JP2012128877A - Suspicious behavior detection system and method - Google Patents

Abstract

An object of the present invention is to provide a suspicious behavior detection system that can identify and identify a suspicious person who is performing abnormal behavior.
In a suspicious behavior detection system 1 that detects a suspicious behavior of a monitoring target using a video of a stereo camera 10, a moving trajectory acquisition unit 21 that acquires the moving trajectory information of the monitoring target, and the movement trajectory information. The behavior identifying unit 22 that identifies the behavior state of the monitoring target and automatically determines the suspicious behavior of the monitoring target.
[Selection] Figure 1

Description

  Embodiments described herein relate generally to a suspicious behavior detection system that uses an optical sensor such as a camera.

  In recent years, a monitoring system for monitoring a suspicious person or the like using a video (moving image) taken by a video camera has been developed. Various systems have been proposed for the monitoring system. For example, a system using a feature value based on three-dimensional higher-order local autocorrelation has been proposed (see, for example, Patent Document 1).

JP 2006-79272 A

  As described above, the monitoring system can detect an abnormal operation in the screen shot by the video camera, but identifies and identifies a suspicious person who is performing an abnormal action from a plurality of people. It is not possible.

  Therefore, an object of the present invention is to provide a suspicious behavior detection system that can identify and identify a suspicious person who is performing an abnormal behavior.

  A suspicious behavior detection system according to an aspect of the present invention includes a sensor unit for detecting a movement position of a monitoring target, and a movement track acquisition unit that acquires movement track information of the monitoring target based on an output of the sensor unit. Using the learning information acquired by learning the behavior according to the movement trajectory, and identifying the behavior state of the monitoring target based on the movement trajectory information acquired by the movement trajectory acquisition means, and the behavior And a determination unit that automatically determines in real time the suspicious behavior to be monitored based on the behavior state identified by the identification unit.

The block diagram which shows the principal part of the suspicious action detection system regarding embodiment. The figure for demonstrating the specific structure of the system regarding embodiment. The block diagram for demonstrating the specific structure of the movement trace integration part and action identification part regarding embodiment. The figure for demonstrating the learning method in the action identification part regarding embodiment. The figure for demonstrating the learning method in the action identification part regarding embodiment. The figure for demonstrating the designation method of the walk locus | trajectory in the action identification part regarding embodiment. The flowchart for demonstrating the process sequence of the suspicious action detection system regarding embodiment.

  Embodiments will be described below with reference to the drawings.

  Embodiments of the present invention will be described below with reference to the drawings.

(Basic system configuration)
FIG. 1 is a block diagram illustrating a main part of the suspicious behavior detection system according to the present embodiment.

  As shown in FIG. 1, the system 1 includes a plurality of stereo cameras 10 and a suspicious behavior detection device 20. The stereo camera 10 operates as a sensor for detecting the movement position of a person who is a monitoring target. The stereo camera 10 has a configuration in which a plurality of cameras arranged at different viewpoints including right and left or top and bottom are combined, and the captured video is transmitted to the suspicious behavior detection device 20. The plurality of cameras may be two cameras arranged at distant positions.

  In addition to the stereo camera 10, an optical system sensor such as an infrared sensor 11 or a laser sensor 12 may be used as the sensor.

  The suspicious behavior detection device 20 is configured by a computer system, and includes functional elements of a movement trajectory acquisition unit 21 and a behavior identification unit 22 realized by a CPU and software. The movement trajectory acquisition unit 21 has a function of performing image processing (stereo image processing) on an image transmitted from the stereo camera 10, and generates movement trajectory information indicating a movement trajectory of a person to be monitored from the image processing result. . Here, as will be described later, the movement trajectory of the person corresponds to a walking trajectory when the person walks and moves.

  The movement trajectory acquisition unit 21 generates movement trajectory information obtained by integrating the movement trajectories corresponding to the shooting range (monitoring area) of the stereo camera 10 based on the images transmitted from the plurality of stereo cameras 10. As will be described later, the integrated movement trajectory information includes information indicating a movement trajectory obtained by connecting (connecting) each monitoring area and the non-monitoring area.

  The action identification unit 22 stores learning information acquired by learning according to the movement locus in advance, and uses this learning information to monitor the object based on the movement locus information sent from the movement locus acquisition unit 21. Determine the suspicious behavior of people.

(Specific system configuration and effects)
FIG. 2 is a diagram for explaining a specific example to which the system of this embodiment is applied.

  Here, it is assumed that the suspicious behavior detection system 1 of the present embodiment is applied to a monitoring system that monitors a passage in a building. In this system, as shown in FIG. 2, monitoring areas 200, 210, 220, and 230 for monitoring the passage by, for example, four stereo cameras 10-1 to 10-4 are set.

