SA517390016B1 - Method for Counting Objects in A Predetermined Spatial Area - Google Patents

Abstract

The invention relates to a method for counting objects in a predetermined spatial area, the method including: - a first unsupervised learning phase, and - a second phase for counting in the area, comprising the following steps: - acquiring an image including the predetermined area via the detector (30), and - estimating the spatial density of objects in the predetermined area by applying the law to the acquired image. Figure 3

Description

Method for Counting Objects in A Predetermined Spatial Area Full Description Background The present invention relates to a method for counting objects in a predetermined spatial 8. The invention [al] relates to a related computer software product and medium information 1010008100. The present invention relates to a detection device capable of performing a counting method. With the increasing use of video surveillance in public places to improve security; Many have been developed

Techniques to analyze the videos provided by the cameras. European Document No. 2,704,060a2 specifically describes a method for estimating the size of the data stack from an image; The method includes the steps of defining a candidate to obtain lel informational data points using an efficient learning algorithm. The method also includes an annotation reception step

From user selects data from most informative data points. The method also includes a step to use the regression function on most of the specified information points along with the relevant numbering to reduce the loss on the footnote data unit. Finally the method includes the step of constructing a regression model using user appended data and inferred appended data. US Document 2000/0031005A1 describes an optical system incorporating a camera

5 and a processor including video analysis software The software is suitable for estimating a background reference frame that is a background; estimation of geometric variables to represent the gradient of change of an object in 1 of the determinant frame; Obtaining a change detection map to distinguish the background from an object in 1 of the specified frame and merging the change detection map with geometric variables to obtain a measure of crowding in the Casall frame

however"; When increasing the spatial densities of objects; The previously mentioned methods provide unreliable results for determining the spatial density of objects in the provided frame. lly there is a function of the method of identifying objects in a predetermined spatial area making it possible to obtain more reliable estimates of the dag of determination when the spatial density of objects is high. General description of the invention at the end; A method for counting objects in a predefined spatial area is proposed. The method involves a first unsupervised learning phase involving the steps of selecting a set of discrete spatial density of objects in a predefined spatial area; Provide variables for 0 to a detector capable of acquiring at least one image » image includes the specified spatial area clue Provided variables include at least one variable related to the focal length of the cass device a variable related to the height of the detector and a variable related to the inclination of the detector For the predetermined spatial area; Generate a database containing many images for each selected spatial density; The generation step uses the variables provided in the custom step; Defining Law 5 facilitates the estimation of spatial density in a predetermined spatial area using the generated education rule. The method also includes a second stage of counting in a pre-determined spatial area that includes steps to obtain an image that includes the pre-determined spatial area through the detection device; Estimating the spatial density of objects in a predetermined spatial area by applying the law to the acquired image. according to specific models; The counting method has one or more of the following features; Alone or in a possible technical combination: - The generation step includes the step of correcting the perspective defects associated with the detector. The method includes the step of slicing an image to identify the pixels that contain an object.

- The selection step is made by learning; Learning includes the extraction of properties; the extracted properties are also used in the estimation step. The selection step is performed by support vector regression. - in the assigned step; The variable associated with the detector is an inherent feature of the detector. - in the allotted step; A position variable for the detector and/or an orientation variable for the detector is provided. A computer software product that includes software instructions is also described; Program instructions that execute a method as previously described; When program instructions are executed by computer 0, an information medium is also proposed on which to store the program product of computer 0 as described above. A monitoring device capable of counting objects in a pre-determined spatial area is also proposed, which includes a detection device capable of obtaining at least one image »The image includes the pre-determined spatial area; detector includes variables»; One of the variables is related to the focal length of the detector; A variable is related to the height of the detector and a variable is related to the inclination of the detector 5 related to the predetermined spatial area. The device includes a processing unit capable of counting objects in a predetermined spatial area. The processing module is able to choose a combination of different spatial densities in a predefined spatial space; receiving variables associated with the detector; Received variables include at least one variable related to the focal length of the pre-selected spatial 0 device; Generate a set of images for each selected spatial AES using received variables; Determining a law that facilitates the estimation of spatial density in a predetermined spatial area using the generated education rule as a learning base and estimating the spatial density of objects in a predetermined area by applying the law to the acquired image.

