Abnormal behavior detection system, abnormal behavior detection method, and abnormal behavior detection program

Abstract

To prevent a normal behavior in a video from being erroneously detected as an abnormal behavior without requiring complexity.SOLUTION: An abnormal behavior detection system detects an abnormal behavior from behaviors of a person in a video, and comprises: a storage unit which stores an exclusion behavior filter in which an exclusion behavior to be excluded from detection of the abnormal behavior is registered; and an exclusion behavior determination unit which determines whether or not a behavior of the person detected as the abnormal behavior corresponds to the exclusion behavior registered in the exclusion behavior filter, and corrects a detection result of the behavior of the person from the abnormal behavior to a normal behavior when the behavior of the person detected as the abnormal behavior corresponds to the exclusion behavior.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to an abnormal behavior detection system, an abnormal behavior detection method, and an abnormal behavior detection program. [Background technology]

[0002] BACKGROUND ART Conventionally, there has been known a system that records the behavior of a person captured in an image taken by a camera, and issues an abnormality alert when the person behaves abnormally different from normal behavior (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6647489 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the above-mentioned conventional technology detects abnormal behavior based on differences from normal behavior patterns, which means that normal behavior that was not anticipated in advance may differ from normal behavior patterns and be mistakenly detected as abnormal behavior.

[0005] To avoid this problem, it is desirable to be able to specify and exclude falsely detected normal actions. However, because 2D posture detection-based action detection only has position information for each part of a person's body, and the amount of information is small, it is difficult to recognize the same normal action in different orientations as the same excluded action by simply defining one excluded action, and this creates the cumbersome task of defining each one as a separate normal action.

[0006] The present invention has been made in consideration of the above problems, and aims to provide a mechanism that prevents normal behavior in a video from being erroneously detected as abnormal behavior without requiring any complication. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems, one aspect of the present invention is an abnormal behavior detection system that detects abnormal behavior from the behavior of a person in a video, and is characterized by having a memory unit that stores an exclusion behavior filter in which exclusion behaviors to be excluded from the detection of abnormal behavior are registered, and an exclusion behavior determination unit that determines whether the behavior of a person detected as abnormal behavior corresponds to the exclusion behavior registered in the exclusion behavior filter, and if so, corrects the detection result of the person's behavior from abnormal behavior to normal behavior. [Effects of the Invention]

[0008] According to the present invention, for example, it is possible to prevent normal behavior in a video from being erroneously detected as abnormal behavior without requiring any complication. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a diagram illustrating an overview of an abnormal behavior detection system according to an embodiment. [Figure 2] FIG. 1 is a block diagram showing an example of the configuration of an abnormal behavior detection system according to an embodiment. [Figure 3] FIG. 2 is a diagram showing an example of the data structure of a detection result storage table of the abnormal behavior detection system according to the embodiment. [Figure 4] FIG. 2 is a diagram showing an example of the data structure of an exclusion behavior filter storage table of the abnormal behavior detection system according to the embodiment. [Figure 5] 10 is a flowchart illustrating an example of abnormal behavior detection processing in the abnormal behavior detection system according to the embodiment. [Figure 6] 10 is a flowchart illustrating an example of an exclusion behavior registration process of the abnormal behavior detection system according to the embodiment. [Figure 7] FIG. 2 is a diagram showing an example of a UI (User Interface) of an exclusion behavior filter registration unit of the abnormal behavior detection system according to the embodiment. [Figure 8] FIG. 2 is a diagram showing an example of the hardware configuration of a computer that realizes the abnormal behavior detection system according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The embodiments are merely examples for explaining the present invention and do not limit the present invention. For clarity of explanation, appropriate omissions and simplifications have been made. The present invention can be implemented in various other forms or in forms that combine some or all of each form. Unless otherwise specified, each component may be singular or plural.

[0011] When there are multiple components with the same or similar functions, they may be described using the same reference numeral with different subscripts. When there is no need to distinguish between these multiple components, the subscripts may be omitted.

[0012] In the description of the following embodiments, differences from the previously described embodiments will be mainly described, and descriptions of overlapping portions will be omitted as appropriate. Furthermore, information expressed in table format in this specification is not limited to table format and may be expressed in other data formats.