  The passage is divided into area A and area B, and there is a non-monitoring area 240 that connects the areas. The handling of the non-monitoring area 240 will be described later. As described above, it is possible to use the infrared sensor 11 or the laser sensor 12 as a sensor in addition to the stereo camera 10, and monitor the same areas A and B with a plurality of types of sensors. Variations are possible. However, in the present embodiment, a case of a method of monitoring with four stereo cameras 10-1 to 10-4 will be described.

  FIG. 3 is a block diagram for explaining specific functional elements of the movement trajectory acquisition unit 21 and the behavior identification unit 22 included in the suspicious behavior detection device 20.

  The movement trajectory acquisition unit 21 performs a stereo image process on the images from the stereo cameras 10-1 to 10-4, and acquires a plurality of walking trajectory acquisition units for acquiring walking trajectory information indicating the walking trajectory of the person to be monitored. 30. Furthermore, the movement trajectory acquisition unit 21 integrates the walking trajectory information acquired by the plurality of walking trajectory acquisition units 30 or performs a walking trajectory integration that performs a complementary process of complementing the walking trajectory of the non-monitoring area from the preceding and following walking trajectories. Part 31. The walking trajectory integration unit 31 integrates walking trajectory information from a plurality of monitoring areas and integration of walking trajectory information acquired by a plurality of types of sensors (for example, a stereo camera and an infrared sensor) in the same monitoring area. Both of these are executed.

  The action identification unit 22 includes a plurality of classifiers, and includes an action integration unit 45 that outputs an identification result (determination result) obtained by combining them as a final output result. The behavior integration unit 45 performs, for example, a method such as majority voting or logical product or a method of executing a discrimination process by a rule as a pre-process and executing a discrimination process by a learning machine when the result is excessive or insufficient. Output the result (judgment result).

  Specifically, the behavior identification unit 22 is constructed by applying a pattern recognition method such as SVM (Support Vector Machine), for example, and mathematically characterizes the characteristics of the walking trajectory information (movement trajectory data) to be monitored. analyse. Thereby, the suspicious behavior is discriminated by teaching the normal pattern and the abnormal pattern in the human system by learning.

  As the plurality of discriminators, a gender discriminator 40, an age discriminator 41, a normal / abnormal discriminator 42, a stay / run discriminator 43, a scale discriminator 44, and the like are provided. Each discriminator stores learning information acquired by learning according to a walking (moving) trajectory in advance, and executes identification processing using the learning information.

  Specifically, for example, the age discriminator 41 stores age information included in human attribute information and information indicating a walking trajectory as if walking steadily as learning information, and walking from the movement trajectory acquisition unit 21. Based on the trajectory information, if the age of walking is an elderly person, it is identified as an elderly person. Moreover, if the age is a child, the age discriminator 41 identifies it as a lost child. Here, the learning information includes, for example, information such as height, walking speed and pace by age.

  Further, for example, the stay / running identifier 43 stores, as learning information, what kind of walking locus can be said to be staying and what kind of walking locus can be said to be running based on the walking locus of a normal pedestrian. Yes. Further, the normal / abnormal identifier 42 is information indicating a walking locus (straight, turning, etc.) determined to be normal based on a walking locus of a pedestrian in a passage, for example, at a position such as in front of a door, Information indicating a walking trajectory determined to be abnormal due to unstable walking (the direction is indefinite, a stop for a certain period of time, or the like) is stored as learning information.

  Note that the behavior integration unit 45 may execute a sensitivity / insensitivity selection process for the identification results from each classifier. For example, the normal / abnormal discriminator 42 is set so that a sensitive process is performed at night and relatively normal and abnormal are discriminated, or an insensitive process is performed during the day. Set to relax the criteria for judging abnormalities.

  Hereinafter, the operation and effect of the system of the present embodiment will be described with reference to FIGS. FIG. 7 is a flowchart showing a processing procedure of the suspicious behavior detection system when applied to the passage shown in FIG.

  First, as shown in FIG. 2, the system inputs images taken by the stereo cameras 10-1 to 10-4 arranged in the passage (step S1). Each of the walking trajectory acquisition units 30 of the moving trajectory acquisition unit 21 acquires the walking trajectory information (moving trajectory information) in the corresponding monitoring areas 200, 210, 220, and 230 after executing the stereo image processing (step S2, S2). S3). The walking trajectory information is specifically information indicating various walking trajectories as shown in FIG.

  Here, the walking trajectory integration unit 31 integrates and outputs the walking trajectory information from the corresponding monitoring areas 200, 210, 220, and 230. Further, as will be described later, the walking trajectory integration unit 31 performs a complementing process for complementing the walking trajectory of the non-monitoring area 240 from the preceding and following walking trajectories in cooperation with the stereo cameras 10-1 to 10-4.

