Information processing device, system, and method

Abstract

To provide an information processing device, a system, and a method capable of clustering a group of images captured by physically separated cameras and specifying an event estimated from the image.SOLUTION: In a machine learning system 100, an inference unit of an information processing device that achieves a collection and analysis device includes: a binary conversion unit that binarizes each of elements constituting images 121 to 12N captured by respective cameras 111 to 11N disposed at a plurality of points to generate binary observation data about the image; a clustering unit that classifies the observation data into one cluster among a plurality of clusters based on a mixed probability distribution having a maximum likelihood parameter estimated by machine learning; and an event assignment unit that determines an event associated with the classified cluster.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, a system, and a method. More specifically, the present disclosure relates to an information processing apparatus, a system, and a method for estimating an occurring event by using a plurality of camera images.

Background Art

[0002] In recent years, research on machine learning has been widely conducted. In particular, with the development of a technique called deep learning, learning modules that exhibit performance equal to or higher than that of human recognition ability have become available.

[0003] As an application example of machine learning, as shown in Patent Document 1, a surveillance camera or the like is arranged, and an object such as a person captured in an image captured by these cameras is detected by a neural network model for learned object detection, and the object detected by using a neural network model for learned object recognition is recognized. Devices and systems that perform such operations are known. In this system, it is possible to recognize the objects captured in the images captured by the surveillance cameras.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in the technique of Patent Document 1, it is not possible to image a monitoring target with a plurality of physically separated cameras and identify an occurring event based on the captured image group. It is very useful to extract features by machine learning from an image group captured by physically separated cameras and identify an event occurring in the monitoring target, such as a natural disaster or a traffic condition.

[0006] In view of the above, an object of the present disclosure is to provide an information processing apparatus, a system, and a method that enable clustering of a group of images captured by physically separated cameras and identification of events estimated from the images.

Means for Solving the Problems

[0007] An information processing apparatus according to an embodiment of the present disclosure includes a binarization unit that generates observation data consisting of binary values for the images by binarizing each of the elements constituting the images captured by each camera arranged at a plurality of locations, a clustering unit that classifies the observation data into one of a plurality of clusters based on a mixture probability distribution having maximum likelihood parameters estimated by machine learning, and an event assignment unit that determines an event associated with the classified cluster.

[0008] A method according to an embodiment of the present disclosure is a method implemented in an information processing apparatus, including generating observation data consisting of binary values for the images by binarizing each of the elements constituting the images captured by each camera arranged at a plurality of locations, classifying the observation data into one of a plurality of clusters based on a mixture probability distribution having maximum likelihood parameters estimated by machine learning, and determining an event associated with the classified cluster.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4A

Figure 4B

Figure 5A

Figure 5B

Figure 6

Mode for Carrying Out the Invention

[0010] [Description of Embodiments of the Present Disclosure] First, the contents of the embodiments of the present disclosure will be listed and described. One embodiment of the present disclosure has the following configuration.

[0011] 〔Configuration 1〕 The characteristic configuration of the information processing apparatus for achieving the above object is a binarization unit (42) that generates observation data consisting of two values for the image by binarizing each of the elements constituting the images captured by each camera arranged at a plurality of points, and a clustering unit (43) that classifies the observation data into one of a plurality of clusters based on a mixed probability distribution having a maximum likelihood parameter estimated by machine learning, and an event assignment unit (44) that determines an event associated with the classified cluster.

[0012] 〔Configuration 2〕 Another characteristic configuration of the information processing apparatus according to the present disclosure is that the mixed probability distribution is a mixed Bernoulli distribution.

[0013] 〔Configuration 3〕 Another characteristic configuration of the information processing apparatus according to the present disclosure lies in that the machine learning is maximum likelihood estimation by the EM (Expectation Maximization) algorithm.

[0014] 〔Configuration 4〕 Another characteristic configuration of the information processing apparatus according to the present disclosure further includes a data storage unit (50) that stores a first lookup table (53) for determining the correspondence between the cluster and the event, and the event assignment unit determines the event to be assigned to the image by referring to the first lookup table.

[0015] 〔Configuration 5〕 Another characteristic configuration of the information processing apparatus according to the present disclosure is that the data storage unit stores a second lookup table for determining the correspondence between the identifier of each camera and the position of the camera, acquires the identifier of the camera associated with the image, and further includes a location specifying unit (45) that specifies the location of the camera corresponding to the acquired camera identifier by referring to the second lookup table.

