KR20180096038A - Crime prediction system based on moving behavior pattern - Google Patents

Abstract

The present invention relates to a crime prediction apparatus based on a behavior pattern, capable of accurately predicting and preventing a crime, which comprises: a communication unit receiving an image transmitted from a camera; an analysis device analyzing behaviors of all people in the image received through the communication unit to predict a crime; and an output unit notifying a user of an object determined to have an abnormal behavior by the analysis device.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a crime prediction system,

The present invention relates to a modeling method for predicting human behavior, and more specifically, it relates to a modeling method for predicting a behavior of a human being by grasping patterns of a person's body, arms, and hands accompanied by an action such as a criminal act, Based crime prediction system.

Recently, a lot of surveillance cameras have been installed in the outskirts or in a quiet place to prevent crime. In addition, a surveillance camera is installed for the purpose of accident detection even in a place where frequent traffic is frequent due to a large number of people with frequent flow (for example, 10-1520446).

When installing such a surveillance camera, it is important to know how much information can be obtained in comparison with the installation cost and whether to solve the square area. On the other hand, in the conventional surveillance system, a plurality of fixed type cameras are generally installed. Here, the fixed camera means fixing of the shooting direction, not the fixed place. Since a plurality of fixed cameras monitor the periphery by dividing the area, a large number of detection cameras must be installed in one place in order to photograph a square area. Therefore, the installation efficiency and monitoring efficiency are low, There is a problem that monitoring personnel are required to be involved in the analysis.

In order to solve these problems, efforts have been made to reduce the installation cost and solve the labor cost problem by combining with a portable type (rotation type) camera called a tracking camera. However, the images obtained through the combination of the fixed camera and the tracking camera are used only for the post-use purpose to investigate or confirm the emergency situation or the crime situation after the occurrence of the emergency situation or the crime. In the case of the emergency situation or the crime, There is a problem that it does not help.

In this regard, there is an attempt to recognize the next human action through the images collected from the camera. Human behavior awareness has received much attention in recent computer vision and artificial intelligence communities, but challenges still remain due to changes in appearance and multi-object interaction. Although previous studies on human behavior recognition used bodily component motion as an input feature, recent approaches have shown that, due to the limitations of image processing techniques, lower (as a local spatial-temporal feature) - is focused on collecting the features of the level. Another approach is that the development from one action to a complex action needs to be taken into account in the interrelationship between human objects. If it is possible to grasp, for example, between a captured image at a certain point of time and an image captured at a different point of time, a person A, for example, is quickly approaching a person named B, The technologies that are being used do not take this into account.

More specifically, the current crime issue has priority in arrest and prevention, and CCTV is an effective means to supplement the scarcity of manpower and equipment in preventing such crime. Therefore, researches on crime prevention using face recognition and crime behavior prediction using CCTV are being actively carried out. Face recognition is a method of extracting human faces using CCTV high-resolution images recognized at a distance, comparing them with faces of criminals (comparison subjects) stored in the database, and monitoring them in real time when there is a dangerous person using VMS . However, it can not be said that all crimes occur only in the comparison list, and there is inconvenience that the screen should be monitored in real time. Therefore, research on behavior prediction, not just face recognition, has been actively conducted. It mainly analyzes information such as sex, distance, and motion of two subjects, and distinguishes between normal behavior and abnormal behavior. However, due to the technical limitations of CCTV, only the overall motion of the object can be detected, but detailed information such as finger movements can not be analyzed. As a result, it is impossible to accurately analyze the relationship or behavior of the objects, and the accuracy is lowered. If the criminal behavior prediction is able to receive the information including the movement of the finger of the target, the observer who identifies the crime actor's intention and predicts and transmits the aggressiveness, the observer who distinguishes the screen in real time decreases the object to be judged, Although the accuracy can be increased, no technique for predicting criminal behavior based on such a behavior pattern has been proposed.

Many researches on crime have been conducted focusing on the characteristics of criminals. And to identify the causes of crime through personal characteristics of criminals and surrounding environment. In addition to this, it is necessary to have a specific time and space condition in order to commit crime, and prevention of crime through environmental design is attracting attention from crime measures. In addition to these studies, studies are being conducted to predict and prevent crime through behavior prediction using CCTV, but no proposal has been made for predicting criminal behavior.