BRIEF EXPLANATION OF THE DRAWINGS Other features and advantages of the invention will be apparent by viewing the following description of the EY models) as an example only and this is done with reference to the drawings; Which is: - Figure 1; Schematic projection of an example processing unit facilitating a method for counting objects in a predetermined spatial area - Fig. 2; An example image generated to form a learning base»; f - fig. 3; Flowchart of a sample method for counting objects; A dallas processing module 10 and a computer program product 12 are shown in Figure 1. 0 Detailed description: The interaction of the 12computer program product with the processing module 10 facilitates the implementation of a method for counting objects in a specific spatial area, of which the processing module 10 is preferably a computer. 5 in existence ple tqut 10 is an electronic computer capable of processing and/or transforming data that is embodied as electronic and physical quantities in the ple processing unit 0 registers and/or memory in other similar data corresponding to physical data in the BIN registers or gal types of screens; transportation or storage devices. Module 10 includes processor 14 includes dallas data processing unit 20 16 memories 18 and information medium reader 20. Module 10 Load includes keyboard 22 and display unit 24.

The malin product includes a computer 12 and an information medium 20 . Information medium 20 is medium readable Information medium 20 is medium readable by the processing unit 0 1 3 Bale by the data processing unit 14. Information medium 20 is a medium suitable for storing electronic instructions and is capable of Pairing with a computer system bus. (JUS) The information medium readable 20 is a floppy disk optical disc ¢ Compact Disc read—only-memory (CD-ROM) magnetic optical disc magnetic-optical disc read-only-memory (ROM) random access memory (RAM) 10 erasable programmable read-only memory (EPROM) memory Erasable programmable read-only memory A magnetic card or optical card A computer program containing program instructions is stored on an information medium 20 15 The computer program can be loaded on the data processing unit 14 and is suitable for implementing the method for counting objects in a predetermined spatial area when the computer program is executed on the processing unit

The process of the processing module 10 interacting with the computer software product 12 is described with reference to Figures 2 and 3; Which schematically illustrates an example of the method model for counting objects. 0 The term 'object' refers to a countable element. In the context of the invention, the 'object is observable', i.e., detector 30 as illustrated in Figure 3 is capable of observing the body in a predetermined area. Said detector 30, for example, is Camera 30. The aforementioned camera 30 is characterized by P parameters.

Thus the detector has 30 Pintrinsic variables; any; It does not depend on the medium Jie the focal length of the detector 30 or the image size obtained by the detector 30. Intrinsic Pintrinsic variables are camera-specific 30. The detector 30 also has extrinsic variables Pextrinsic . For example, a variable linked to the position of camera 30 on the basis of the medium in question, and it is a Pextrinsic variable. preferred; The Pposition variable of the camera 30 is the height of the camera 30 According to another example, the PPOSItion variable of the camera 30 is the distance of the camera 30 0 with respect to the predefined spatial area. stereotypically The distance of the camera 30 from the predetermined spatial space is defined as the distance between the plane of the detectors of the camera 30 and the point of the predetermined spatial space that is closer to the camera 30. According to another Jil, assuming that the homogeneous orthogonal plane of the reference (YX 2) of the desired space is determined 5 monitor it The other position variable, Pposition of the Camera 30, is the position along the X axis along the axis and the position along the Y axis along the axis. like that»; The variable related to the orientation of the camera 30, which is dependent on the mechanical system holding the camera 30, is a pextrinsic variable. Preferably » The orientation variable 'Porientation of camera 30' is the angle of inclination of the 0 axis of view of camera 30 associated with the horizontal axis. The said angle is generally called the angle of divergence. additionally; Other orientation variables are taken into consideration.

The Orientation af variable of camera 30 for example is the angular orientation of the direction of the light axis of camera 30 with respect to the absolute frame of reference 0 The orientation of the light axis of camera 30 is typically called the orientation angle. The inclination of the camera is also 30 about the axis of view, which is also called the rolling angle; Another example of the orientation variable of Camera 30. It should be noted that the direction of the orientation variables of Camera 30 Load aie can be expressed in the form of the “Ugh” rotated Opler angles; Qualitative wobble and rotation. In general, the pitch angle is the preferred variable. Lad yet in this document; The objects are individuals, and after the detection device 30, it is a camera, and there is a video monitor. in that case; The pre-determined spatial area is the area to be monitored (Aa Tag) for example aT. Objects are objects of biological interest that are observed using the detection device 30 and a lens that performs a microscope function. According to another Jl also; The bodies are carbon nanotubes 5 and in this case; The detection device is an electron microscope. with general sympathy.” This method can potentially be applied to virtually any type of object whose behavior can be induced. dag: general The predefined spatial area ZS depends on the object. As an illustration » when the object in question 1 is an object of biological interest; The predetermined spatial 0 is the Ble ZS for the gia of the sample observed using a microscope.