[0016] 〔Configuration 6〕 Another characteristic configuration of the information processing apparatus according to the present disclosure is that the data storage unit stores a set of training data obtained by binarizing each element constituting the image, and further includes a machine learning unit (48) that executes machine learning based on a mixture probability distribution using the set of training data to estimate the maximum likelihood parameters of the mixture probability distribution. The training data of the image is associated with an event in advance.

[0017] 〔Configuration 7〕 Another characteristic configuration of the information processing apparatus according to the present disclosure lies in that the event is the traffic congestion degree of the road monitored by the camera or the water level degree of the river monitored by the camera.

[0018] 〔Configuration 8〕 Another characteristic configuration of the system according to the present disclosure is that it includes the information processing apparatus according to any one of claims 1 to 7 and a plurality of the cameras respectively arranged at different locations of the monitoring target.

[0019] 〔Configuration 9〕The characteristic configuration of the method for achieving the above object is a method implemented in an information processing apparatus, which includes generating observation data consisting of binary values for the image by binarizing each of the elements constituting the image captured by each camera arranged at a plurality of locations (S304), classifying the observation data into one of a plurality of clusters based on a mixed probability distribution having maximum likelihood parameters estimated by machine learning (S306), and determining an event associated with the classified cluster (S308).

[0020] 〔Configuration 10〕Another characteristic configuration of the method according to the present disclosure is that the determination of the event is made by referring to a first look-up table that defines the correspondence between the cluster and the event (S308).

[0021] 〔Configuration 11〕Another characteristic configuration of the method according to the present disclosure further includes acquiring an identifier of the camera associated with the image, and specifying the location of the camera corresponding to the acquired camera identifier by referring to a second look-up table that defines the correspondence between the identifier of each camera and the location of the camera (S310).

[0022] [Details of Embodiments of the Present Disclosure] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. It should be noted that the following description is merely an example and is not intended to limit the technical scope of the present invention to the following embodiments. In the drawings, the same or similar elements are assigned the same or similar reference numerals, and duplicate descriptions regarding the same or similar elements may be omitted in the description of each embodiment. Also, the features shown in each embodiment are applicable to other embodiments as long as they do not conflict with each other. However, the embodiments of the present disclosure are not necessarily limited to such a mode. It will be apparent to those skilled in the art that the embodiments of the present disclosure can take various forms included in the scope defined in the claims.

[0023] FIG. 1 is a diagram schematically showing a configuration example of a machine learning system 100 according to an embodiment of the present disclosure. As shown in FIG. 1, the machine learning system 100 includes a plurality of cameras 11 1 ~11 N and a collection analysis device 40. N is an integer of 2 or more. The cameras 11 1 ~11 N are surveillance cameras that acquire images of the locations where they are arranged in real time. The cameras 11 1 ~11 N are respectively arranged at different locations to be monitored. For example, when the object to be monitored is a road, the cameras 11 1 ~11 N are traffic surveillance cameras arranged at different locations on the road (such as highway interchanges and bridges), and when the object to be monitored is a river, they are watershed cameras arranged at different locations on the river (such as upstream, midstream, and downstream). The cameras 11 1 ~11 N may be cameras that acquire black-and-white images or cameras that capture color images or videos.

[0024] Hereinafter, each entity constituting the machine learning system 100 shown in FIG. 1 will be described.

[0025] The cameras 11 1 ~11 N are respectively connected to the collection analysis device 40 via a data network (DN) 70. The cameras 11 1 ~11 N respectively acquire images and transmit the acquired images to the collection analysis device 40 via the DN 70. FIG. 1 shows the image 12 1 acquired by the camera 11 1 , the image 12 2 acquired by the camera 11 2 , the image 12 3 acquired by the camera 11 3 , the image 12 4 acquired by the camera 11 4 , the image 12 N acquired by the camera 11 N as examples respectively. The images 12 1 from 12N They are acquired at approximately the same time respectively.

[0026] The collection analysis device 40 includes an inference unit 41, a parameter generation unit 47, and a data storage unit 50.

[0027] The inference unit 41 includes a binary conversion unit 42, a clustering unit 43, an event assignment unit 44, and a location identification unit 45. The configuration of the inference unit 41 will be described in detail later.