[0013] (Overview of Abnormal Behavior Detection System 2) First, an overview of an abnormal behavior detection system 2 according to this embodiment will be described with reference to Fig. 1. Fig. 1 is an explanatory diagram of an overview of the abnormal behavior detection system 2 according to this embodiment. As shown in Fig. 1, the abnormal behavior detection system 2 is designed to detect a person exhibiting suspicious behavior from video captured by a camera 10. The same behavior may or may not be treated as abnormal behavior depending on the location, and it may be desired to have different detection results.

[0014] For example, even if the system uniformly determines that the act of "running" is abnormal behavior, there may be cases where it is desired to treat running outdoors as not abnormal behavior and running indoors as abnormal behavior. Therefore, the abnormal behavior detection system 2 corrects the detection result of the behavior of person 110 from abnormal behavior to normal behavior, so that person 110 running outdoors in the video is determined to be a person not behaving abnormally as shown in Figure 1(a), and person 111 running indoors in the video is determined to be a person behaving abnormally as shown in Figure 1(b).

[0015] (Configuration of Abnormal Behavior Detection System 2) Next, the configuration of an abnormal behavior detection system 2 according to an embodiment will be described with reference to Fig. 2. Fig. 2 is a block diagram showing an example configuration of the abnormal behavior detection system 2 according to an embodiment. The abnormal behavior detection system 2 includes a storage unit 20, a calculation unit 21, and an exclusion behavior filter registration unit 22.

[0016] The storage unit 20 is a storage device that stores a detection model 201, a detection result storage table 202, and an exclusion behavior filter storage table 203.

[0017] The detection model 201 includes a human body part position detection model 2011 used by the behavior detection unit 211 in a "human body part position detection process" described later, a feature amount abnormal value estimation model 2012 used in a "feature amount abnormal value estimation process" described later, and an abnormal part determination model 2013 used by the abnormal part determination unit 212 in an "abnormal part determination process" described later. These models are designed using machine learning models such as neural networks and are trained in advance.

[0018] (Detection result storage table 202) Next, the detection result storage table 202 according to this embodiment will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of the data structure of the detection result storage table 202 of the abnormal behavior detection system 2 according to this embodiment. The detection result storage table 202 includes a behavior ID 301, a behavior feature vector 302, an abnormal value 303, and an abnormal part pattern 304.

[0019] The behavior ID 301 is identification information for each behavior, and a positive integer value is stored. The behavior feature vector 302 indicates the feature amount of the target person's behavior in consecutive frames, and stores a bd-dimensional real vector obtained by concatenating d-dimensional (d is an integer of 2 or more) real vectors (head 3021, left arm 3022, ...) of each body part (in this embodiment, the number of body parts is b=9: head, left arm, right arm, left hand, right hand, left elbow, right elbow, left foot, and right foot) extracted from the displacement of the positions of b (b is an integer of 1 or more) body parts. The behavior feature vector 302 is a concatenated behavior feature vector obtained by concatenating the behavior feature vectors of multiple body parts of the person.

[0020] The abnormal value 303 is a real value indicating the degree of abnormality. The abnormal value 303 is the output of the feature abnormal value estimation model 2012, which receives the behavior feature vector 302 as input. When the abnormal value 303 is determined to be equal to or greater than the threshold τ (abnormal behavior), the abnormal body part pattern 304 is a binary integer value indicating whether the movement of each body part is abnormal or not, with 1 indicating abnormal if the abnormal value of each body part is equal to or greater than the threshold τ and 0 indicating normal if the abnormal value is less than the threshold τ. When the abnormal value 303 is less than the threshold τ and is determined to be normal behavior, a null value is stored in the abnormal body part pattern 304.

[0021] (Exclusion Action Filter Saved Table 203) Next, the exclusion behavior filter save table 203 according to this embodiment will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of the data structure of the exclusion behavior filter save table 203 of the abnormal behavior detection system according to this embodiment. The exclusion behavior filter save table 203 includes an abnormal part pattern 401, a person direction 402, an exclusion behavior type ID 403, and a behavior feature vector (behavior ID) 404. An exclusion behavior refers to a behavior that is not subject to detection of abnormal behavior.

[0022] The abnormal part pattern 401 is expressed as a vector in which binary expressions (for example, 0 and 1) indicating the presence or absence of an abnormality in each part are arranged for all parts. The abnormal part pattern 401 is a vectorized version of the pattern of abnormal parts of the behavior to be excluded, and is similar to the abnormal part pattern 304.