In the context of the given example; When the body is a person; The specified spatial area

preset 25 ble for part of the median for example; gia from the street. It clearly appears in

Figure 2 image.

in that case; The camera 30 is able to acquire the predetermined spatial space image 25; stereotypically Camera 30 is able to get a picture of the street. Furthermore it;

The enumeration method is a method of counting the stack of data in a medium.

As shown schematically in Figure 3, the method includes two consecutive phases 01 P25

The first phase P1 corresponds to the upper part of Fig. 3 (marked by dotted lines 32);

The P2 second phase corresponds to the lower part of Figure 3. It is the first phase

0 01 is an unsupervised learning phase; While the second phase (P2) is a counting phase in a pre-defined spatial area (ZS), the first unsupervised learning phase (Plage) leads to obtaining a law that facilitates the estimation of the spatial density of people in the pre-specified space. The expression "unsupervised learning" means that the learning data is generated automatically.

5 in this context; The law is a link between the characteristics of an image and the value of the spatial density of people in a predetermined spatial area (ZS). The law facilitates the identification of common features in the increasing spatial density of people.

During learning phase PT the detector variables 30 are used to generate the learning rule All from the learning rule; The regression model is learned by sang modular processing 10.

0 Searches the second phase©2 for counting in a pre-determined spatial space ZS Applying the law to actual cases. 'Actual state' refers to images acquired by camera 30 and not generated LT as in unsupervised learning phase I dlls1© .

— 0 1 — more specifically; The video stream is collected from the camera 30, then the features are extracted from the video stream to obtain the spatial density on the overall image due to the regression model during the implementation of the first learning phase 01. Thus, the implementation of the method facilitates obtaining the number of people in a predetermined spatial area ZS 5

The first unsupervised learning phase includes 01 four steps: selection step 5100; setup step 5102; generation step 5104 and selection step 5106. in the selection step 5100; A group of different spatial densities of people in a predefined spatial space (ZS) is selected

0 According to the model » each spatial density is uniformly distributed between spatial density zero and spatial density high. general dag”; The spatial density is high based on eg lll in particular the viewing conditions. stereotypically The spatial density is greater than or equal to 50 people in the previously defined area, which is a high spatial density.

1 At the end of step f of choosing 1 00 5 a set of spatial densities is obtained. according to the model; The selection step 5100 is performed for several separate regions of the image. in step 5102; Camera-related 30 variants are provided. Available variants include at least one detection-lead focal length-related 30; A variable related to the detector's height 30 A variable related to the detector's inclination 30 in relation to the area

0 predefined spatial 25 and a variable linked to the position of the main point. The variable associated with the position of the main point is for example the position of the optical center relative to the center of the image.

— 1 1 — According to Example » The variable associated with camera 30 is a Pintrinsicdabic variable. For example »the focal length of the camera is provided as 30 or the number of camera pixels as 30. According to another example; The variable associated with camera 30 is only one of the combination of position variable 0116018100 and direction variable 0116018100 of camera 30 with respect to the medium.

at the end of the custom step 5102; The variables associated with the camera 30 are obtained. Among those 1 for variables’ there is at least one variable associated with the focal length detection lead 30; A predefined variable ZS and a variable associated with the position of the key point.

0 in generation step 5104; A learning base is generated. in the end; In the generation step (S104), a set of images and annotation element corresponding to the ground truth are generated. We refer to the expression “ground truth” to use the English expression “ground truth”. A set of images is generated for each spatial density chosen in step f to select 00) 1 5

Generate means that the processing module 10 generates synthetic images using the variable associated with camera 30. Synthetics are not images obtained from camera 30. These images are generated as if they were captured by camera 30; But they are synthetic images. In the “unde” format, pedestrian positions including synthetic images are computed for symmetry of the spatial density field in the 3D medium included in the space comprising the camera's field of view.