  Based on the walking trajectory information output from the movement trajectory acquisition unit 21, the behavior identification unit 22 identifies the behavior of the person 100 who is walking along the passage that is the monitoring target (step S4). Specifically, as described above, a process for identifying an action by each of the classifiers 40 to 44 is executed.

  Here, the normal / abnormal identifier 42 will be specifically described.

  Each discriminator 40-44 performs the process which identifies action using the learning information acquired by the learning process. Here, the learning methods are roughly classified into an unsupervised learning method as shown in FIG. 4 and a supervised learning method as shown in FIG. In unsupervised learning, as shown in FIGS. 4B and 4C, the walking trajectory is clustered into various classes to obtain learning information in which each walking class is labeled normal / abnormal.

  The normal / abnormal discriminator 42 uses the learning information based on the walking trajectory information from the walking trajectory integration unit 31 to identify the acquired walking trajectory as a walking class, and normal / abnormal according to the attached label. Identify anomalies. Specifically, the normal / abnormal discriminator 42 identifies that the walking trajectory in the monitoring region 200 shown in FIG. 2 is abnormal from the learning information shown in FIGS.

  In the supervised learning shown in FIGS. 5A and 5B, learning information in which normal / abnormal labels 50 and 51 are added to the walking trajectory in the human system is obtained. The normal / abnormal discriminator 42 uses the learning information based on the walking trajectory information from the walking trajectory integration unit 31 to determine whether the acquired walking trajectory corresponds to a normal walking trajectory or an abnormal walking trajectory. And it identifies that it is abnormal with respect to the walking locus in the monitoring area | region 200 shown in FIG.

  FIG. 6 is a diagram for explaining a method for specifying and selecting walking trajectory data used for learning. That is, each discriminator 40-44 searches corresponding trajectories 60-62 from the accumulated walking trajectory information by designating various conditions. For example, the designation of a place is designation of a person who has passed a certain area, a person who has moved from a certain place to a certain place, or the like. The designation of time is designation of a person who has passed on a specific day or a person who has passed at a specific time. The designation of the locus is the designation of the locus by drawing the locus on the screen (GUI). As a walking trajectory used for learning, abstract feature amounts such as continuous position coordinates, speed and number of direction changes, continuous images obtained from walking trajectories, and feature amounts obtainable from continuous images are used.

  That is, each discriminator 40 to 44 applies a so-called sequential learning method so that walking trajectory information (moving trajectory data) used in sequential learning can be arbitrarily selected from the stored walking trajectory information data group. Select regularly and automatically based on conditions (time zone, location, human attribute information). Alternatively, a method in which the operator designates or selects arbitrary walking trajectory information (movement trajectory data) from the terminal may be used.

  The behavior integrating unit 45 of the behavior identifying unit 22 integrates the identification results of the normal / abnormal classifier 42 and the identification results of other classifiers, and finally determines a person who is performing suspicious behavior (step) S5). Here, when the behavior integration unit 45 takes a walking trajectory different from the normal walking trajectory in the monitoring area 200 and is identified as abnormal by the normal / abnormal discriminator 42, the behavior of the corresponding person 110 is It determines with it being suspicious (YES of step S5).

  When the determination result that the system is suspicious is output from the behavior identifying unit 22, the system executes a process of reporting that there is a person 110 who performs suspicious behavior, for example (step S6).

  By the way, in the case of a wide-area passage, it may not be possible to determine whether or not each monitoring area is suspicious (NO in step S5). In such a case, the walking trajectory integration unit 31 of the system links the walking trajectories of the monitoring areas 200, 210, 220, and 230 in cooperation with the stereo cameras 10-1 to 10-4 as described above ( YES in step S7, S8). In addition, the system also performs complementary processing for complementing the walking trajectory from the preceding and following walking trajectories for the non-monitoring region 240, and outputs walking trajectory information obtained by connecting and integrating all the walking trajectories.

  Whether or not a person showing an abnormal walking trajectory is finally suspiciously based on the walking trajectory information obtained by connecting and integrating all the walking trajectories even in a wide passage. Can be determined.

  In the system of the present embodiment, when the behavior integration unit 45 of the behavior identification unit 22 finally determines a person who is suspicious, tracking / removal of the installed cameras 10-1 to 10-4 is performed. A configuration including an apparatus for controlling the zoom imaging function to close up the suspicious person on the screen may be employed.