[0023] The person direction 402 stores the angle value of the direction in which the person is facing. This angle is estimated by the person direction estimation unit 213, and in this embodiment, is a discrete value at 45-degree intervals of 0 degrees, 45 degrees, and 90 degrees. The exclusion behavior type ID 403 stores a filter ID defined for each type of exclusion behavior.

[0024] The behavior feature vector (behavior ID) 404 stores an arbitrary number of behavior IDs 301 (FIG. 3) corresponding to representative behavior feature vectors 302 (FIG. 3) corresponding to the abnormal part pattern 401 and person orientation 402 stored in the detection result storage table 202. When the behavior feature vector of a certain person matches (or is similar to) the behavior feature vector (behavior ID) 404 corresponding to the abnormal part pattern 401 stored in the exclusion behavior filter storage table 203 and the person orientation matches the person orientation 402 stored in the exclusion behavior filter storage table 203, the abnormal value determined as an abnormality detection is changed to a normal value.

[0025] In the exclusion behavior filter storage table 203 in Figure 4, the first and second rows are assigned the same exclusion behavior type ID even though the abnormal part pattern 401 and person orientation 402 are different. This is because the abnormal parts of the left arm and the right arm can be considered to be the same abnormal part when the person orientation is rotated by 90 degrees.

[0026] The calculation unit 21 includes a processor and a memory, and has a behavior detection unit 211 , an abnormal portion determination unit 212 , a person direction estimation unit 213 , and an exclusion behavior determination unit 214 .

[0027] The behavior detection unit 211 performs the following two types of processing. One is a processing (personal body part position detection processing) that uses a person body part position detection model 2011 to detect the position of a person and each body part (personal body part position information) from the latest frame captured by the camera 10. The other is a processing (feature quantity abnormal value estimation processing) that extracts a behavior feature vector from the person body part position information of a specific person in consecutive frames of the video captured by the camera 10, and estimates an abnormal value from the behavior feature vector using a feature quantity abnormal value estimation model 2012. The detected behavior feature vector and the estimated abnormal value are stored in the detection result storage table 202.

[0028] When a person's behavior is determined to be abnormal, the abnormal part determination unit 212 identifies which part of the body is causing the abnormal movement using the abnormal part determination model 2013 (abnormal part determination process). The combination of whether each part is abnormal or not is called an abnormal part pattern.

[0029] The person direction estimation unit 213 estimates the direction in which the person is facing from the positions of the person's body parts detected by the behavior detection unit 211. The direction in which the person is facing can be determined, for example, from the direction of a perpendicular line to a line connecting the positions of both shoulders detected by the behavior detection unit 211. The person direction is a discrete value, and is one of values at 45-degree intervals, such as 0 degrees, 45 degrees, or 90 degrees. Each person is assigned the value that is closest to the angle of the perpendicular line to the line connecting both shoulders as the person direction.

[0030] The exclusion behavior determination unit 214 determines whether the input behavior is a behavior that is not subject to anomaly detection. That is, when the behavior detection unit 211 determines that a certain behavior is abnormal, the exclusion behavior determination unit 214 determines whether the input behavior corresponds to any of the exclusion behaviors registered in the exclusion behavior filter storage table 203 that corresponds to the abnormal part identified by the abnormal part determination unit 212 and the direction estimated by the person direction estimation unit 213, as will be described later, and changes the abnormal value to a normal value.

[0031] The exclusion behavior determination unit 214 stores a distance threshold θ of the abnormal body part pattern used when determining exclusion behavior and changing to a normal value, a distance threshold η of the behavior feature vector, and a normal value m to change from the abnormal value when determined to be an exclusion target. Hamming distance or the like is used as the distance of the abnormal body part pattern. Euclidean distance, cosine similarity or the like is used as the distance of the behavior feature vector.

[0032] The exclusion behavior filter registration unit 22 registers the behavior feature vector (behavior ID) of the exclusion behavior in the exclusion behavior filter storage table 203. The exclusion behavior filter registration unit 22 includes a display device (not shown) that outputs a UI (User Interface).

[0033] (Abnormal behavior detection processing) Next, the abnormal behavior detection process according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of the abnormal behavior detection process of the abnormal behavior detection system 2 according to this embodiment. First, when a frame image is acquired from the camera 10 (S501), the behavior detection unit 211 performs a single-frame analysis of the frame image (S502). The single-frame analysis involves identifying the position of a person in the latest frame, identifying the positions of body parts of the person, and associating the positions of the same person between consecutive frames.

