JP2018093266A - Information processing device, information processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To appropriately set a region for accurately detecting abnormal behaviors from a monitoring object image.SOLUTION: An information processing device 40 includes: acquisition means for acquiring movement information on an imaging object in an image; and division means for dividing, on the basis of the movement information acquired by the acquisition means, a movement region of the imaging object in the image into a plurality of sub-regions corresponding to a movement inclination of the imaging object. The information processing device learns a movement of a learning object by the sub-region in the sub-regions into which the division means divided the movement region, performs an abnormal movement detection process using learning data corresponding to a sub-region, and thereby can reduce erroneous detection of abnormal movements.SELECTED DRAWING: Figure 4

Description

  The present invention relates to an information processing apparatus, an information processing method, and a program.

  In recent years, with a shift from an analog camera to a network camera, a surveillance system has been developed to have a high-resolution camera, a high frame rate, and a large-scale system using more than several hundred cameras. Such a monitoring system is required to monitor not only a single store such as a conventional department store, a supermarket, a convenience store, and a rental video store, but also a shopping mall and the entire region.

Under these circumstances, image analysis technology that recognizes people and cars from the image information of surveillance cameras has entered the practical stage, and the system automatically monitors the images of hundreds of surveillance cameras on behalf of people. There is a growing need for monitoring systems using technology.
As such a monitoring support technique, Patent Document 1 discloses a technique for learning a human normal behavior and detecting a behavior deviating from the learned normal behavior as an abnormal behavior. This patent document 1 discloses that one learning data is selected from a plurality of learning data according to the number of people in the monitoring target image, and abnormal behavior is detected using the selected learning data. Yes.

JP 2009-266052 A

However, in the technique described in Patent Document 1, abnormal behavior is detected using the selected single learning data, and thus there are a plurality of regions in which the behavior tendency of a person is different in an image to be monitored If you do not support.
For example, a scene where an escalator and a staircase are provided can be considered. On the stairs, it is normal behavior for the user to stop in the middle of the stairs or to go up and down the stairs. On the other hand, in an escalator, it is abnormal behavior that a user stops moving on the way or the user runs backward on the escalator. Thus, if an abnormal behavior detection process using a single learning data is performed as in the technique of Patent Document 1 on an image in which regions having different abnormal behavior and normal behavior exist at the same time, false detection is performed. Many will occur.

If an instruction to divide the area is received from the system setter in advance and normal behavior can be learned for each divided area, it is possible to detect abnormal behavior with high accuracy, but setting to divide into multiple areas for each camera. It takes a lot of work to do. In particular, when the number of cameras is several hundreds as in city monitoring, the effort is enormous.
Therefore, an object of the present invention is to appropriately perform region setting for accurately detecting abnormal behavior from a monitoring target image.

  In order to solve the above-described problem, an aspect of the information processing apparatus according to the present invention includes an acquisition unit that acquires motion information of an imaging target in an image, and the image based on the motion information acquired by the acquisition unit. Splitting means for splitting the motion area of the imaging target in a plurality of partial areas corresponding to the motion tendency of the imaging target.

  According to the present invention, it is possible to appropriately perform region setting for accurately detecting abnormal behavior from a monitoring target image.

It is a network block diagram which shows an example of the monitoring system. It is a figure which shows an example of the hardware constitutions of information processing apparatus. It is an example of the scene containing the some area | region from which an operation | movement tendency differs. It is a functional block diagram of an information processor. It is a conceptual diagram which shows the flow from learning to operation | use. It is a flowchart which shows the procedure of an area | region division process. It is an example of the division | segmentation into the small area | region in an area | region division process. It is an example of the feature-value used by area | region division processing. It is an example of a clustering result. It is a flowchart which shows the procedure of an abnormal action detection process. It is an example of the warning display in abnormal behavior detection. It is a conceptual diagram which shows the flow from the learning which added the area | region correction process to operation. It is an example of the user interface for area correction.

DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings.
The embodiment described below is an example as means for realizing the present invention, and should be appropriately modified or changed depending on the configuration and various conditions of the apparatus to which the present invention is applied. It is not limited to the embodiment.
(First embodiment)
FIG. 1 is a network configuration diagram showing an example of the operating environment of the monitoring system in the present embodiment. In this embodiment, the monitoring system is applied to a network camera system.
The network camera system 10 includes at least one network camera (hereinafter simply referred to as “camera”) 20, a recording server device 30, an information processing device 40, and a client device 50. The camera 20, the recording server device 30, the information processing device 40, and the client device 50 are connected by a network 60. The network 60 may be a LAN (Local Area Network), the Internet, a WAN (Wide Area Network), or the like. The physical connection form to the network 60 may be wired or wireless. Further, in FIG. 1, the number of cameras 20, the recording server device 30, the information processing device 40, and the client device 50 connected to the network 60 is not limited to the number shown in FIG.

The camera 20 is an imaging device capable of imaging a monitoring target and transmitting the captured video (image) via the network 60. The camera 20 transmits an image to the recording server device 30 via the network 60. The camera 20 can be installed in a large commercial facility or city. The recording server device 30 receives an image from the camera 20, stores the received image, and transmits the received image to the information processing device 40 and the client device 50.
The information processing apparatus 40 is a video analysis server. The information processing device 40 can perform image analysis processing on the image received from the recording server device 30 and transmit the analysis result to the client device 50. In this embodiment, the case where an image captured by the camera 20 is transmitted to the information processing apparatus 40 via the recording server apparatus 30 will be described. However, the information processing apparatus 40 directly receives an image from the camera 20. You may receive it.

The client device 50 can be configured by a personal computer (PC), for example. In the present embodiment, the client device 50 is a terminal for an operator deployed in a security room to refer to the analysis result of the information processing device 40, for example. The client device 50 is not limited to a PC, and a tablet terminal or the like can be used instead. The client device 50 may have a function of receiving an instruction from an operator and acquiring an analysis result from the information processing device 40 based on the instruction. Further, the client device 50 may have a function of receiving a request from an operator and instructing the recording server device 30 to perform recording based on the request.
In the present embodiment, the recording server device 30 and the information processing device 40 are described as being built on an on-premises server. However, the present invention is not limited to the above, and these devices may be constructed on a cloud computing platform.

FIG. 2 is a diagram illustrating an example of a hardware configuration of the information processing apparatus 40.
The information processing apparatus 40 includes a CPU 41, a RAM 42, a ROM 43, an HDD 44, an output device 45, an input device 46, a communication device 47, and a system bus 48.
The CPU 41 is a processor that comprehensively controls the operation of the information processing apparatus 40, and controls each component (42 to 47) via the system bus 48. The RAM 42 has an area for temporarily storing programs and data loaded from the ROM 43 and the HDD 44. Further, the RAM 42 has an area for temporarily storing data received from an external device via the communication device 47. The RAM 42 also has a work area used when the CPU 41 executes each process.

  The ROM 43 is a non-volatile memory that stores an image analysis program and the like necessary for the CPU 41 to execute image analysis processing. This image analysis program may be an application program that runs on an OS (operating system). The program may be recorded on the HDD 44 or a recording medium (not shown). The CPU 41 can implement various functional operations by reading the OS and application programs from the ROM 43 or the HDD 44, loading them into the RAM 42, and executing them.

The output device 45 includes a monitor that displays various processing results and a user interface screen (UI screen). The input device 46 can be configured by a pointing device such as a keyboard and a mouse, a touch panel, and the like. A user of the information processing apparatus 40 (for example, a system setting person) can input an instruction to the information processing apparatus 40 by operating the input device 46. The communication device 47 is an interface for communicating with an external device connected to the network 60. The system bus 48 connects the CPU 41, the RAM 42, the ROM 43, the HDD 44, the output device 45, the input device 46, and the communication device 47 so that they can communicate with each other.
The recording server device 30 and the client device 50 can also have the same hardware configuration as that in FIG.

  In the present embodiment, the information processing apparatus 40 performs an abnormal operation detection process that detects an abnormal operation that is an operation deviating from the normal operation of the monitoring target in the image as the image analysis process, and the processing result is transmitted to the client device. 50 has a display control function to be displayed on the display screen. In the present embodiment, a case where the monitoring target is a human and the information processing apparatus 40 executes an abnormal behavior detection process for detecting a human abnormal behavior in an image will be described.