[0028] The parameter generation unit 47 includes a machine learning unit 48 that performs machine learning based on a mixture probability distribution using the training data set 51 and estimates the maximum likelihood parameters of the mixture probability distribution. The configuration of the parameter generation unit 47 will be described in detail later.

[0029] Stored in the data storage unit 50 are a training data set 51, parameter data 52 indicating a parameter group of a mixture probability distribution having maximum likelihood parameters estimated by machine learning using the training data set 51, a first look-up table 53, and a second look-up table 54. The training data set 51 is a data set used for generating the parameter data 52 for the parameter generation unit 47.

[0030] The parameter generation unit 47 stores the parameter data 52 in the data storage unit 50 at an appropriate timing. The inference unit 41 can read out and utilize the parameter data 52.

[0031] The collection and analysis device 40 of the above-described machine learning system 100 may be implemented by a single computer including one or more processors, or alternatively, may be implemented by a plurality of computers interconnected via a communication path. All or part of each functional unit of the collection and analysis device 40 can be realized by one or more processors including one or more arithmetic units (Processing Units) that execute processing by the code (instruction group) of a computer program read from a non-volatile memory (computer-readable recording medium). For example, as the arithmetic unit, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or an NPU (Neural Network Processing Unit) can be used.

[0032] FIG. 2 is a schematic configuration diagram of an information processing device (computer) 200, which is an example of a hardware configuration for realizing the collection and analysis device 40 according to an embodiment of the present disclosure. The information processing device 200 includes a processor 201 including a plurality of processor cores μC,..., μC, a random access memory (RAM: Random Access Memory) 202, a non-volatile memory 203, a large-capacity memory 204, an input / output interface 205, and a signal path 206. The signal path 206 is a bus for interconnecting the processor 201, the RAM 202, the non-volatile memory 203, the large-capacity memory 204, and the input / output interface 205. The RAM 202 is a data storage area used when the processor 201 executes digital signal processing. The non-volatile memory 203 has a data storage area in which the code (instruction group) of a computer program executed by the processor 201 is stored.

[0033] Next, with reference to FIGS. 3 to 5, the processing flow of the inference unit 41 of the collection and analysis device 40 will be described in detail below. FIG. 3 is a flowchart showing an example of a procedure 300 such as event assignment processing by the inference unit 41 according to an embodiment of the present disclosure.

[0034] First, in step S302, the binary conversion unit 42 (FIG. 1) collects the images captured by each camera 11 1 ~11 N respectively. The images are the images captured in real time by each camera 11 1 ~11 N at approximately the same time.

[0035] Next, in step S304, the binary conversion unit 42 generates observation data x n consisting of binary values representing black and white two-tone for each of the collected images for all elements constituting the image. The elements constituting the image may be pixels or blocks consisting of a plurality of pixels. If the number of pixels of the images captured by each camera 11 1 ~11 N is different, the binary conversion unit 42 can adjust the number of blocks for dividing the image according to the number of pixels so that the images captured by each camera are composed of the same number of blocks. That is, the number of divisions may be fixed or variable according to the conditions of the input image.

[0036] FIGS. 4A and 4B are schematic diagrams for explaining the binarization of each of a plurality of elements included in the observed image 400 according to an embodiment of the present disclosure. FIG. 4A illustrates a part of the image 400 captured by one camera 11 at a certain time. The image 400 is composed of a plurality of elements, and each element is represented in grayscale. In the illustrated example, elements 401 and 403 are close to black, and element 402 is close to white.

[0037] The binary conversion unit 42 binarizes each element (block or pixel) constituting the image, and for the image, observation data x ncan be generated. For example, when the element is a block, the average pixel value of the colors of each pixel constituting the block is obtained, and it is determined whether this average pixel value is greater than or less than a threshold value, and the color of each block is simplified (binary conversion to white and black). Alternatively, the pixel value of the brightest pixel within the block may be used as the pixel value of the block, and the color of the block may be binary-converted based on whether this pixel value is greater than or less than the threshold value. Even if the observed image 400 is a color image, the binary conversion unit 42 can binary-convert each element based on the pixel value of each element. The binary conversion unit 42 binary-converts all the elements in FIG. 4A of the image 400, for example, in order from the upper left to the lower right of the image 400, and the observation data x of the image 400 n is generated. The binary conversion unit 42 1 ~11 N performs binary conversion on all the images captured by each camera 11 n in the same procedure, and generates the observation data x n respectively.