[0034] If a certain person P has been detected for n or more consecutive frames (n is a predetermined positive integer) from the latest frame (YES in S503), the behavior detection unit 211 performs abnormal behavior detection processing on the person P (S504). In the abnormal behavior detection processing, the behavior detection unit 211 detects person body part position information of the person P in consecutive frames by person body part position detection processing, extracts a behavior feature vector F of the person P by feature amount abnormal value estimation processing based on the person body part position information, and estimates an abnormal value V based on the behavior feature vector F. The behavior feature vector F is a concatenated vector obtained by concatenating vectors indicating the movements of multiple body parts of the person P.

[0035] Next, if the abnormal value V estimated in S504 is equal to or greater than the threshold τ (S505 YES), the abnormal region determination unit 212 performs abnormal region determination processing for the person P (S506), and if it is less than the threshold τ (S505 NO), the process proceeds to S510. In S506, the abnormal region determination unit 212 decomposes the behavior feature vector F extracted in S504 into vectors for each region, and calculates abnormal values for each region through region-by-region feature abnormal value estimation processing. An abnormal region pattern A, which is a binary combination in which 1 is assigned to regions where the abnormal value is equal to or greater than the threshold τ and 0 is assigned to regions where the abnormal value is less than the threshold τ, is stored in the detection result storage table 202 in S510, which will be described later.

[0036] Next, the person direction estimation unit 213 estimates the direction D of the person P whose abnormal value V is determined to be equal to or greater than the threshold value τ in S505 (S507). Next, the exclusion behavior determination unit 214 refers to the exclusion behavior filter storage table 203, and narrows down the candidates to the behavior feature vectors (behavior IDs) 404 of records that have an abnormal body part pattern 401 whose distance from the abnormal body part pattern A of the person P is less than the threshold value θ and that have a person direction 402 that matches the person direction D. By setting it to "the distance between the abnormal body part pattern A and the abnormal body part pattern 401 is less than the threshold value θ," it is possible to take a margin that allows for a certain degree of similarity rather than an exact match.

[0037] If the distance between the behavior feature vector F and any of the behavior feature vectors 302 ( FIG. 3 ) corresponding to the narrowed-down behavior feature vector (behavior ID) 404 is equal to or less than the threshold η (YES in S508), the exclusion behavior determination unit 214 determines that the behavior of person P is a behavior to be excluded from abnormality determination, and corrects the abnormal value stored in the abnormal value 303 field to a normal value m (m is a predetermined positive number) (S509). By setting "the distance between the behavior feature vector F and any of the behavior feature vectors 302 is equal to or less than the threshold η," it is possible to provide a margin that allows for a certain degree of similarity rather than an exact match. By correcting the value in the abnormal value 303 field from an abnormal value to a normal value, the detection result of the corresponding behavior is corrected from abnormal behavior to normal behavior.

[0038] On the other hand, if the distance between the behavior feature vector F and any of the behavior feature vectors 302 (FIG. 3) corresponding to the extracted behavior feature vector (behavior ID) 404 is greater than the threshold value η (S508 NO), the exclusion behavior determination unit 214 proceeds to S510.

[0039] Next, the calculation unit 21 stores the behavior feature vector F, the abnormal value V (or the corrected normal value m), and the abnormal part pattern A calculated in this abnormal behavior detection process in the detection result storage table 202 (S510).

[0040] In the explanation of this abnormal behavior detection process, we have described a case where excluded behaviors that are not subject to abnormality detection are defined for each abnormal part pattern and person orientation, but there may be cases where there is no person orientation. In that case, the process of S507 is skipped, and the matching of person orientation is not taken into account in S508.

[0041] (Exclusion behavior registration process) The above describes an example in which a person's behavior detected as abnormal behavior is corrected to normal behavior based on the exclusion behavior filter storage table 203. The following describes how to create records in the exclusion behavior filter storage table 203. Figure 6 is a flowchart showing an example of the exclusion behavior registration process of the abnormal behavior detection system 2 according to the embodiment.

[0042] First, the exclusion behavior filter registration unit 22 accepts input of abnormal behavior to be excluded on the UI 7 (FIG. 7) displayed on a display device (not shown) (S601). Next, the exclusion behavior filter registration unit 22 specifies the abnormal part pattern and person orientation of the abnormal behavior to be excluded as search conditions, and searches for the corresponding behavior from the exclusion behavior filter storage table 203 (S602).