  In addition, the information processing apparatus 40 has a function of performing learning processing for learning the operation of an imaging target in an image and generating learning data used for the image analysis processing prior to the image analysis processing. In the present embodiment, a case will be described in which the imaging target is a human and the learning target operation in the learning process is a normal operation (normal behavior). That is, in the present embodiment, the abnormal behavior detection process detects the abnormal behavior of the human in the image using learning data obtained by learning the normal behavior of the human from an image in which the normal behavior is captured in advance. It is processing.

The merit of the method of learning normal behavior instead of abnormal behavior is that all behaviors deviating from normal behavior can be detected as abnormal behavior. Further, when learning an abnormal behavior, it is necessary to actually cause the abnormal behavior to be learned, which takes time for learning. However, when learning a normal behavior, the time can be saved. On the other hand, a disadvantage of the method of learning normal behavior is that the type of abnormal behavior cannot be detected.
In addition, although this embodiment demonstrates the case where a normal action is learned, it is applicable also when learning an abnormal action.

FIG. 3 is an example of an image to be subjected to the abnormal behavior detection process. The scene shown in FIG. 3 is an example of a scene including a plurality of regions having different human behavior tendencies. Here, the human behavior tendency is a tendency of human movement assumed for each region, such as going up and down for stairs, two-way traffic for passages, and movement in a random direction for squares.
In the image 100, a staircase 101 and an up escalator 102 are captured. On the stairs 101, the user moves up or down, respectively. Also, the user can stop in the middle of the stairs 101 or turn back in the middle of the stairs 101. On the other hand, on the escalator 102, the user basically moves in one direction (for example, the ascending direction).

  In the escalator 102, for example, an injury or the like is accompanied when the user runs backward, and therefore it is desirable that the escalator 102 can detect an abnormality when the user moves in a descending direction. However, because the user moves up and down on the stairs 101, if the user moves in the descending direction like the escalator 102 and an abnormality is detected, an undesirable abnormality detection is notified. End up. That is, if an abnormal behavior is detected using the same learning data in the area corresponding to the stairs 101 and the area corresponding to the escalator 102, many false detections occur.

Therefore, in the present embodiment, considering the case where a plurality of regions having different motion trends of the imaging target are included in one image, the motion region of the imaging target in the image is determined based on the motion information of the imaging target in the image. It is automatically divided into a plurality of partial areas corresponding to the movement tendency of the imaging target. Then, abnormal behavior detection is performed using different learning data for each divided partial region.
FIG. 4 is a block diagram illustrating a functional configuration of the information processing apparatus 40. The information processing apparatus 40 includes a video acquisition unit 401, a human body detection unit 402, a region division unit 403, a learning unit 404, a learning data storage unit 405, a region determination unit 406, an abnormal behavior detection unit 407, a warning A notification unit 408 and an area data storage unit 409 are provided. The function of each unit of the information processing apparatus 40 shown in FIG. 4 can be realized by the CPU 41 executing a program. However, at least some of the elements shown in FIG. 4 may operate as dedicated hardware. In this case, the dedicated hardware operates based on the control of the CPU 41.

  The video acquisition unit 401 acquires an image from the recording server device 30 and outputs the acquired image to the human body detection unit 402. The human body detection unit 402 detects a human body from the image input from the video acquisition unit 401 and outputs human body information such as the coordinates, width, height, movement direction, and movement speed of the human body on the image. In the present embodiment, the case where the human body detection unit 402 detects a human body will be described. However, a face of a specific person may be detected. When the monitoring target is a specific object other than a human, the human body detection unit 402 can be an object detection unit that detects the object.

  The area dividing unit 403 aggregates human body motion information (movement direction and movement speed) detected by the human body detection unit 402 and performs image area division. Here, the area dividing unit 403 performs an area dividing process for dividing the motion area of the human body in the image into a plurality of partial areas corresponding to the behavior tendency of the human body. Then, the area dividing unit 403 outputs the result of area division to the area data storage unit 409 as area data. The area data is data indicating each partial area divided by area division, and is generated for each camera 20. An example of the area data will be described later. The area data storage unit 409 stores the area data in an area data storage area of the RAM 42 or the HDD 44, for example.