[0038] FIG. 4B is a diagram illustrating the observation data x generated from the elements (for example, blocks) constituting the observed image shown in FIG. 4A according to an embodiment of the present disclosure. In the example of FIG. 4B, when the color of the element shown in FIG. 4A is close to black (elements 401, 403), the observation data value is "1", and when the color of the element shown in FIG. 4A is close to white (element 402), the observation data value is "0". The observation data x n is expressed as a binary vector having binary variables taking values of "0" or "1" as elements. In the example of FIG. 4B, the observation data x n of the image corresponding to FIG. 4A n is the bit string (1,0,0…0,1,1,0,0,1…0,1,1,0,0,1…1,1,1,……0,0,1…1,1,1,0,1,1…0,1,0) vector.

[0039] As described above, the parameter generation unit 47 (Fig. 1) stores parameter data 52 indicating a parameter group of a mixed probability distribution having the maximum likelihood parameters estimated by machine learning in the data storage unit 50. The inference unit 41 can read the parameter data 52 from the data storage unit 50 and use the parameter data 52.

[0040] Next, in step S306, the clustering unit 43 (Fig. 1) uses the parameter data 52 read from the data storage unit 50 to perform clustering (cluster classification) using the parameter group of the mixed probability distribution indicated by the parameter data 52. That is, the clustering unit 43 classifies the observation data x n into one of a plurality of clusters based on the mixed probability distribution indicated by the parameter data 52.

[0041] In this embodiment, since each of the binary variables that are elements of the observation data x n can be considered to follow a Bernoulli distribution, a mixed Bernoulli distribution can be used as the mixed probability distribution indicated by the parameter data 52. As the machine learning, a likelihood estimation method based on the EM (Expectation-Maximization) algorithm can be used.

[0042] According to the maximum likelihood estimation method based on the EM algorithm of the mixed Bernoulli distribution, the ratio of the observation data x n belonging to the k-th cluster can be calculated by the burden rate γ nk . The burden rate γ nk is represented by the following equation (1).

[0043]

Equation

[0044] Here, π k , μ k are the maximum likelihood parameters estimated by the maximum likelihood estimation method based on the EM algorithm.

[0045] In step S306, the clustering unit 43 uses the observation data x n to classify it into the cluster corresponding to the highest burden rate among the burden rates γ n1 ,…,γ nK . The observation data x n is classified into a cluster having common features.

[0046] Next, in step S308, the event assignment unit 44 (FIG. 1) refers to the first lookup table 53 and determines the event associated with the cluster into which the observation data x n is classified. In the first lookup table 53 in the data storage unit 50, the correspondence between the cluster and the event is predetermined. The event can be, for example, the traffic congestion level (congested, slightly congested, free) of the road to be monitored, or the water level degree (low water level, medium level, high level) of the river to be monitored.

[0047] FIG. 5A illustrates a first lookup table 53 according to an embodiment of the present disclosure when the event to be estimated is the degree of congestion. In the illustrated example, event 1 (the road to be monitored is free) is assigned to cluster 1, event 2 (slightly congested) is assigned to cluster 2, and event 3 (congested) is assigned to cluster 3. Alternatively, when the event to be estimated is the degree of river flooding, event 1 (the river to be monitored has a low water level lower than the normal water level) can be assigned to cluster 1, event 2 (medium water level) to cluster 2, event 3 (high water level higher than the normal water level) to cluster 3, event 4 (dangerous water level where the risk of flooding increases) to cluster 4, event 5 (flooding) to cluster 5, and so on.

[0048] Next, in step S310, the location specifying unit 45 refers to the second lookup table 54 and specifies the section where the event (for example, the degree of congestion) is occurring.

[0049] FIG. 5B illustrates a second lookup table 54 according to an embodiment of the present disclosure. In the second lookup table 54, camera 11 1~11 N and the identifier of N (e.g., the IP address of the camera), and the camera 11 arranged for the object to be monitored 1 ~11 N has a predefined correspondence with the location of N (e.g., the distance from the starting point). In FIG. 5B, for example, the camera 11 1 is arranged at the starting point and the distance from the starting point is 0 km, and the camera 11 N has a distance of 100 km from the starting point. The location specifying unit 45 obtains the camera identifier corresponding to each observation data x n respectively, and refers to the second lookup table 54 to determine the location (e.g., the distance from the starting point) where the camera corresponding to each camera identifier is arranged. The location specifying unit 45 associates the event associated with the observation data x n with the location of the camera associated with the observation data x n to specify the section or location where the event is occurring. When the object to be monitored is a road, the location specifying unit 45 can specify the section or location where the road is congested, or when the object to be monitored is a river, the section or location where the river is flooded. In the example of FIG. 1, the location specifying unit 45 specifies the section where the camera 11 2 to the camera 11 4 is installed as a congested section.