According to the distance, the risk of crime is different. It distinguishes the dangerous distance by crime based on the proximity study of Edward Hall which classifies the distance to the other side. Edward Hall categorized human distance into four categories: personal distance, intimate distance, social distance, and public distance (personal distances include intimate distances because they are not so different from personal distances and intimate distances). Crime risk is extremely low for public distance (eg, over 3.7 m), but social distance (eg, 1.2 m to 3.7 m) or intimate distance (eg, less than 1.2 m) . However, according to the existing technology using CCTV such as cameras, attempts to predict the possibility of crime in consideration of the possibility of crime due to such distance, possibility of crime according to behavior pattern of a specific part of the body involved in crime behavior, I can not.

The present invention has been made to solve the problems of the prior art described above. It is assumed that CCTV can grasp the detailed movement of the human hand, and in addition to predicting crime behavior studied up to now, Through the behavior analysis, it is aimed to perform more accurate crime prediction and crime prevention.

The present invention also aims at predicting the crime behavior by sequentially grasping the distance between the objects, the movement of the arms, and the movement of the hands. In the action pointing to the opponent, pay attention to the target before the crime occurs, And to provide a crime prevention system that can respond immediately when an act of wielding occurs when an assault occurs.

In order to achieve the above object, according to the present invention, a behavior pattern-based crime predicting apparatus is provided. The apparatus includes a communication unit for receiving an image transmitted from at least one of a fixed camera and a rotary camera, an analyzing unit for analyzing the behavior of all the persons in the image received through the communication unit, And an output unit for notifying an object determined to have an abnormal behavior by the analyzing apparatus, wherein the analyzing apparatus is a distance recognizing module for recognizing distances between all the characters in the image, Or the distance range, a distance recognition module configured to specify a person within the distance range, and an omni view pattern recognition module for analyzing a body pattern and an arm pattern of the specified person, Acquires the coordinates of at least a part of the skeleton, stores the coordinates as three-dimensional coordinates, Analyzing the pattern of the body of water and the pattern of the arm and analyzing the analysis result that the body and arm patterns of the person are out of the predetermined pattern that is the normal pattern, A hand pattern recognition module for analyzing a hand of a person based on the three-dimensional coordinates and transmitting a coordinate change occurring in an abnormal behavior to a determination module, A hand pattern recognition module configured to detect the number of fingers of a person and predict a risk, and a decision module for predicting an abnormal behavior by combining analytical results of each module. The decision module includes a body pattern, an arm pattern, Patterns are identified sequentially, and if abnormal behavior is recognized at each step, proceed to the next step, And a determination module configured to notify, via the output unit, the result of the determination, if it is determined that the behavior deviates from a predetermined reference pattern.

In one embodiment, the skeletal coordinates are selected from the group consisting of head, shoulder, vertebra, central hip, right shoulder, left shoulder, right elbow, left elbow, right wrist, left wrist, right hand, Knee, left knee, right ankle, left ankle, right foot and left foot.

In one embodiment, the omni-view pattern recognition module determines that the pattern of the body is abnormal when the arm of the person is entirely moving, and if the arm is judged to be a third person, It can be judged that the pattern of the arm is an abnormal behavior.

In one embodiment, the hand pattern recognition module may be configured to determine a risk based on the number of fingers analyzed of the person.

In one embodiment, the determination module may determine an abnormal behavior if the abnormal behavior of the body pattern, the arm pattern, and the hand pattern according to the distance recognized by the distance recognition module coincide with each other.

In one embodiment, the determination module determines whether the body pattern of the person in the omni view pattern recognition module is abnormal if the distance between the persons recognized by the distance sensing module is within a social distance or an intimate distance range It may be configured to continuously analyze the body pattern of the person when the distance between the persons is out of the social distance range, even if it is determined that the person is not invisible.

According to the present invention, it is possible to predict the criminal behavior by sequentially grasping the distance between the objects, the movement of the arms, and the movements of the hands.

FIG. 1 is a diagram schematically showing the overall configuration of a crime prevention system according to an embodiment of the present invention.
2 is a flowchart of an omni-view based crime prediction process according to the present invention.
FIG. 3 is a diagram schematically showing a body part that can be considered in predicting a criminal act.
FIG. 4 is a diagram showing a configuration of an analysis apparatus of a crime prevention system according to an embodiment of the present invention in a module format.
5 is a diagram showing an example of an omni view pattern recognition algorithm applied to a crime prevention system according to an embodiment of the present invention.
FIG. 6 is a diagram illustrating an example of a hand pattern recognition algorithm applied to a crime prevention system according to an embodiment of the present invention.
7 is a diagram showing an example of the algorithm of the decision module.
8 is a diagram showing an example of behavior pattern analysis results predicted by applying the crime prevention system according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following description, the description of the algorithm itself, which is already known in association with the behavior pattern recognition, will be omitted. Even if these explanations are omitted, those skilled in the art will readily understand the technical features of the present invention to be protected by the following description.