[0043] Next, the exclusion behavior filter registration unit 22 determines whether or not an exclusion behavior that meets the search criteria specified in S602 exists in the exclusion behavior filter save table 203. If an exclusion behavior that meets the search criteria specified in S602 exists in the exclusion behavior filter save table 203 (S603 YES), the exclusion behavior filter registration unit 22 displays candidates of the same behavior narrowed down by the search criteria on the display device, assigns the same exclusion behavior type ID as the behavior selected by the user from these candidates to the abnormal behavior to be excluded that was input in S601 (S604), and proceeds to S606.

[0044] On the other hand, if the exclusion behavior filter storage table 203 does not contain any exclusion behavior that matches the search criteria specified in S602 (S603 NO), the exclusion behavior filter registration unit 22 assigns a new exclusion behavior type ID to the abnormal behavior to be excluded that was input in S601 (S605), and moves the process to S606. In S606, the abnormal part pattern 401, person direction 402, exclusion behavior type ID 403, and behavior feature vector (behavior ID) 404 of the abnormal behavior to be excluded are stored in the exclusion behavior filter storage table 203.

[0045] (UI of the exclusion behavior filter registration unit 22 of the abnormal behavior detection system 2) Next, the UI7 of the exclusion behavior filter registration unit 22 will be described with reference to Fig. 7. Fig. 7 is a diagram showing an example of the UI7 of the exclusion behavior filter registration unit 22 of the abnormal behavior detection system 2 according to the embodiment. The UI7 is displayed on a display device (not shown), and is a screen configured to include a selection frame 701, search conditions 702, a search button 703, a search result window 704, and a corresponding behavior selection button 705.

[0046] The selection frame 701 displays (in a frame in the example of FIG. 7) "behaviors to be excluded" selected from abnormal behaviors among those determined to be abnormal behaviors based on abnormal values estimated by the behavior detection unit 211. The selection frame 701 displays the behaviors determined to be abnormal behaviors for each frame of the video so that they can be selected by moving the frames forward and backward ("forward" and "backward"), and accepts the user's selection of an action. The above corresponds to S601 (FIG. 6).

[0047] The search criteria 702 include two search criteria: “Abnormal Body Part Pattern” and “Different Directions Only.” The “Abnormal Body Part Pattern” search criteria are set to the abnormal body parts of the behavior selected in the selection frame 701. In the framed area in FIG. 7, the abnormal body parts of the person are identified by white circles: “Left Hand,” “Right Hand,” “Left Elbow,” and “Right Elbow.” Correspondingly, “Left Hand,” “Right Hand,” “Left Elbow,” and “Right Elbow” are set in the “Abnormal Body Part Pattern” field. The search criteria for the abnormal body part pattern can be modified by manually entering information in the “Abnormal Body Part Pattern” field of the search criteria 702. The “Different Directions Only” search criteria allows you to set whether or not to exclude people with the same orientation. Checking the “Different Directions Only” search criteria to narrow down the search to people with different orientations can prevent duplicate exclusion behaviors of people with the same orientation. The search button 703 is pressed when searching the exclusion behavior filter storage table 203 using the search criteria set in the search criteria 702. Note that only the “Abnormal Body Part Pattern” may be specified as a search condition. The above corresponds to S602 (FIG. 6).

[0048] The search results window 704 displays the search results. The search results are videos of abnormal behavior that match the search criteria and have been acquired as corresponding videos from a video storage unit (not shown). If the search results window 704 contains a behavior that the user determines is the same as the behavior selected in the selection frame 701, the user selects it. The rectangular area marked "Select" in the search results window 704 is the behavior that the user determines is the same as the behavior selected in the selection frame 701. If not, no selection is made. This corresponds to S603 (FIG. 6).

[0049] The relevant behavior selection button 705 includes two types of buttons, "Yes" and "No." When "Yes" is selected, the exclusion behavior filter registration unit 22 assigns the same exclusion behavior type ID 403 as the behavior selected in the search result window 704, and registers the abnormal part pattern, person direction, and behavior feature vector (behavior ID) in the exclusion behavior filter storage table 203. The above corresponds to S604 and S606 (FIG. 6).

[0050] When "None" is selected, the exclusion behavior filter registration unit 22 assigns a new exclusion behavior type ID 403 to the behavior selected in the selection frame 701, and registers the abnormal part pattern, person direction, and behavior feature vector (behavior ID) in the exclusion behavior filter storage table 203. The above corresponds to S605 and S606 (FIG. 6).