[0050] The location specifying unit 45 can transmit information including the specified event and the section or location where the event is occurring via the network to another terminal (server or user terminal) not shown in the figure.

[0051] As described above, the event assignment unit 44 performs clustering (cluster classification) based on the mixture probability distribution with the maximum likelihood parameters estimated by machine learning using the observation data x 1 ~11 N for each of the acquired images for each camera 11 n composed of the binary values of each element constituting each image, and can determine the event associated with the cluster to which the observation data x n is classified. The location specifying unit 45 determines the event associated with the cluster to which each observation data x nFrom the identifier of the camera associated therewith, the location where the event is occurring can be specified.

[0052] Next, with reference to FIG. 6, the processing of the parameter generation unit 47 (FIG. 1) will be described in detail below. FIG. 6 is a flowchart showing an example of a procedure 600 for parameter generation processing by the parameter generation unit 47 according to an embodiment of the present disclosure.

[0053] The training data set 51 (FIG. 1) consists of a set of training data x corresponding to the observation data in FIG. 5. n The training data set 51 can be generated by an algorithm that generates a random bit sequence of 0 and 1. The training data set 51 may be created from an image acquired by the camera. Each training data x n is respectively associated with each event in advance. Hereinafter, a method for generating the parameter data 52 based on the training data set 51 will be described.

[0054] In step S602, the machine learning unit 48 (FIG. 1) reads the training data set 51 from the data storage unit 50.

[0055] Next, in step S604, the machine learning unit 48 uses the read training data set 51 to perform machine learning based on the mixture probability distribution and estimates the maximum likelihood parameters of the mixture probability distribution.

[0056] Next, in step S606, the machine learning unit 48 stores, as the parameter data 52, the parameter group of the mixture probability distribution having the estimated maximum likelihood parameters in the data storage unit 50.

[0057] Here, the procedure of the maximum likelihood estimation method by the EM algorithm of the mixture Bernoulli distribution executed by the machine learning unit 48 in step S604 will be described below.

[0058] N binary vectors x 1 , x 2 , …, x NThe set, i.e., the binary dataset X, follows a mixture Bernoulli distribution with a set of K parameter vectors μ 1 , μ 2 , …, μ K Let it be the case that the binary dataset X and the parameter dataset M, which is a set of the above, follow a mixture Bernoulli distribution. The binary dataset X and the parameter dataset M are each expressed as follows (T is the transpose symbol).

[0059]

Equation

[0060] Here, the n-th binary vector x n and the k-th parameter vector μ k are each represented as vectors having D elements as follows.

[0061]

Equation

[0062] The elements x n of the binary vector x n (i) (i = 1, 2, …, D) are binary variables taking values of 0 or 1. Each binary vector x n is assumed to be independently generated by the following mixture Bernoulli distribution.

[0063]

Equation

[0064] Here, π is a parameter dataset consisting of K mixing ratios and can be expressed by the following equation.

[0065]

Equation

[0066] The log-likelihood function P(X|M, π) is expressed by the following equation.

[0067] [Mathematics]

[0068] The logarithmic likelihood function P(X|M,π) is expressed by the following formula.

[0069] [Mathematics]

[0070] The EM algorithm is a method for finding the maximum likelihood solution of the parameter datasets M,π of the mixture Bernoulli distribution such that the likelihood (the above likelihood function (X|M,π) or logarithmic likelihood function P(X|M,π)) is maximized.

[0071] First, the machine learning unit 48 executes the initial step. That is, it initializes the parameter data M = {μ 1 , μ 2 , …, μ K}, π = {π 1 , π 2 , …, π K} and the likelihood. The initial value of π K is set to 1 / K so as to be uniform, and the initial value of μ k is set to a normal distribution (mean value = 1 / 2, variance σ 2 ).