FIG. 1 schematically shows a general configuration of a crime prevention system according to an embodiment of the present invention. As shown, the system includes, for example, a plurality of fixed cameras 10, a plurality of rotary cameras 20, and a criminal behavior prediction device 100. [

The fixed camera 10 is a camera provided for photographing an object (such as a person) in a predetermined area. The rotary camera 20 is a camera provided for photographing an object while rotating within a predetermined range, unlike a fixed camera. An image of a person photographed from these cameras is transmitted to the apparatus 100 of the present invention via a wired / wireless communication network and is received through the communication unit 110. The communication unit 110 transmits a human image to the analysis apparatus 120. The analysis apparatus 120 models the human behavior in the photographed image based on a predetermined behavior pattern recognition modeling algorithm And outputs the result through the output unit 130, for example, a speaker or a display device. The configuration of each of these devices or the camera itself is a well-known configuration. In the present invention, it is desirable to track the distance between a person and a third person, a predetermined behavior of a person such as a body, an arm, and a hand from a person's image transmitted to the analysis apparatus 120, If a particular event, such as a three-way approach, is followed, it is intended to prevent a situation such as a crime by tracking it. Hereinafter, a specific example of such a crime prediction will be described. It should be understood that the configuration described below is only one example presented for achieving the object of the present invention, and that the present invention is not limited to this configuration itself.

1. Omni-view-based crime prediction flow

In the present invention, the distances between objects, body movements, arm movements, and hand movements are sequentially grasped and the criminal behavior is predicted. If you are doing something that points to your opponent, you should be careful with targeting before the crime occurs, and be prepared to take immediate action if the act of swinging your arms shows that an assault can occur. Hereinafter, the overall concept of a behavior pattern employed by the present invention will be described in detail.

The present invention proposes a crime prediction model that further expands the existing crime prediction. FIG. 2 shows an overall analysis model for predicting criminal behavior according to the present invention. In the present invention, in order to predict a criminal behavior, a predetermined result is sequentially derived through distance, body pattern, arm pattern, and hand pattern.

The present invention grasps the overall inclination of the body and the position of the legs, and grasps the movement of the arms when the risk is determined. Movement of the arm is judged to be a dangerous behavior when stretched toward the preceding object and when raised. Finally, we grasp the movements of the hand, agree with all of the above dangerous behaviors, judge that when the movement of the hand points to the previous object, it attacks when targeting, stretching or holding the fist.

On the other hand, since the distance has already been much studied, in the present invention, after recognizing distances according to the conventional method (for example, distance using CCTV), body patterns, arm patterns and hand patterns, Provide a predictable technical configuration.

2. Body pattern recognition

When you grasp the body pattern, you judge the behavior by the inclination of the body and movement of the arms and legs. When walking normally, the lower body moves constantly, and the body is stretched as a whole, or the upper body is slightly inclined forward. Otherwise, the body of the subject is tilted toward the partner as a whole, and the lower body is judged to be abnormal if the center of the lower leg is concentrated on the leg. If you see the movement of the body in accordance with each abnormal behavior, in the case of targeting, the movement of the arm and the hand changes, and there is not a big change in the body, but when the assault is applied, the center of the lower body concentrates on one leg And the body tilts toward the other side as a whole. Also, arm movements are usually shaken back and forth around thighs, holding something, holding hands when touching a cell phone, but the position of the elbow to shoulder is almost unchanged. On the other hand, when an abnormal behavior is shown, the movement of the entire arm is clearly visible. In the case of targeting, the arm is stretched forward, and in the case of an attack, the position of the hand is raised above the head, and the elbow is raised to the vicinity of the shoulder.

Hand pattern recognition is the most important because targeting is difficult to judge by arm pattern alone. In the case of an attack, since the entire arm is wielded, when the object is continuously photographed, the entire movement of the arm becomes short in a short time.