[0051] In this embodiment, for abnormal behavior that is difficult to exclude from detection, candidates are narrowed down by focusing on commonalities in the abnormal part patterns 401 and person orientations 402 stored in the exclusion behavior filter storage table 203, and then a determination is made as to whether or not the behavior should be excluded from detection based on the similarity of the feature amount (behavior feature vector (behavior ID) 404). Thus, in an abnormal behavior detection system that detects suspicious people captured in time-series images, the amount of information defined by the exclusion behavior filter is reduced while improving detection accuracy. Furthermore, even when manually associating identical abnormal behaviors that differ only in orientation, narrowing down candidates based on commonalities in the abnormal part patterns 401 can reduce the workload of linking identical abnormal behaviors in different orientations.

[0052] (Hardware of the computer 500 that realizes the abnormal behavior detection system 2) 8 is a diagram showing an example of the hardware configuration of a computer 500 that realizes the abnormal behavior detection system 2 according to the embodiment. In the computer 500, a processor 510 such as a CPU (Central Processing Unit), a memory 520 such as a RAM (Random Access Memory), a storage 530 such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive), a network I / F (Interface) 540, an input / output device 550 (e.g., a keyboard, a mouse, a touch panel, a display, etc.), and a peripheral device 560 are connected via a bus.

[0053] In computer 500, a program for realizing abnormal behavior detection system 2 is read from storage 530 and executed by processor 510 and memory 520 in cooperation with each other, thereby realizing each system. Alternatively, the program for realizing abnormal behavior detection system 2 may be obtained from an external computer equipped with a non-transitory storage device by communication via network I / F 540. Alternatively, the program for realizing abnormal behavior detection system 2 may be recorded on a non-transitory recording medium and read and acquired by a medium reading device.

[0054] The present invention is not limited to the above-described embodiments and includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and are not necessarily limited to those including all of the described configurations. Furthermore, as long as there is no contradiction, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of one embodiment to the configuration of another embodiment. Furthermore, it is possible to add, delete, replace, integrate, or distribute part of the configuration of each embodiment. Furthermore, the configurations and processes described in the embodiments can be distributed, integrated, or replaced as appropriate based on processing efficiency or implementation efficiency. [Explanation of symbols]

[0055] 2: Abnormal behavior detection system, 20: Memory unit, 22: Exclusion behavior filter registration unit, 203: Exclusion behavior filter storage table, 211: Behavior detection unit, 212: Abnormal part determination unit, 213: Person direction estimation unit, 214: Exclusion behavior determination unit

Claims

1. An abnormal behavior detection system that detects abnormal behavior from the behavior of a person in a video, comprising: A vector indicating the movement of a body part of a person in the action of the person is defined as an action feature vector, a storage unit for storing an exclusion behavior filter in which the behavior feature vector representing an exclusion behavior to be excluded from abnormal behavior detection is registered; an exclusion behavior determination unit that determines whether or not the behavior of a person whose abnormal behavior has been detected corresponds to the exclusion behavior based on a distance between the behavior feature vector of the person and the behavior feature vector representing the exclusion behavior registered in the exclusion behavior filter, and corrects the detection result of the behavior of the person from the abnormal behavior to a normal behavior if the behavior corresponds to the exclusion behavior; An abnormal behavior detection system comprising:

2. An abnormal behavior detection system that detects abnormal behavior from the behavior of a person in a video, comprising: A vector indicating the movement of a body part of a person in the action of the person is defined as an action feature vector, a vector obtained by concatenating the action feature vectors for each of a plurality of body parts of the person in the action of the person is defined as a concatenated action feature vector; a storage unit for storing an exclusion behavior filter in which the linked behavior feature vector representing an exclusion behavior to be excluded from abnormal behavior detection is registered; an exclusion behavior determination unit that determines whether or not the behavior of the person in which abnormal behavior has been detected corresponds to the exclusion behavior based on the distance between the linked behavior feature vector of the behavior of the person in which abnormal behavior has been detected and the linked behavior feature vector representing the exclusion behavior registered in the exclusion behavior filter, and if so, corrects the detection result of the behavior of the person from abnormal behavior to normal behavior; An abnormal behavior detection system comprising:

3. The abnormal behavior detection system according to claim 2, The excluded behavior is stored in association with an abnormal part pattern that combines whether or not abnormal behavior is present for each of a plurality of body parts of the person, The exclusion behavior determination unit Narrowing down the excluded behaviors registered in the excluded behavior filter based on a distance between a combination pattern of whether or not abnormal behaviors for multiple body parts of a person in the behavior of the person in which abnormal behavior has been detected and the abnormal part pattern; Whether or not the behavior of the person in which the abnormal behavior has been detected corresponds to the exclusion behavior is determined based on the distance between the linked behavior feature vector of the behavior of the person in which the abnormal behavior has been detected and the linked behavior feature vector of any of the narrowed-down exclusion behaviors. An abnormal behavior detection system characterized by:

4. The abnormal behavior detection system according to claim 3, The exclusion behavior is further stored in association with a person direction indicating the direction of the person; The exclusion behavior determination unit narrowing down the excluded behaviors registered in the excluded behavior filter based on the direction of the person in the behavior of the person in which the abnormal behavior was detected and the distance between the abnormal part pattern and a combination pattern of whether or not the abnormal behavior is related to each of a plurality of parts of the person's body in the behavior of the person; Whether or not the behavior of the person in which the abnormal behavior has been detected corresponds to the exclusion behavior is determined based on the distance between the linked behavior feature vector of the behavior of the person in which the abnormal behavior has been detected and the linked behavior feature vector of any of the narrowed-down exclusion behaviors. An abnormal behavior detection system characterized by:

5. The abnormal behavior detection system according to claim 3, When the distance between the combination pattern and the abnormal portion pattern is equal to or less than a first predetermined value, and the distance between the linked behavior feature vector in the behavior of the person in which the abnormal behavior is detected and the linked behavior feature vector of any of the narrowed-down excluded behaviors is equal to or less than a second predetermined value, the behavior of the person is determined to be the excluded behavior. An abnormal behavior detection system characterized by:

6. The abnormal behavior detection system according to claim 1 or 2, an exclusion behavior filter registration unit that registers the exclusion behavior in the exclusion behavior filter; The excluded behavior includes an abnormal body part pattern that combines whether or not abnormal behaviors for multiple body parts of the person are true, a person orientation that indicates the orientation of the person's body, and a behavior feature vector that indicates the movement of the body parts of the person, The exclusion behavior filter registration unit When registering the exclusion behavior based on the behavior of a person in the video, the exclusion behavior registered in the exclusion behavior filter is narrowed down based on the abnormal part pattern or based on the abnormal part pattern and the person's orientation, and candidates of the same behavior as the behavior of the person are output on a screen, and the same behavior selected by the user from the candidates is registered in the exclusion behavior filter. An abnormal behavior detection system characterized by:

7. An abnormal behavior detection method executed by an abnormal behavior detection system that detects abnormal behavior from the behavior of a person in a video, comprising: A vector indicating the movement of a body part of a person in the action of the person is defined as an action feature vector, The abnormal behavior detection system includes: a storage unit for storing an exclusion behavior filter in which the behavior feature vector representing an exclusion behavior to be excluded from abnormal behavior detection is registered; determining whether or not the behavior of a person detected as abnormal behavior corresponds to the exclusion behavior based on a distance between the behavior feature vector of the person and the behavior feature vector representing the exclusion behavior registered in the exclusion behavior filter; If applicable, correct the detected behavior of the person from abnormal behavior to normal behavior. An abnormal behavior detection method characterized by including each process.

8. An abnormal behavior detection method executed by an abnormal behavior detection system that detects abnormal behavior from the behavior of a person in a video, comprising: A vector indicating the movement of a body part of a person in the action of the person is defined as an action feature vector, a vector obtained by concatenating the action feature vectors for each of a plurality of body parts of the person in the action of the person is defined as a concatenated action feature vector; the abnormal behavior detection system includes a storage unit that stores an exclusion behavior filter in which the linked behavior feature vector representing an exclusion behavior to be excluded from detection of abnormal behavior is registered; The abnormal behavior detection system, determining whether or not the behavior of the person in which abnormal behavior has been detected corresponds to the exclusion behavior based on a distance between the linked behavior feature vector of the behavior of the person in which abnormal behavior has been detected and the linked behavior feature vector representing the exclusion behavior registered in the exclusion behavior filter; If applicable, correct the detected behavior of the person from abnormal behavior to normal behavior. An abnormal behavior detection system characterized by having each process.

9. An abnormal behavior detection program for causing a computer to function as the abnormal behavior detection system according to any one of claims 1 to 6.

Images