[0072] Next, it executes the E step. That is, according to the following formula, it calculates a value γ nk called the responsibility using the current parameter sets M,π. The responsibility γ nk represents the ratio of the binary vector x n belonging to the k-th cluster.

[0073] [Mathematics]

[0074] Next, the machine learning unit 48 executes the M step. That is, according to the following formula, using the current load factor γ nk the parameter sets M and π are updated.

[0075]

Equation

[0076] Next, the machine learning unit 48 determines whether a predetermined convergence condition is satisfied. Specifically, if the likelihood converges within a predetermined numerical range, the parameter sets M and π converge, or both the likelihood and the parameter sets M and π converge, it may be determined that the convergence condition is satisfied. When it is determined that the convergence condition is not satisfied, return to the E step and perform iterative calculations. When the predetermined convergence condition cannot be obtained even after iterative calculations for a predetermined number of times or a certain period of time, any one or all of the parameter sets M and π and the number of clusters K may be changed, and the process from the initial step may be executed.

[0077] Although the embodiments of the present invention have been described above, the above-described embodiments of the invention are for facilitating the understanding of the present invention and do not limit the present invention. The present invention can be changed and improved without departing from its gist, and it goes without saying that equivalents of the present invention are included. Further, any combination of embodiments and modifications is possible within the range that can solve at least a part of the above-described problems or exhibit at least a part of the effects, and any combination or omission of each component described in the claims and the specification is possible.

Explanation of Reference Numerals

[0078] 100... Machine learning system 11 1 ~11 N ... Camera 40... Collection and analysis device 41... Inference unit 42... Binary conversion unit 43... Clustering unit 44... Event Assignment Section 45... Location Identification Section 47... Parameter Generation Section 48... Machine Learning Section 50... Data Storage Section 51... Training Data 52... Parameter Data 53... First Lookup Table 54... Second Lookup Table 70... Data Network

Claims

1. a binary conversion unit that generates binary observation data for an image captured by each of the cameras disposed at a plurality of locations by binarizing each of elements that constitute the image; a clustering unit that classifies the observation data into one of a plurality of clusters based on a mixture probability distribution having a maximum likelihood parameter estimated by machine learning; an event assignment unit for determining an event associated with the classified cluster; An information processing device comprising:

2. The information processing device according to claim 1 , wherein the mixed probability distribution is a mixed Bernoulli distribution.

3. The information processing device according to claim 1 , wherein the machine learning is maximum likelihood estimation using an Expectation Maximization (EM) algorithm.

4. A data storage unit stores a first lookup table that defines a correspondence between the clusters and the events, The information processing apparatus according to claim 1 , wherein the event allocation unit determines the event to be allocated to the image by referring to the first lookup table.

5. the data storage unit stores a second lookup table that defines a correspondence between an identifier of each of the cameras and a position of the camera; The information processing apparatus according to claim 4 , further comprising a location identification unit that acquires an identifier of the camera associated with the image, and identifies a location of the camera corresponding to the acquired identifier of the camera by referring to the second lookup table.

6. the data storage unit stores a set of training data in which each element constituting the image is binarized; A machine learning unit that performs machine learning based on a mixture probability distribution using the set of training data to estimate maximum likelihood parameters of the mixture probability distribution, The information processing apparatus according to claim 4 , wherein the training data of the images is previously associated with an event.

7. The information processing device according to claim 1 , wherein the event is a degree of traffic congestion on a road monitored by the camera, or a water level of a river monitored by the camera.

8. An information processing device according to any one of claims 1 to 7; A plurality of cameras are arranged at different points of a monitoring target, The system according to claim 1 , comprising:

9. A method implemented in an information processing device, comprising: generating binary observation data for an image captured by each of the cameras disposed at a plurality of locations by binarizing each of the elements constituting the image; classifying the observed data into one of a plurality of clusters based on a mixture probability distribution having a maximum likelihood parameter estimated by machine learning; determining events associated with the classified clusters; A method comprising:

10. The method according to claim 9 , wherein the event is determined by referring to a first lookup table that defines a correspondence between the cluster and the event.

11. acquiring an identifier of the camera associated with the image, and referring to a second lookup table that defines a correspondence relationship between the identifier of each camera and a position of the camera, to identify a location of the camera that corresponds to the acquired identifier of the camera; The method of claim 9 further comprising:

Images