0 30 and close proximity to it to avoid edge effects. (JUS) to form a spatial density domain; a continuous spatial density gradient from maximum spatial density to minimum spatial density is found through the image. The gradient is directed along fixed depth lines in

— 2 1 — Fig. 3 where for depth (AAs a) CAS spatial density intervals range from very to very rare. This corresponds to the default for a given depth” and that the appearance of data stacks at constant spatial density is approximately the same as along a constant depth line. As can be used, a combination of randomly placed Gaussian functions in a predetermined spatial space 25 with random standard deviation is used.The said combination facilitates the generation of an image close to One method facilitates ground truth calculation to determine the spatial density of a ground truth function; which is called a function LF function calculates the © function for each pixel of the generated image; the spatial density of the truth function Faia VY is determined by the following formula: N(p; 1, 62. 15,5) 10 > = (1,7)00 6 70 PEP where: GSP - is the list of positions in the image where pedestrians are located; Pp (N - 62) is a 2D Gaussian kernel evaluated at the pixel at the mean position PP and 5 - C about coefficient. The position of the transient in the image is calculated using a variable of the 30 camera by displaying in the image; A point present at the position of a traverse around which the invention rotates and at the corresponding height to describe human height. stereotypically The average human height is 1.70 metres. similar to that; The coefficient is calculated as one-third of the half of the apparent height of the human being at position 080510000 in the image. Next” The value assumed by the o coefficient of the P pedestrian depends on its position in the image. stereotypically The parameter 0m approaches several pixels.

— 1 3 —

Ground truth is calculated automatically and footnote avoided by humans. lL This process is more cdi as the process is performed automatically. at the end of generation step 5104; An lly database is obtained that acts as a learning database. The database includes a set of synthetic images that are related to each image's ground reality.

in selection step 5106; Code is determined. The law facilitates the estimation of spatial density in a predetermined spatial area. in the end; The learning rule generated in the generation step 5104 is used as the learning rule. As Talis explained, it should be noted that the learning rule is generated without manual annotation execution by the operator.

0 According to Jud of Figure 3, the determination step 5106 is performed through four successive steps: sshd 98 for perspective correction » step 51065 for slicing the data stack » step ©5106 for feature extraction and learning step 51060. In the correction step 38m a is taken Due to doses, in picture A, the objects close to the detector 30 appear larger than the distant objects. The influence of said perspective must be taken into account

5 When analyzing the characteristics to be extracted from the image; Where the features extracted from the body of the front panel represent a small part of the body compared to the extracted properties that are extracted away from the image. In the 38m correction step, two successive steps are taken, i.e. a step to measure the characteristics using a perspective measurement map, and a step to divide the image into strips of semi-constant depth.

0. The properties are measured using a perspective scale map where the weight of the pixel is related to the available depth of the object generating the pixel. Pixels are weighed for distant objects. The weight step increases the weight taken into account if the body is further away from the front panel body.

— 4 1 — The perspective metering map is generated using the camera variable 30 more specifically; The correction factor is obtained by the differential straightening. Subsequently; instead of using a two-dimensional variable such as height or width; It would be preferable to explain the three-dimensional nature of human beings. in the end; The change in volume DE shows the dimensions occupied by the passer-by. The (p)W perspective correction factor for pixel m is derived from this. for example; The perspective correction factor //a(0) is obtained using the following formula: ((f 4 + max(4;(p)) lat id + ri l where: The human being is expressed by a cylinder with a circular base cylinder with a circular base « (P)AL is the apparent area spall of the lateral area; and 2) is the apparent area of the upper circular base. Use the obtained map to measure the properties by applying the 5-calculated weights to each of the pixels that are located in a pre-defined spatial space ZS if the properties are surface properties and by applying the square root of the weights to the properties associated with edge effects if the properties are properties. Of the contour type. For cutting strips of fixed depth in the image, the angle of view from which the transient can be viewed is represented by a cylinder having a hog lis) and camera radius of 30" which is positioned at H height 0.

