DE3234608A1 - Method and circuit arrangement for generating a position-independent object signature - Google Patents

Abstract

The method of functioning and its circuit implementation enables a position-independent object signature to be formed. The formation of such a signature significantly simplifies and accelerates the position-invariant classification of two-dimensional objects (industrial pattern recognition). The invention is based on the formation of a structure-specific data string (object signature) which is independent of position and orientation. The hardware implementation is effected with the aid of an image processing processor which contains the logic operations (dyadic operators) suitable for the digital computation. The separated object is split into an object-characteristic data string (object signature) independently of its instantaneous position via dyadic operations by a set of translation- and rotation-invariant decomposition functions. Part 2 of the system, the segment area correlator, is a specific correlation circuit which correlates the generated object signature by hardware means, that is to say at high speed by direct comparison of the "unknown" pattern in all possible positions with each reference element. The advantage of the structure according to the invention lies in the simplicity of the methods used, that is to say in simple operations suitable for digital computation.

Description

Description:

The invention relates to a method and a circuit arrangement for the identification and determination of the rotational position of a data-separated object. For a fast digital image pattern recognition (work pieces) the recognition problem including the determination of the position with fast algorithms carried out in special computers will. The determination of the rotational position of an object turns out to be in contrast to Detection of a translational deviation is very difficult from a technical point of view problem to be solved. To generate structure-specific features independent of The following methods are known to the position and position or orientation: - Calculation of the area moments of inertia, - correlative methods and - polar coding.

A known method for determining the rotational position is based on the determination of circle-workpiece contour intersections (German patent application P 2544 544.7). Another method determines the angle of rotation by measuring the area F and the main moments of inertia / 1 /. The use of moments of inertia results in symmetrical Objects cause difficulties because their principal moments of inertia become identical. A publication / 1 / describes a polar coding by means of circular scanning, which is much less sensitive to interference. Despite the hardware implementation of the process the time for determining the rotational position of the object is 100 ms and that of the learning phase should be set at 25 seconds. A patent application with the number P 25 42 904.3 shows an approach to centering a singular.

Sample and subsequent quick comparison in the rotating register in the case of polar coordinate deviation.

By hardware development work on an image processor there was the following possibility of position-invariant transformations with simple dyadic operations to be carried out. They are the main characteristic of ours Invention.

For the pattern classification, an unknown pattern, which is within of the image field can be shifted and rotated into a position-invariant object-characteristic Structure to be reshaped. Any object movement can be used for a two-dimensional pattern can be described via a point that is fixed to the body. This fixed point is in the In the present case, the object's center of gravity (xs, ys). By shifting the determined Object focus in the pattern focus of the data network r = O and through the application zonal and sectoral dyadic operators as linking elements result in a Sequence of individual elements (Fig. La + b). The area value of an individual element results in one.

Parameter Fzi + Fz2 + Fsl + Fs2 The sum of the parameters leads to a object-characteristic data string (object signature).

The object of the circuit arrangement according to the invention is the object signature to generate and to specify the object and its Specify the rotational position in relation to a reference.

Literature: 1. Martini, G .; Nehr, G .: Recognition of angular orientation of objects with the help of optical sensors.

THE INDUSTRIAL ROBOT, Volume 6, No.2, June 1979. UK In accordance with the invention this method is implemented with the following circuit arrangement. - Of the The image processing processor consists of the following assemblies: 1. 2. 3 or more binary image memories, 3. Mux / Demux unit, 4. Internal address counter, 5. x, y image area control (window VGL), 6. central control unit with ALU, 6 bit VGL and flags, 7th control word register, 8th area counter and 9th sector memory (read-only memory in the memory area of the administration computer or additional memory in the image processor) These assemblies provide the segmental area correlator with the object signature a data string. The data string is made up of the zonal and sectoral data together.

The assemblies of the segment area correlator are: 1. 2. Reference memory, 3. sector counter, 4. angle increment shifter, 5. subtracter (ALU), 6. Amount generator (ALU), 7. Amount totalizer (correlation counter), 8. Reference selection, 9th control unit and 10th bus coupler.

They form an absolute difference function. Is what you are looking for If the object is in the correct rotational position, the function achieves an absolute minimum.

The invention is described below using an exemplary embodiment Described in more detail with reference to the block diagram (Figure 2).

The block diagram (Figure 2) shows the structure of the device according to the invention again.