3. Hand pattern recognition

The hand pattern along with the arm pattern plays an important role in determining abnormal behavior. In the case of an attack, the hand is holding or stretching a fist, and in the case of targeting, the tip of the fingertip faces the other side with only one finger.

4. Omni-view-based prediction module

Through the above process, it is possible to judge the abnormal behavior of the object. It is judged to be abnormal if all of the patterns defined in the present invention are matched with the recognition of the distance from the recognition of the hand. If the distance is a social distance and the direction of the body and arm is pointing toward the other side and the finger is pointing to the other side, the target is a close distance, the body is inclined toward the other side, the arm position is above the shoulder, If you are on this side or holding your fist, you can judge it as an attack. Using the flowchart of FIG. 2, if the behavior of the object is observed as an abnormal behavior, the next step is observed. Otherwise, the object is observed again from the distance recognition module to prevent the crime.

5. Omni-View based criminal behavior prediction method

Hereinafter, an example of a method for specifically embodying the above-described pattern will be described.

4 is a block diagram of a configuration of an analysis apparatus 120 according to an embodiment of the present invention. As shown, the analyzer 120 includes an omni view pattern recognition module 122 for recognizing body patterns and arm patterns, a head pattern recognition module 124 for recognizing hand patterns, And a decision module 126 for determining a behavior pattern of the person based on the perceived pattern. It may also include a distance-aware module to recognize the distance between people in the video shot via CCTV, though this is not shown.

5-1. Implementation of omni view pattern

With infrared sensors, you can not only capture clear images at night, but also observe objects in various directions. CCTV-based distances are assumed to be possible because they have already been used, and they recognize the omni-view pattern of objects. When the skeleton of the object is recognized, a total of 20 points including the joint, head, hand, and foot are obtained as shown in FIG. The coordinates are stored in a storage device (not shown) in a three-dimensional coordinate system, and the body coordinates can be grasped to recognize the body pattern. Table 1 shows these skeleton point indices.

In relation to the recognition of the pattern of the body, the pattern of the body judges the behavior through the inclination of the body and the movement of the legs. When walking normally, the lower body moves constantly, and the body is stretched as a whole, or the upper body is slightly inclined forward. Otherwise, the body of the object B is inclined toward A as a whole, and the lower body is judged to be abnormal if the center of the lower body is centered on the leg. According to each abnormal behavior, in the case of targeting, the movement of the arm and the hand changes and there is no big change in the body, but when the assault is applied, the center of the lower body is concentrated on one leg as mentioned above, A is tilted toward.

Regarding the recognition of the pattern of the arm, when walking normally, the arm is shaken back and forth around the thigh on a daily basis, and the position of the hand is raised when touching the cell phone, but the position from the elbow to the shoulder hardly changes. On the other hand, when an abnormal behavior is shown, the movement of the entire arm is clearly visible. In targeting, the arm is stretched forward, in the case of assault, the position of the hand is raised above the head, and the elbow is raised to the vicinity of the shoulder.

In other words, in the three-dimensional coordinate system, the right side of the object is placed in the x-axis, the top in the y-axis, and the front in the z-axis. In the case of assault, the ELBOW y coordinate of one arm is increased by SHOULDER y coordinate or more, and the y coordinate of WRIST is raised above the HEAD y coordinate. The body tilts toward object A, so that the z-coordinates of HIP_CENTER, SPINE, SHOULDER_CENTER become closer to object A in order. Also, the angle from SHOULDER_CENTER to HIP_CENTER is inclined by more than 15 degrees in the direction of object A. Because the width of the leg is greater than the width of the shoulder, the difference between the x coordinates of both FOOTs is equal to or greater than the difference between the two SHOULDER x coordinates, and one leg extends forward toward A, It shows a difference. In the case of targeting, the body pattern is hard to find a big change, but the target's arm is stretched straight toward the other side. Since the SHOULDER, ELBOW, and WRIST are placed on a straight line, the z-axis extends in a direction close to A, and the x and y coordinates do not show a large difference. The angle of the arm is close to 90 degrees in the body.