The learning unit 404 performs learning processing with reference to the region data, and generates learning data for each partial region divided by the region division processing. Specifically, the learning unit 404 learns human normal behavior for each partial region based on an image of normal behavior captured, and generates learning data corresponding to each of the partial regions. Then, the learning unit 404 outputs the generated plurality of learning data to the learning data storage unit 405.
The learning data can be composed of feature amounts for each partial region extracted from the image. As the feature amount, for example, cubic higher-order local auto-correlation (CHLAC) capable of expressing the motion feature of an object in a moving image can be used. Note that when handling feature quantities in time series as in CHLAC, it is necessary to acquire feature quantities over a plurality of frames. Therefore, it is assumed that a feature amount is acquired for a human body region that is large enough to accommodate human behavior.

The learning data storage unit 405 stores the learning data in a learning data storage area of the RAM 42 or the HDD 44, for example. The area determination unit 406 refers to the area data and determines in which partial area the human body that is the monitoring target of the abnormal behavior is divided by the area dividing unit 403. The abnormal behavior detection unit 407 detects the abnormal behavior to be monitored using the learning data corresponding to the partial region determined by the region determination unit 406.
The warning notification unit 408 notifies the client device 50 that the abnormal behavior has been detected when the abnormal behavior detection unit 407 detects the abnormal behavior. The client device 50 that has received the notification displays on the display screen that abnormal behavior has occurred. Thereby, the operator in the security room can recognize that an abnormality has occurred via the client device 50. A display example in the client device 50 will be described later.

Next, a flow until the information processing apparatus 40 learns normal behavior and executes abnormal behavior detection will be described with reference to FIG. Hereinafter, the alphabet S means a step in the flowchart.
First, the camera 20 captures an actual behavior of a person for a predetermined period (several hours to several days), and records the captured image on the recording server device 30. Then, the information processing apparatus 40 performs region division processing and learning processing based on the recorded image. This is called the “learning phase”. Next, the information processing apparatus 40 performs abnormal behavior detection using the learning data generated in the learning phase. This is called the “operational phase”.

  In S <b> 1, the recording server device 30 records a state in which a person actually captured by the camera 20 passes for a predetermined period (several hours to several days). Next, in S <b> 2, the information processing apparatus 40 divides the image into a plurality of partial regions according to the human behavior tendency based on the image of the predetermined period (several hours to several days) recorded in S <b> 1. Split processing is performed. Next, in S <b> 3, the information processing apparatus 40 generates learning data for each partial region divided by the region dividing process. The information processing apparatus 40 may generate the learning data using only the image of the predetermined period used in the region division processing, or the learning data may be generated using an image recorded for an additional several days to improve accuracy. It may be generated. The processing from S1 to S3 is processing in the learning phase.

In S <b> 4, the information processing apparatus 40 performs an abnormal action detection process for detecting an abnormal action on an image to be monitored using the learning data generated in the learning phase.
Hereinafter, the region division processing executed by the information processing apparatus 40 in S2 of FIG. 5 will be described in detail with reference to FIGS.
The process illustrated in FIG. 6 is started, for example, at a timing when the information processing apparatus 40 receives a process start instruction from the user. However, the start timing of the process of FIG. 6 is not limited to the above timing. The information processing apparatus 40 can implement each process illustrated in FIG. 6 by the CPU 41 reading and executing a necessary program. However, as described above, the processing of FIG. 6 may be realized by operating at least a part of each element shown in FIG. 4 as dedicated hardware. In this case, the dedicated hardware operates based on the control of the CPU 41.

  In S <b> 21, the area dividing unit 403 divides the image into a plurality of areas (small areas) having a size equal to or smaller than the partial area. FIG. 7 shows an example in which the scene shown in FIG. 3 is divided into small areas. In the example of FIG. 7, the size of the small area 110 is set to a predetermined fixed size. Note that the size of the small area 110 may be set in advance in a program setting file or the like, or may be set by receiving an input from a system setting person. Further, the size of the small area 110 may be variable, such as increasing the size in front of the screen and decreasing the size in the back of the screen.

In S <b> 22, the area dividing unit 403 calculates a feature amount for each small area 110 with respect to an image recorded for a predetermined period (several hours to several days). In the present embodiment, the region dividing unit 403 can add up the movement direction that is the movement information of the human body detected by the human body detection unit 402 and use the occurrence frequency of the movement direction as a feature amount. An example of the feature amount is shown in FIG.
FIG. 8A is a conceptual diagram in which numbers are assigned in directions in which the human body can move. Here, 16 directions are set as the moving direction of the human body, but more or fewer directions may be set. FIG. 8B shows the result of calculating the feature amount for the two small regions 111 and 112. The feature amount 111a is a feature amount of the small region 111, and the feature amount 112a is a feature amount of the small region 112. Since the small region 111 is a region corresponding to the rising escalator, the occurrence frequency of the direction 1 appears high as the feature amount 111a. On the other hand, since the small area 112 is an area corresponding to a staircase, the moving direction of the user is bidirectional up and down, and there are also users who do not walk straight ahead. For this reason, as the feature amount 112a, a high occurrence frequency appears around direction 1 and around direction 9.