— 5 1 — This angle appears to vary based on the position of the transient in the image. The slice represented in the image is the extreme values which do not diverge from each other. in the end; The maximum angular shift is specified in the slice corresponding to the pixel. as an illustration; Two of the strips 81 and 82 are shown in Figure 2; and that Sharbins 81 and 82 are bound to the predetermined spatial space ZS at the end of correction step 51068; Defects associated with a specific camera shot30 are taken into account when analyzing the synthetic images. In step S106b of data stack segmentation, the pixels that belong to the data stack in the overall image are determined. in the end; in the learning base; Ground truth data is used. 0 in the extraction step ©5106; properties are extracted. Properties are connected to a specific pixel 0; A pixel Ble is about a pixel that was sliced in slicing step 51065. This ensures that the pixel is associated with the existence of the data stack. The properties associated with pixel P are computed on a Wprw sliding window having radius TW centered on pixel m. for example; The contour features are obtained using the Canny edge detector. The contour features are for example the direction of the contour lines; or their lengths. As can step ©5106 to extract the contour features using the Minkowski fractal dimension ; The step is denoted by the abbreviation (MFD) 0 .

— 6 1 — sshall ©5106 can be done to extract textual properties eg by gray based arrays. These layers are typically denoted by the acronym .GLDM The Lad 51060 extraction texture sshd includes a step to represent the extracted properties in a vector de sans dimension by differentiating radius values .r 'w values' more specifically; A reference value of the radius TW corresponding to one-third of the middle of the average height shown in the image of a human as previously described is taken into account for coefficient 008006601 6. The mentioned reference values are used; referred to as 61©; to get 6 different Chal values of diameter 0 aw which are “according to the example of Figure 3:” ref 1/2. rref 1/4. rref .3/2 p rref et 3. rref .2. dam therefore; according to a form by the applicant; per pixel of the image; A vector of the properties is obtained with dimensions 144 * 1. In the learning step (S106) according to the example shown; support vector regression is used. 5 The aforementioned bridging is also referred to in 4a/a5. The regression is facilitated by linking the properties vector to the assumed values by the spatial density in the density function F. The performance depends on The technology depends largely on the appropriate tuning of a variable that is chosen in the kernel and the technology itself. In the example shown, the kernel is a Gaussian kernel; RBF ( Radial Basis Function); the kernel is denoted by the expression kernel kernel O. To distinguish the said kernel, three variables must be distinguished: the kernel variable oy, the penalty variable for the error expression C, and the loss function parameter €.

— 7 1 — The kernel variable y kernel parameter is selected such that kernel K satisfies the following equation: 2 ) and — k(xp xq) = exp (-r [xp Cus is Xp and 0* is About .qth 4 pth characteristics vectors Optimizing the choice of three variables makes it easier to get good performance levels The modular approach is that mutual validity is time consuming The following constraints are used for the penalty variable and the variable function loss parameter loss function parameter all dla p. C= max(|y — 3. ay, ly + 3. ay) logn £E=T.0p |[— n where 0 < 7 is dad mean value of ground truth values “ - 0 is the standard deviation of the ground truth values “ ob - is the standard deviation of the noise of the ground truth values; - © is the number of learning data ; and se - 5 is a constant value .constant value The student chooses to set the constant value > at 3 and for the kernel variable 7 the variable is set at the average value of a list of all values for the The benchmark is between XP and 0* for all possible 0's and 0's. at the end of learning shad 51060; A law is obtained that facilitates the estimation of spatial density in a predetermined spatial area.

— 8 1 — in the specific case described herein; The law is a regression model, where the learning technique used is SVR technology. The second phase 02 of counting in the region includes three steps: acquisition step (S202), acquisition step 4, and estimation step 5206. In the acquisition step 2m, an image is obtained By asst 30. The image includes at least the predefined spatial area ZS in the acquisition step 5204; the properties specified in the extraction step ©5106 are acquired for the image available in the acquisition step 5202. The acquisition step 5204 includes a slicing step 52048 and an extraction step 52045.

A slicing step of 38 causes is performed using an algorithm to detect the stacking of the data in the image. The slicing process makes it easier to extract the properties associated with data stacking rather than the properties related to locations where there is no data stacking. Slicing optimization enhances the overall performance of the counting method.

5 The 52040 extraction step is similar to the ©5106 extraction step described in the first phase .P1 in that only the pixels where the data stack is detected are extracted; At the end of the acquisition step S204 ¢ the feature vectors are obtained in each pixel of the image for which the data stack has been detected.

0 in estimation step 5206; The spatial density of objects in a predetermined area is estimated. in the end; The law obtained at the end of the first phase (PL) is applied to the acquired image. The link is represented by Arrow 50 in Figure 3.