FIG. 5 shows an example of an algorithm applied to the omni-view pattern recognition module 122 in order to predict an abnormal behavior through the pattern of the body and the arm. This algorithm follows the skeleton indices in Table 1 for convenience

5.2 convexhull-based hand pattern recognition module

To grasp the hand pattern, we use convexhull and defect. Using convexhull, you can create a polygon that contains the end points of all the fingers. The convexhull algorithm sets a region of interest in the depth image, adjusts the size, scans the contour, approximates the polygon, and then performs a series of processes to detect convexhull and defect. That is, the image of the vicinity of the palm is set as the region of interest using the depth information, and the contour of the depth image is detected. ConvexHull is created using the points generated by polygonal approximation on the detected contour. Using this convexhull and defect, the pattern of the hand can be recognized. By using the convexhull's corners and the number of defects, you can recognize the hand pattern. The convexhull algorithm itself is already known, and further detailed description is omitted. In the present invention, a hand pattern is found by applying the convexhull algorithm. FIG. 6 shows an example of a method for finding a convex hull, that is, an algorithm adopted by the hand pattern recognition module 124. In the present invention, the number of fingers is found through the convexhull and the defect, and the risk is predicted as shown in Table 2 based on this.

Attack (when you open your hand) Assault (when holding a fist) Targeting Number of Corner Points 5 Variance 1 or 2 Number of defects 4 0 1 or 2

5.3. Decision module

Combine these to predict abnormal behavior. Starting from the distance between the objects, the body pattern, the arm pattern, and the hand pattern are grasped sequentially. If abnormal behavior is recognized at each step, the next step is taken. Even if it is not judged abnormal behavior during recognition, continuous observation is done. If the distance between the two objects in the distance cognition module is close to the social distance, the body pattern is observed continuously until the distance is far from the body pattern, even if it does not show abnormal behavior. If an abnormal behavior is recognized at each step, the next step is recognized. For each step, the distance and the corresponding abnormal behavior of body pattern, arm pattern, and hand pattern must match. For example, even if a body pattern such as assault is shown, if the distance between two objects is social distance, it is not judged as abnormal behavior. It is judged to be an abnormal behavior if the pattern of the body (arm) along the distance and the pattern of the hand coincide with the pattern defined in the present invention. FIG. 7 shows an algorithm model adopted by the decision module 126 to derive the outcome as a whole.

5.4 Implementing Results

When the above-described algorithm is applied, the result as shown in FIG. 8 may be dangerous. Depth When viewing from one direction with a camera, it is possible to obtain the coordinates when viewing from various directions as above. Therefore, the skeletal coordinates of the front view can be known even when the object is not in front of the CCTV. As described in the algorithm of the target coordinates, if the height values of the wrist, elbow and shoulder are constant and the direction of the body is directed toward another object, it can be said as targeting. Also, there is a risk of attack if the elbow of the subject is similar to or higher than the shoulder, the hand is similar to or higher than the head, and the leg is extended to the shoulder width. The Hand Pattern shows a polygonal approximation to find the Cunvexhull and Defect of the hand. If a danger is detected, an audible alarm is sounded through the output unit 130 or information is transmitted to the VMS to monitor the object.

6. Concrete examples

First, the criminal behavior predicting apparatus 100 receives an image photographed through the cameras 10 and 20 and transmitted through a wired / wireless communication network through the communication unit 110. Then, the analysis apparatus 120 analyzes the image to determine whether a behavior corresponding to the criminal activity occurs.

In particular, the distance perception module (not shown) of the analysis device determines the distance between all the characters in the image. At this time, the distance cognitive module excludes these persons because there is no risk of crime if the distance between the persons corresponds to the first distance (public distance, more than 3.7 m). However, if characters are included in the second distance (1.2m to 3.7m as a social distance) or the third distance (as an intimate distance, within 1.2m), there is a possibility of a crime. Therefore, .

The omni-view pattern recognition module 122 analyzes a body pattern and an arm pattern of a person. At this time, the omni-view pattern recognition module analyzes the skeleton of the target, and calculates 20 points including the head, shoulder, vertebra, center heap, right shoulder, left shoulder, right elbow, left elbow, right wrist, left wrist, The right hip, the left hip, the right knee, the left knee, the right ankle, the left ankle, the right foot, and the left foot. The omni-view pattern recognition module 1220 stores these coordinates as three-dimensional coordinates and analyzes the body pattern. As a result of this analysis, it is found that the body patterns of the objects A and B are normal patterns (This is a pattern that is preset according to various experiments or criteria. The pattern of normal state is not fixed but may be changed according to a change of a crime pattern or a pattern. If the pattern change of the body is normal It is judged that there is no abnormal behavior. On the other hand, for example, in the case where the body of the analyzed object A is the reference pattern to be used When the pattern of the object is a pattern deviating from a predetermined reference pattern, for example, an object B is inclined as a whole and the lower body is centered on the leg When analyzed by, that the omni pattern view module 122 determines as abnormal behavior, and transmits the result to the determining module 126. The