  In S <b> 23, the region dividing unit 403 derives the motion region of the human body by excluding the small region 110 where the moving human body does not exist from the image based on the feature amount for each small region 110. The process of S23 is a process of excluding the small area 110 through which a human body does not pass from the clustering target of S25 described later. In S24, the area dividing unit 403 sets the number of clusters. The number of clusters may be received from the system setter, or may be a fixed value set in advance.

In S <b> 25, the region dividing unit 403 clusters the small regions 110 after the region exclusion in S <b> 23 so that the number of clusters set in S <b> 24 is reached based on the feature amount of each small region 110. For the clustering, unsupervised learning such as the k-means method can be used. An example of the region division result obtained by clustering is shown in FIG.
FIG. 9A shows an original image, and FIG. 9B shows area data 120 obtained as a result of area division. In FIG. 9B, a small area to which the same index is assigned corresponds to one divided partial area. FIG. 9B shows an example in which the number of clusters is 2, and 0: an area where a human body has not passed, 1: an area corresponding to a staircase, and 2: an area corresponding to an escalator, which is divided into partial areas. Is shown. The area data 120 includes information (upper left coordinates, width, height, index) of each small area.

  As described above, the region dividing unit 403 performs clustering of the feature amounts related to the motion information for each small region, so that a set of small regions having the same or similar behavior tendency of the human body is obtained as a partial region corresponding to the behavior tendency. Derived as In addition, the region dividing unit 403 performs clustering after excluding the small region 110 through which people cannot pass from the clustering target. For this reason, when the number of clusters is 2, the image is prevented from being divided into “regions through which people cannot pass” and “regions through which people pass”, and a plurality of operation regions that are “regions through which people pass” Can be divided into subregions.

In addition, although the case where only the moving direction of the human body is used as the human body motion information in the feature amount calculation processing in S22 has been described, the motion information is not limited to the moving direction. For example, if the imaging scene shows a one-way staircase and an escalator that moves in the same direction as the passage of the staircase, the human body is divided into a small area corresponding to the staircase and a small area corresponding to the escalator. The calculation result of the feature amount based on the moving direction of is substantially the same. Therefore, when only the moving direction is used as the human body motion information, the area corresponding to the stairs and the area corresponding to the escalator cannot be divided. Therefore, the movement speed of the human body may be added as the movement information of the human body, and the variation in the movement speed may be added to the feature amount.
In S25, the case of performing clustering according to the predetermined number of clusters set in S24 has been described. However, the number of clusters may be automatically determined by the x-means method or the like.

Next, the abnormal behavior detection process executed by the information processing apparatus 40 in S4 of FIG. 5 will be described with reference to FIG.
The process illustrated in FIG. 10 is started at a timing when the information processing apparatus 40 acquires an image to be monitored, for example. However, the start timing of the process of FIG. 10 is not limited to the above timing. The information processing apparatus 40 can implement each process illustrated in FIG. 10 by the CPU 41 reading and executing a necessary program. However, as described above, at least a part of each element shown in FIG. 4 may operate as dedicated hardware so that the processing in FIG. 10 is realized. In this case, the dedicated hardware operates based on the control of the CPU 41.

The information processing apparatus 40 inputs an image of a predetermined frame (for example, 30 frames) for each camera 20 and executes abnormal behavior detection processing according to the flowchart shown in FIG.
In S41, the human body detection unit 402 detects the human body imaged in the image and outputs human body information (coordinates, widths, and heights of the human body on the image). In S42, the region determination unit 406 refers to the region data 120 and determines in which partial region the human body detected in S41 exists. Here, when a human body exists over a plurality of partial regions, a partial region over which more areas are straddled is selected.