The method according to the invention for forming the object signature was implemented in the image processor. To explain the function, assumed a separated binary object. This object can be in one of the binary Function memory are provided for processing.

By calculating the center of gravity coordinates during the separation process this data is available. The read-only memories are converted one after the other zonal and sectoral area functions. During this operation you can both the position of the center of gravity of the object are corrected (central position of the image) as well as the area function through translational displacement into the object position to be brought. The central control unit logically summarizes the two binary images together (AND function) and form an area value from the result. The size the area value is e.g. stored in a byte. After executing the zonal Operation, the object is initially encoded in n area values Fz. The function Fzi..n can become ambiguous with some objects (e.g. concave contours). In a second Sectoral operation, sectoral data are generated, with the object with the segments of a circle of a 2 n-gon is coded one after the other. The consequence of the sectoral area values Fsi..n results in the comparison data necessary for a sharp correlation and calculation of the rotational position required are. In the system developed there were 16 zonal circular ring pieces and 128 sectoral circular segments implemented. An object data string is thus set in this system made up of 144 bytes. An angular resolution of 2:80 is used achieved.

An increase in the number of sectoral sectors improves the angular resolution up to (Template Matching).

The generated data string is fed to the segment area correlator.

In order to be able to carry out an object identification, you must first the reference data are generated. For this purpose, the object is in the system's reference position offered. The processing unit I generates the object-characteristic from this Data string that is stored in the reference memory of unit II. The learning phase runs fully automatically without any interactive user intervention.

In the work phase, the data string to be tested is stored in the device under test memory enrolled. The circuit arrangement is formed from the test and reference data string the absolute value of the difference between the individual function values. The difference formation takes place by addressing the storage space of the individual area values in the test object and reference memory via the sector counter, with the individual References are addressable. The angular incremental shifter takes on the task of based on an initial value, change the reference data string after each run move a position. This results in a comparison data set for the implemented circuit of 18432 bytes. With the system clock of 3 MHz used, it is a single operation carried out in 333 ns, i.e. an object difference function is formed in 5.4 ms.

A shortening of this time is by using faster logic (FAST-TTL) possible by a factor of 10. By operating several units in parallel the segment area correlator can have several object difference functions in this Time to be generated. The method according to the invention is therefore not only in the individual correlation of objects much faster than known methods (factor 200) i but By using the basic unit multiple times, you can also choose from a large number of Objeklent individual objects can be correlated out at high speed.

The inventive method thus forms the basis for a information-oriented image pattern recognition. The patent application -kraft from 10.9.1982- specified method for capturing a selected image information area receives after a quick localization (pre-classification) the contained in the image field Patterns from this system the limit data (e.g. contour coordinates) to then possibly edit again with a higher resolution.

Claims (2)

Method and circuit arrangement for generating a position-independent Object signature PATENT CLAIMS 0- method and circuit arrangement for formation a position-independent object signature, consisting of an image processor with several binary image memories. The binary image memories can also have image planes of the Be gray value memory that the separated image objects or artificially generated Include link objects in binary surface or edge representation. By the application of dyadic operations on two image levels (arithmetic or logical Combination of two units) becomes a position-independent object signature in this way generates that the separated image object with certain decomposition functions (artificial Objects) is decomposed and the decomposition product is brought into a numerical form will. The sequence of these numerical data results in the object signature in the form of a Data strings. This rotatable data string is then combined with a reference data string correlated in a hardware correlator. 2. The method and the circuit arrangement is characterized in that that by a first device (image processing processor) a separated binary Object area image (image object) dyadic with stored binary area functions (artificial objects) is linked. The artificial objects are made of a network formed from data, which the circular rings and sections of a circle (Fig. 1) for decomposition can be removed. The link between the separated object and these functions results in an object-characteristic signature consisting of a sequence of area values the annulus link (zonal data) and the sequence of area values from the link with the section of a circle (sectoral data) By a second circuit arrangement, the so-called segment area correlator (SFK) with several reference memories, one DUT memory and the necessary control and arithmetic logic units, the "unknown" object signature is correlated with the reference signatures. the Correlation takes place in two stages. At the first stage the zonal data correlated (object pre-classification). Because the data set is a much smaller The classification takes place very quickly (in the US area). The pre-classification excludes all reference objects whose correlation values have a certain value to be specified Exceed limit. Mostly remain for the second stage of the sectoral correlation only a reference or two left. Identification takes place in this second stage with a high degree of correlation as well as the determination of the rotational position.