— 9 1 — more specifically; The regression model obtained at the end of prophase 1 is used to obtain the spatial density dad. In coal terms the model is used on each property vector to infer the die estimate of the value assumed by the function] at the corresponding pixel m. Furthermore, the estimation is done only in interpolated areas and the function F is the result of Gaussian stagnation ¢ It is possible for the function F in some cases to assume values that cannot be ignored outside the interpolated area. To estimate the values assumed by the function “outside the sectioned area”, the spatial density values obtained using image convolution are extrapolated using the measured kernel K such that: (ky — ug)? + (ky — vp)? = V (ky, kg), (ks, kp) and is 0 - the constant 0n is defined by the equation 'Ug =U, — f. (ky — uy) VOCE - is defined by the equation (m7 - =v, — B.(k, v and o .

= Q. —————dllxdll mode is constant n -

pe is delimited by po > 2 --6) +2( - 1) + . for VC UC - are the coordinates of the kernel center; These coordinates are tunable by the processing unit 10. The variable a is for example in the binary slicing technique. At the end of the estimation step 5206; An estimation of the spatial density of a number of pedestrians in a predetermined spatial area (ZS) is obtained optionally; The second phase, P2, includes an induction step through which a number of passers-by are inferred

0 from the estimated spatial intensities in the estimation step.

For example; The number of pedestrians in a predetermined area is obtained by merging over a predetermined area altogether. The method makes it easy to obtain an elemental analysis and act like a self-learning system. For the current field method, the proposed method uses a learning method that actually includes an easily obtainable database. As a matter of fact in the methods of the current field; The method is done by learning using actual data; Which includes manual annotation by data engine for a learning base. Furthermore it"; The learning base of the proposed method is of better quality than the learning base of the current domain method. In practice, the variability and representability of learning data is controversial for current field methods. on dag select; when used; in the current domain rules; There are extreme cases» i.e.; When the spatial density is greater than 50 people in the area; Rarely in edule data as this condition rarely occurs in application. Furthermore it; The footnote entered by the operator is virtually unreliable; Especially when the data stack is very dense or the accuracy is limited; These problems can be added to others. In contrast, 5 leaders are generated to represent future use cases for Detector 30; any; Numerous images generated facilitating the examination of very different spatial densities ranging from very non-density to very dense. Furthermore it; The method is severe in terms of changing the position and/or orientation of the detector. In fact, the vision variable is valuable. The lad learning data is generated using the same 30 Jie 0 tec camera that is used in a row. in the end; Camera-related data is provided 30; Jie Inherent Focal Length or Pintrinsic Image Size Variable or Pposition Orientation Variable (Porientation) Thanks to the variable defined by the camera 30; the learning (LT) database is built which is statistically modeled Sas He can be counted on.

Solidity is confirmed by the fact that the change to camera 30 only assumes a re-implementation of the PT without acquiring ground images; which are irreplaceable in the current field method. The method is useful where the ground truth is immediately available; The annotation is automated and any change on camera 30 can be taken into account directly by learning the regression model without assumption

Provide a new learning group. The method specifically applies data stack oversight through video monitoring for security or transmission purposes. in said use; The method facilitates the estimation of the spatial density of the data stack and the counting of the number of passers-by in the data stack.

0 repeatedly” in this mode; The acquisition step is performed continuously so that phase 2 can be performed repeatedly on a continuous video stream. The proposed method is related to computer-assisted vision; The computer that performs the learning process. Hence, the method can be performed using any computer or other device. Multiple systems can be used with programs that perform the previous method; However, it is possible to consider using dedicated hardware

shay 5 the previous method; The method can be implemented by means of a monitoring device capable of counting objects in a predetermined spatial area (ZS) that includes a 30 detector capable of acquiring one image At least the image includes a predefined spatial space ZS and a processing module 10. A processing unit 10 is suitable for counting objects in a predefined spatial space ZS and is capable of

0 select » saving step 5102; Generation step (S104 determination step 5106; acquisition step 5204 and estimation step 5206. Module 10 is suitable for receiving detector-bound variable 30) Received variables include at least one variable associated with detector focal length 30; variable associated with detector height 30 A variable related to the inclination of the detector 30 with respect to the predetermined spatial area ZS

Furthermore it"; The proposed models are not tied to a specific programming language. casually; This means that many programming languages can be used to perform the previously described method. The previously described method and models can be combined with each other; AS or partially; To find other models of the invention.