  In S43, the abnormal behavior detecting unit 407 calculates the abnormal behavior score for each of the human bodies detected in S41. First, the abnormal behavior detection unit 407 extracts feature amounts for detecting abnormal behavior with respect to the detected human body. The feature amount extracted here is the same as the feature amount constituting the learning data, and for example, CHLAC or the like can be used. Then, the abnormal behavior detection unit 407 calculates the distance between the detected feature quantity of the human body and the learning data (feature quantity) corresponding to the partial region where the detected human body exists as a score of the abnormal behavior of the human body. To do.

In S44, the abnormal behavior detecting unit 407 determines whether or not the abnormal behavior score calculated in S43 is equal to or greater than a threshold value. If the abnormal behavior score is equal to or greater than the threshold value, the process proceeds to S45. If the abnormal behavior score is lower than the threshold value, the process ends.
In S45, the warning notification unit 408 notifies the client device 50 that abnormal behavior has been detected. At the time of notification, identification information (for example, ID) of the camera 20 that has imaged the abnormal behavior, position information (for example, coordinates) on the image where the abnormal behavior has occurred, and the like are also transmitted. The client device 50 that has received the notification displays a warning on the display screen. Thereby, the operator deployed in the security room can grasp that abnormal behavior has occurred. A display example of the display screen in the client device 50 is shown in FIG.

FIG. 11A is an example of the display screen 200 in a situation where no abnormal behavior has occurred. The display screen 200 includes a video display unit 201 and a map display unit 202. The video display unit 201 displays live view videos of a plurality (9 in FIG. 11) of cameras 20 arranged in a tile shape. The map display unit 202 displays which camera 20 is installed in the building along with a map of the building in which the camera 20 is installed.
FIG. 11B is an example of the display screen 200 in a situation where abnormal behavior has occurred. The display screen 200 includes, in the video display unit 201, an enlarged display unit 203 that displays an enlarged video of the camera 20 that has captured the abnormal behavior. In the enlarged display portion 203, as shown in FIG. 11B, the position where the abnormal action has occurred in the image may be displayed so as to be visible.

  In addition, the map display unit 202 displays the installation position of the camera 20 that captured the abnormal behavior in the building on the map. For example, when the camera 20 installed on the first floor of the building images abnormal behavior, the marker 204 may be displayed at a position corresponding to the installation position of the camera 20 as shown in FIG. With such a display, an operator deployed in the security room can easily grasp where the abnormal behavior has occurred in the building, and can quickly go to the site as necessary.

  As described above, the information processing apparatus 40 according to the present embodiment acquires the motion information of the imaging target in the image, and the motion area of the imaging target in the image is determined based on the acquired motion information. Is divided into a plurality of partial areas corresponding to. When the imaging target is a human body, the information processing apparatus 40 divides the human body motion area in the image into a plurality of partial areas corresponding to the human body behavior tendency based on the motion information of the human body in the image. As described above, the information processing apparatus 40 can automatically perform region division when a plurality of regions having different motion tendencies (behavior tendencies) are included in one scene. Therefore, it is possible to save time and labor for the system setter to divide the area when setting the system.

  The movement information can be information related to the moving direction of the human body. In a scene where an escalator and a staircase are provided side by side, a user usually moves in a certain direction in an area corresponding to the escalator, and a user moves in both directions in an area corresponding to the stairs. Therefore, if information regarding the moving direction is used as the movement information, the area corresponding to the escalator and the area corresponding to the stairs can be appropriately divided. For example, the same applies to a scene where an ascending escalator and a descending escalator are provided, and a scene where an escalator and a passage are provided.

  The movement information may be information related to the moving speed of the human body. In the case of a scene with an ascending stairs and an ascending escalator, the moving direction of the stairs user and the moving direction of the escalator user are the same. I can't. By using the information regarding the moving speed, it is possible to deal with the scene as described above, and appropriate area division is possible. For example, the same applies to a scene in which a one-way passage and a moving sidewalk are provided.

In addition, the information processing apparatus 40 derives a feature amount related to motion information for each of a plurality of small regions obtained by dividing an image by a size equal to or smaller than a partial region, and clusteres the feature amounts of the small regions to respond to the behavior tendency. Perform area division. Here, the feature amount related to the motion information can be the occurrence frequency of the moving direction or the variation of the moving speed. Thereby, the information processing apparatus 40 can divide a plurality of regions having different behavior tendencies easily and appropriately.
Furthermore, the information processing apparatus 40 performs clustering after excluding small regions from which no humans pass from the clustering target. Therefore, the motion region of the human body in the image can be divided into a plurality of regions according to the behavior tendency. That is, as a result of the region dividing process, it is possible to avoid the division into a region where a person passes and a region where a person does not pass.