  As described above, according to this embodiment, based on the walking trajectory (movement trajectory) of the person to be monitored, the behavior is identified by using the learning information, and finally the abnormal behavior is performed. Suspicious person can be determined. Therefore, if the system according to the present embodiment is applied to, for example, a monitoring system in a building, a suspicious person can be automatically specified, so that an effective monitoring function can be realized.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

1 ... Suspicious behavior detection system, 10, 10-1 to 10-4 ... Stereo camera,
11 ... Infrared sensor, 12 ... Laser sensor, 20 ... Suspicious behavior detection device,
21 ... movement locus acquisition unit, 22 ... action identification unit, 30 ... walking locus acquisition unit,
31 ... walking locus integration unit, 40 ... gender discriminator, 41 ... age discriminator,
42 ... Normal / abnormal identifier, 43 ... Standing / running identifier, 44 ... Owl identifier,
45 ... Action integration unit, 200, 210, 220, 230 ... Monitoring area,
240 ... non-monitoring area.

Claims (13)

Sensor means for detecting the movement position of the monitoring object;
Based on the output of the sensor means, movement trajectory acquisition means for acquiring the movement trajectory information of the monitoring target;
Using the learning information acquired by learning the behavior according to the movement trajectory, and the action identification means for identifying the monitoring target behavior state based on the movement trajectory information acquired by the movement trajectory acquisition means;
Determination means for determining the suspicious behavior of the monitoring target based on the behavior state identified by the behavior identification means,
The action identifying means includes
A suspicious behavior detection system comprising: identification means for identifying whether a movement locus of the monitoring target indicated by the movement locus information is normal or abnormal by using human attribute information as the learning information.   The suspicious behavior detection system according to claim 1, wherein the sensor means includes a stereo camera, a monocular camera, or another optical system sensor. The sensor means includes
A stereo camera for photographing the monitoring target;
The suspicious behavior detection system according to claim 1, further comprising an image processing unit that processes a video signal output from the stereo camera. The movement trajectory acquisition means includes
Sensor means for detecting the movement position of the monitoring target in a plurality of monitoring areas corresponding to the shooting range of each camera, having a plurality of cameras;
2. The suspicious device according to claim 1, further comprising: a generation unit that integrates outputs of the sensor unit to generate integrated movement trajectory information indicating the movement trajectory of the monitoring target across the plurality of monitoring regions. Behavior detection system.   5. The suspicious behavior detection system according to claim 4, wherein the sensor means includes a stereo camera, a monocular camera, or another optical system sensor as the plurality of cameras. The movement trajectory acquisition means includes
In a non-monitoring area that is not included in the shooting range of each camera, a complementary process for connecting the movement trajectory of the monitoring target is performed based on the output of the sensor means, and the movement trajectory information including the non-monitoring area is generated. The suspicious behavior detection system according to claim 4, further comprising a complementing unit that performs the following. The supplement means is
The suspicious behavior detection system according to claim 6, wherein the supplementary processing is executed based on attribute information including a height of the monitoring target and a feature amount of a pattern.   The behavior identifying means is constructed by a pattern recognition method, mathematically analyzing the characteristics of the movement trajectory information to be monitored, and teaching either or both of a normal pattern and an abnormal pattern as learning information. The suspicious behavior detection system according to claim 1, wherein the suspicious behavior detection system is configured to output information for determining suspicious behavior.   The behavior identifying means, as a method for acquiring the learning information, periodically and automatically based on an arbitrary condition (time zone, location, human attribute information, etc.) from the accumulated data group of movement trajectory information. The suspicious behavior detection system according to claim 1, further comprising means for executing a sequential learning operation to be selected.   The behavior identification means includes a plurality of types of behavior identification means for identifying behavior states having different characteristics based on the movement trajectory information acquired by the movement trajectory acquisition means using learning information having different characteristics as the learning information. The suspicious behavior detection system according to claim 1, comprising:   2. The apparatus according to claim 1, further comprising means for controlling a tracking / zoom imaging function of a camera included in the sensor means to close up a monitoring target determined to be suspicious by the determination means. Suspicious behavior detection system. The action identifying means includes
A means for changing the sensitivity of the identification processing of the identification means according to daytime or nighttime, setting the sensitivity relatively sensitive at night, and setting the sensitivity relatively insensitive during the daytime. The suspicious behavior detection system according to claim 1. A suspicious behavior detection method applied to a suspicious behavior detection system that uses sensor means for detecting a moving position of a monitoring target,
Obtaining the movement trajectory information of the monitoring target based on the output of the sensor means;
Using learning information acquired by learning an action according to a movement trajectory, and identifying an action state of the monitoring target based on the movement trajectory information;
Determining the suspicious behavior of the monitoring target based on the behavior state identified by the step,
The behavior identification step includes
A suspicious behavior detection method, wherein human attribute information is used as the learning information to identify whether the movement trajectory of the monitoring target indicated by the movement trajectory information is normal or abnormal.

Images