a zone and at least two modules, the first module operating in the world region TD, the replenishment operating in the space region SD Allows you to determine the centroid associated with at least one-half module. 184. Apparatus according to any of claims 179-184 (characterized in that the relevant expression is Cartesian). 179-185 (characterized in that the relevant expression is polar), a device for matching the centroid corresponding to the couple (each of the modules of the distance between centroids s, and Logarithm (LD) of the distance between the object and the sensor. Device according to any of claims 179 to 186, characterized in that the means comprises at least one STN module with at least two orientation devices p (p (of the shaft (150, 151) input The centroid for reference axis rotation of at least two Cartesian coordinates of the parameter (and the module determining the centroid for the input parameter), and by means of its input by means of a univariate linear module Allows the determination of the first space Zi containing BarZi (296), processes the first parameter, and processes the coordinates of the second two-line module (297), the second Relevant (298) (299) determined inside said first space Zi second center of gravity BarZi + 1, evenly distributed (Zri0, Zri1, Zri2 ...) divided into distinct corner sectors Each of the first spaces, Zi (second barycenter BarZi + 1 BarZi and even signal) being processed by a two-line module of the sector (300, 301 ... 307) receiving each Sector events
BarZi + 1 presence allowed to divide a sector that owns the second center of gravity BarZi + 1 into a sharp corner sector in relation to a set of two-line modules in a later rank sector The two-line module of the sector corresponding to the second centroid placed to be fine-tuned to uniformly determine the sectorization for ascending and at least one more angle (Zra, Zrb, Zrc ...) (substantially the center of gravity s BarZi and BarZi + 1 merged with the reference axis (1071) perpendicular to the direction of the straight line and the module (both centroids s The distance between BarZi and BarZi + 1 j (1070) beside the straight line. 187. An apparatus consistent with claim 187 characterized in angle angle ((1071) and allows the module to determine the value C (by means of (j (1070) (1072)) Centroid s on the reference axis according to the projection of the straight line merging BarZi and BarZi + 1 ( 187 or 188 (characterized in that the object can be observed from different distances and that it further comprises at least one size invariant device (450) (451) (410) ) With at least the input of the device described in 1), with one hand, the value of the logarithm (400) of the distance LD between the reference points and the size invariant device received at least at one point of the object, and the module (2 The distance between two centroids s BarZi and BarZi + 1 (j of the device determining at least one value C) (1072) corresponds to a substantially constant projection and an angle (relationship (1071) of rotation on something (and LD. Consistent with claim 189 is characterized in that the size invariance device (410) includes means for controlling the distance allowing the use of an external measuring distance or an internal determination of the distance. apparatus. 190. The apparatus according to any of claims 187-190, characterized in that the object can be observed according to different angles and that it further comprises at least one rotation correction device (900) Rotation corrector allows the angle value to be corrected (a reference axis for a set of values ((angle and previously determined modules). Device characterized in accordance with any of claims 179-191, characterized in that it comprises at least one device for deforming the parameter reference system by rotation of the angle ((parameter two References that are at least above the dimensions) Rotate in the case of a two-dimensional reference for the polar coordinate parameters of pixel X (Y) corresponding to the following matrix behavior: Matching any of claims 179 to 192 is possible to determine several centroids s of the zone that appear with respect to the first zone and its own first centroid, followed by time that it includes means The device that describes the feature of storing the barycentric coordinates in the form of approval data for the label. 193. Consistent with claim 193 is that the observed new object matches or is memorized, and determining whether it does not correspond to an object previously classified as such includes new means A device that depicts the feature of determining the approval data of and enabling it to be compared with previously stored label approval data. Rotation angle (means (920) (921) depicting features of an apparatus consistent with claim 193 or 194 allowing it to further include that means to be associated with label approval data ) And then the average distance LD ”makes it possible to determine the angle of rotation (and the average distance LD). Claims 193-195 coincide with any of the processing modules (670) that can determine and store a label and a set of data by means of histogram calculation that the means includes means to classify A device that describes the features of enabling analysis. By comparison of consecutive sequences between values of this derived signal corresponding to exactly the same location of the result (object hierarchy-organized details) by forming at least two histograms of the extracted digital signal From the derived signal, there is an absolute value representative of the variability of Gaussian differences in both of these sequences where the difference is above a certain threshold and for each same spatial location of the extracted signal above 1. Obtain an extracted digital signal relating to the digital magnitude of the signal at least from various locations (which provide information related to characteristic details of the object), and at least about what else of the signal For local location identification, it provides information related to the location of the details That.

Images