  In addition, the information processing apparatus 40 learns the operation to be learned for each partial region for a plurality of partial regions divided into regions, and generates learning data corresponding to each partial region. Then, the information processing device 40 determines in which partial area the human body that is the monitoring target in the image exists, and uses the learning data corresponding to the determined partial area, Detect abnormal behavior. In this way, abnormal behavior detection processing is performed using different learning data for each region having a different behavior tendency. Therefore, in the escalator, it is determined as abnormal behavior that the user's movement stops or reverse running on the way, and the above behavior is determined as normal behavior on the stairs, etc. Detection processing can be performed.

  Here, if the operation to be learned is a normal operation (normal behavior), all the behaviors deviating from the normal behavior can be detected as abnormal behavior in the abnormal behavior detection process. Therefore, it is possible to suppress omission of detection of abnormal behavior. In addition, learning is easy because there is no need to actually cause abnormal behavior to learn. On the other hand, if the operation to be learned is an abnormal operation (abnormal behavior), it is possible to determine the type of abnormal behavior in the abnormal behavior detection process, and when notifying the occurrence of abnormal behavior, It can also be notified.

As described above, according to the information processing apparatus 40 in the present embodiment, it is possible to appropriately perform area division for accurately detecting abnormal behavior without taking the effort of the system setter, Abnormal behavior of a human body can be detected with high accuracy.
The monitoring target is not limited to a person, and may be an object other than a person. For example, when the monitoring target is an automobile, abnormal operations such as reverse running of the automobile can be detected appropriately.

(Second embodiment)
Next, a second embodiment of the present invention will be described.
In the first embodiment described above, the information processing apparatus 40 has been described with reference to the case where the abnormal action detection process is executed using the result of the area division process as it is. In the second embodiment, a case will be described in which the information processing apparatus 40 has a function of correcting a partial area divided based on a correction instruction from the system setter after the area dividing process.
The region division processing in the first embodiment described above can automatically divide the region. However, if the image includes a plurality of regions with similar moving directions of the human body, the region division is not performed correctly. There is a fear. If the region division is not performed correctly, normal behavior cannot be learned correctly, and erroneous detection of abnormal behavior may occur.

FIG. 12 is a diagram illustrating a flow until the information processing apparatus 40 learns normal behavior and executes abnormal behavior detection. In FIG. 12, the same step numbers as those in FIG.
After the area dividing process in S2, in S5, the information processing apparatus 40 performs display control for displaying the partial area divided by the area dividing process on the display unit (output device 45). Then, the information processing apparatus 40 receives a correction of the partial area divided by the area dividing process from the system setting person, and corrects the partial area. In S3, the information processing apparatus 40 performs learning for each partial region corrected in S5.

Hereinafter, the area correction process executed in S5 will be specifically described.
FIG. 13 is an example of a UI screen 130 for correcting an area divided by the area dividing process. Although not shown in FIG. 13, it is assumed that an image captured by the camera 20 is displayed as a background image on the UI screen 130 and a small area is displayed in a transparent manner.
FIG. 13A shows a UI screen 130 when a region is corrected in units of small regions. The UI screen 130 in FIG. 13A displays an index indicating which partial area each small area belongs to based on the area data 120 of the partial area obtained by the area dividing process. Yes. For example, the small area 113 belongs to the partial area “1”, and the small area 114 belongs to the partial area “2”. Here, a case where it is desired to correct the small area 114 to the partial area “1” is considered. In this case, the system setting person designates (for example, clicks) the small area 113 belonging to the partial area “1” as indicated by the arrow 131 and then designates the small area 132 to be corrected as indicated by the arrow 132. (For example, click). As a result, the small area 114 can be corrected to the same partial area “1” as the small area 113 specified immediately before.