Claims (1)

Claims 1- A method for counting objects in a predetermined spatial area (25); The method includes: - a first unsupervised learning phase alas; It includes the following steps: - Selecting a set of different spatial densities for objects in a predefined spatial area (25); Provide variables for a detector (30) capable of acquiring a single image on (BY) the image includes the predefined spatial area (ZS) provided variables include at least one variable related to the focal length of the detector (30); a variable related to the height of the detector (30) and a variable related to the inclination of the detector (30) with respect to a predetermined spatial area (25); 0 - Generate a database containing many images for each selected spatial density » The generation step uses the variables provided in the custom step; f- Determining a law that facilitates the estimation of spatial densities in a predetermined spatial area using the generated education rule; - A second phase of counting in a pre-determined spatial area (ZS) that includes the following steps: - Obtaining an image that includes the pre-determined spatial area (ZS) through the detector (30); f- Estimating the spatial densities of objects in a predefined spatial area (ZS) by applying the law to the acquired image, where: the selection step is implemented by a processing module (10); The receiving step is performed by a processing module (10), the generation step is performed by a processing module (10); the selection step is performed by a processing module (10); the estimation step is performed by a processing module (10); — 4 2 — The acquisition step is performed by the detector (30(¢) and the receive step is performed by a processing module (10). 2- The method according to claim 1, where the generation step includes the step of correcting the perspective defects associated with the detector (30). 3- The method according to claim 1; The method involves the step of slicing an image to identify the pixels that contain an object. 4- The method according to claim 1; where the selection step is made by learning; Learning includes 0 feature extraction; the extracted properties are also used in the estimation step. 5- method according to claim 1 where the selection step is made by support vector regression 5 6- method according to claim 1; where in the allotted step; The variable associated with the detector (30) is a Pintrinsic property associated with the detector (30) 7- Method according to claim 1, where in the assigned step, a position variable Pposition 0 is provided for the detector Detector and/or the variable orientation of the detector (30). ; when executing 5 software instructions by a computer. 9- a storage medium on which a computer software product is stored; The Computer Software Product includes software code; A software instruction that implements a method for counting objects in a predetermined spatial area pursuant to any of claims 1 through 7 when the software instruction is executed by a computer. ja-10 5 A surveillance device capable of counting objects in a predefined 8 spatial space (ZS) includes: - A detector (30) capable of acquiring at least one image The image includes the predefined spatial area (25) The detector (30) has a variable One of the variables is related to the focal length of the detector (30) A variable 0 is related to the height of the detector (30) A variable related to the inclination of the detector (30) with respect to a predefined spatial area (25); and - a processing module (10) capable of counting objects in a specified spatial area predefined processing module (10) capable of: - selecting a set of different spatial densities in a predefined spatial area (25); - receiving variables associated with the detector (30); the received variables include at least one variable related to the focal length of the detector (30); the variable related to the height of the detector (30); The variable associated with the inclination of the detector (30) 0 with respect to the previously defined spatial area (25); - generating a set of images for each chosen spatial density using the received variables; and - defining a law that facilitates the estimation of spatial densities in a predefined spatial area (ZS) using the learning rule generated as a learning rule; and - estimating the spatial densities of objects in a predefined spatial area 5 (25) by applying the law to the image acquired acquired image M 1 1 B 8: - Nint met p “ i 81 Rt ® FRR & 3 i 1 , ° 1 : k 5 ER RE SER PANE EX ; a I b and a NN 3 except 0 | LJ SL | frend what i ; k Es SER EY 3 k 5 than not | 1 tman mm moh na n mm 5 + c and 1 Fo 9 and k ol x RR or at least no doubt oF Ca I 1 is the surface of the bit for the bit of the bit under the bit of our pS 3 iy i § i YX © 3 bd > 8 i LY IE i i § 3 : x XA x N = 10 2 EB 3 * 5 x :0 w x i i : ? h:kh: i EY 3 3 ka YA but SN 5 REE 3 bos 3 case 8: ES a N 4 ? the RE Cais WEE i 3 - TR 3 EY 3 A : 9 CC SE i.e. x N 2 ? + SEY ILE a ur a a a a a 8 ER SN a RE 3 3 > 3 3 a 5 By HUE ANG J oR RNR 3X 1S x ¥ bo bY pS 3 except BR, aa aa & Hy Nay NE X 3 § = kkkk: Xx 3} ¥ : b CR OE FEE Ae Go wd RN : oA Sa % % 3 § pe the l lt FRR lmgi hk > 1 § i for sa and Eo i 3 RR RRR Ree 3 § i kB i a EE i AE LE 8 “5 8 i i # - 3 : ?? ORF TES IE NY 3 3 Ihe NSM YA Ix 3 oS »SSE REI hand ha pS = 3 ?TN v3 3 ER: JS ED XR TE 8 OX EK kas ¥ 3% ¥ FLT OWE OY OW & NS RB RASS ¥ FA 0 : 0 ?? BEE Sl. (EE i NLT - 1] 8 0 XEN i 3 xX § 3 & OEE 3 8 a : BS N X X 3 © Lech #3 l > ee 3 a : ; 3 : solution TA Ee 8 0 + EY % 3 by 3 A 2 iw : WER RN yA BV SE Bow x 3 8 + 8 & & L : 8 REN RR SE k 8 : 83 N EY # 3 @ : 8 k : : RN : 4 3 3 We om SH UWE Yi Lod 8 3 ?X RIS 8 > 87 A x NEM pat Nd A x 8 . C REE MM AAM: 4 EY i 3 = F : 1 8 i : ESS Ro BE B : ne BF 8 RAR: 13 “835 ¥ = RIN 8 2 5” : O : 8 bos * 8 >< 3 : 8 EN 3 : 8 : EY x x 3 N i X § i b 85 > i Re. b 8” 1 a N 0 3 s 5 EB N : 5 # the IX 3 = 1 3 2 8 I 33 3 « 8 ¥ 8 i FN bY 8 = # 8 for © :5 i + & 3 RY 3 ba X § : 88 } Ia} 8 ¥ + : i 85 EY BS = IN: 8 3 bo 8 B AF 3 H i > RY N i i xX § & N EY Ey N a 5 1 iE X 3 H i X ¥ # i x pS = + & CR RR Tahhat RN Al-Mahd RR Born Laht Al-Hah SS SR RR Dah Laht Al-Hamed SR Al-Jedd Dahi Lajh Mad 5 X i B: AN § FEN 0 HY i . etc m neighborhood i 3 | SIRS 3 3 L 8 SHE ! § 5 ces “ § ra = aE 3 § 5 1 oo § § REE bi 58 RAN “ ee : A 3 8 3 BLE SNE Pas 5, 1 tes stat § ; TX bd Fe 8 3 SUE WE WS, 3 RAEN 3 3 i RAL & : A by ta ood ta WR Nagy, : Fd 3 3 9 i NG 1 days a RS : : 8 OO a J i SE is 1 3 3 5 5 C: F B Ne § 3 “Si 4 A E E E >> T B 1 A § E 0 E < M 1 RE? B 3 LEA THE UNIQUE ST, 1 3 > ¢ Rk RR i 1 3 Fost a s ee SEE i SES EE SEE 3 EY bY RNG, 1 ? ? + 0 Sashat Nouni & SORE Ny 1 TER N § 3 8 > a 1 2 ? A 3 # 100 : EY ES EY ES RL B ENR #À< to wd § 3 Noa REG SRS B SE i M 3 3 WINER A X > N ? m * s: 3 § 3 £ 1 m 3 b. No 5 yh mmm mmm mmm mmm JRE oN : § EN PARIS 3 i 3 oe EONAR or 4 N mein Rs § en Ted ES Sued Authority for Intellectual Property RE .¥ + \ A 0 § 8 Ss o + < M SNE A J > E K J TE I UN BE Ca a a a ww > Ld Ed H Ed - 2 Ld, provided that the annual financial consideration is paid for the patent and that it is not null and void for violating any of the provisions of the patent system, layout designs of integrated circuits, plant varieties and industrial designs, or its implementing regulations. Ad Issued by + bb 0.b The Saudi Authority for Intellectual Property > > > This is PO Box 1011 .| for ria 1*1 v= ; Kingdom | Arabic | For Saudi Arabia, SAIP@SAIP.GOV.SA