FIG. 13B shows a UI screen 130 when a region is corrected in units of partial regions obtained by the region dividing process. The modification of the area in units of partial areas corresponds to a process of integrating a plurality of partial areas. For example, the system setting person can set a mode for correcting an area in units of partial areas by pressing a button 133 for instructing area integration.
Here, consider a case where partial area “2” and partial area “3” are integrated to be corrected to partial area “2”. In this case, the system setter inputs an area integration instruction by pressing the button 133 and designates (for example, clicks) the small area 115 belonging to the partial area “2” of the integration source as indicated by the arrow 134. . Next, as shown by an arrow 135, the system setting person designates (for example, clicks) the small area 116 belonging to the partial area “3” to be integrated. Thereby, the area of the partial area “3” is integrated into the partial area “2”.

The region correction method is not limited to the above. For example, after specifying a small area or partial area to be corrected, the index number of the corrected partial area may be directly designated.
As described above, the information processing apparatus 40 according to the present embodiment performs display control for displaying the partial area divided by the area division process on the display unit, and accepts correction of the partial area from the system setter. At this time, the information processing apparatus 40 may accept correction of the range of the partial area in units of small areas that are smaller than the partial area, or may accept correction that integrates a plurality of partial areas. Then, the information processing apparatus 40 corrects the partial area based on a correction instruction from the system setting person. Therefore, it is possible to perform region division more appropriately and to detect abnormal behavior with higher accuracy.

(Other embodiments)
The present invention supplies a program that realizes one or more functions of the above-described embodiments to a system or apparatus via a network or a recording medium, and one or more processors in the computer of the system or apparatus read and execute the program This process can be realized. It can also be realized by a circuit (for example, ASIC) that realizes one or more functions.

  DESCRIPTION OF SYMBOLS 10 ... Network camera system, 20 ... Camera, 30 ... Recording server apparatus, 40 ... Information processing apparatus, 50 ... Client apparatus, 60 ... LAN, 401 ... Video acquisition part, 402 ... Human body detection part, 403 ... Area division part, 404 ... Learning unit, 405 ... Learning data storage unit, 406 ... Area determination unit, 407 ... Abnormal behavior detection unit, 408 ... Warning notification unit, 409 ... Area data storage unit

Claims (15)

Acquisition means for acquiring movement information of an imaging target in an image;
Dividing means for dividing the motion area of the imaging target in the image into a plurality of partial areas corresponding to the motion tendency of the imaging target, based on the motion information acquired by the acquisition means. Information processing apparatus.   The information processing apparatus according to claim 1, wherein the motion information includes information related to a moving direction of the imaging target.   The information processing apparatus according to claim 1, wherein the movement information includes information regarding a moving speed of the imaging target. The acquisition means acquires the motion information for each of a plurality of small areas obtained by dividing the image by a size equal to or smaller than the partial area,
4. The division unit according to claim 1, wherein the dividing unit derives a feature amount related to the motion information for each of the small regions, and clusters the feature amounts of the small regions corresponding to the motion region. The information processing apparatus according to item.   The dividing unit excludes the small region in which the imaging target moving from the image does not exist based on the motion information for each small region acquired by the acquiring unit, and the operation to be a target of the clustering The information processing apparatus according to claim 4, wherein an area is derived. Determining means for determining in which partial area the monitoring target in the image is divided by the dividing means;
6. The apparatus according to claim 1, further comprising a detection unit that detects an abnormal operation of the monitoring target using learning data corresponding to the partial area determined by the determination unit. The information processing apparatus described.   The learning apparatus further comprises learning means for learning the operation of the learning target for each of the partial areas divided by the dividing means and generating learning data corresponding to the partial areas, respectively. Item 7. The information processing device according to any one of items 1 to 6.   The information processing apparatus according to claim 7, wherein the operation to be learned is a normal operation.   The information processing apparatus according to claim 7, wherein the operation to be learned is an abnormal operation. Display control means for displaying a partial area divided by the dividing means on a display unit;
The information processing apparatus according to claim 1, further comprising: a correction unit that receives correction of the partial area divided by the dividing unit and corrects the partial area.   The information processing apparatus according to claim 10, wherein the correction unit receives correction of a range of the partial area in units of areas smaller than the partial area.   The information processing apparatus according to claim 10, wherein the correction unit receives a correction that integrates a plurality of the partial areas. Obtaining motion information of an imaging target in an image;
Dividing the motion area of the imaging target in the image into a plurality of partial areas corresponding to the motion tendency of the imaging target based on the motion information.   The program for functioning a computer as each means of the information processing apparatus of any one of Claim 1 to 12.   A computer-readable recording medium on which the program according to claim 14 is recorded.

Images