The omni-view pattern recognition module 122 also analyzes the pattern of the arms of the object A. If it is a normal pattern, the arm of the object A shows a pattern such as shaking back and forth around the thigh. However, if it is an ideal pattern, for example, if the object A is moving toward the preceding object B or the object is moved toward or away from the object B, it is analyzed that the arm moves overall, 122 determines that the pattern of the arm shows an abnormal pattern, and the result is transmitted to the determination module 126. On the other hand, as described above, when analyzing the pattern of the body and the arm, for example, an algorithm as shown in Fig. 5 is used.

The hand pattern recognition module 124 analyzes the hand of the object based on the three-dimensional coordinates. When the hand of an object points to another object, stretches a hand, grasps a fist, or grasps another object, the position of the hand goes up to the head, or the elbow goes up to the shoulder. Since a coordinate change occurs, it is possible to determine an abnormal behavior. In this case, a convexhull algorithm as shown in FIG. 6 is used.

The determination module 126 predicts abnormal behavior by combining the analysis results of each module as described above. That is, starting from the distance between the objects, the body pattern, the arm pattern, and the hand pattern are sequentially grasped, and if an abnormal behavior is recognized at each step, it is determined that the abnormal behavior is deviated from a predetermined reference pattern at each step , It may lead to a criminal action, so the alarm is sounded through the output unit 130 or information is transmitted to the VMS to monitor the object. At this time, if the abnormal behavior is recognized for each step from the distance recognition to the hand recognition, the process proceeds to the next step. In each step, the distance and the corresponding abnormal behavior of the body, arm, do. That is, even if an arm or a hand pattern such as an assault is analyzed, if the distance between the two objects is within the first distance range out of the second distance, it is not judged as an abnormal behavior. That is, if the distance between the two objects is within the second distance or the third distance and the body pattern analysis result indicates that the object is out of the predetermined reference body pattern, do.

On the other hand, in the case of the present embodiment, if the decision module determines that the decision is normal, it is divided into the targeting in which the object A is determined to be directed to the object B and the attack stage in which the object A is likely to attack the object B, .

As described above, according to the present invention, it is intended to effectively utilize a technique for capturing and warning a large frame of abnormal behavior in order to prevent crime in advance. Instead of using time consuming and financially expensive manpower, if CCTV can observe the behavior of the subject with unattended and predict future gestures in the database, the installation of a simple alarm will be enough to prevent crime. For this purpose, the most important technology is to monitor the distance and instantaneous movement between objects by quickly and precisely grasping the coordinates of the objects to be monitored. Further, the technology can be developed to monitor continuous motion changes, and if the database containing the motion is available, the accuracy of the prediction becomes higher. Secondly, if the conventional face recognition or VMS technology is applied, it will be possible to carry out precise monitoring for 24 hours.

While the present invention has been described with reference to the particular embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims (6)

As a behavior pattern-based crime predicting device,
A communication unit for receiving an image transmitted from at least one of a fixed camera and a rotary camera,
An analyzing unit for analyzing the behavior of all the persons in the image received through the communication unit and for predicting the crime,
An output unit for notifying an object judged to have an abnormal behavior by the analyzing apparatus;
Lt; / RTI >
The analyzer
A distance recognition module for recognizing a distance between all the characters in the image, the distance recognition module being configured to specify a person within the distance range if the distance corresponds to a social distance range or an intimate distance range;
An omni view pattern recognition module for analyzing a body pattern and an arm pattern of the specified person, wherein coordinates of a skeleton of at least a part of the skeleton of the person are acquired and stored as three-dimensional coordinates, And analyzing the pattern of the arm and analyzing the result of the analysis so that the pattern of the body and arm of the person is determined to be a normal pattern and deviating from a predetermined reference pattern, A view pattern recognition module,
A hand pattern recognition module for analyzing the hand of the person based on the three-dimensional coordinates and transmitting a coordinate change occurring in an abnormal behavior to a determination module, the convexhull algorithm is used to find the number of fingers of a person, A hand pattern recognition module,
The module is a module for predicting anomalous behavior by combining the analysis results of each module. It is a module that sequentially grasps the body pattern, the arm pattern, and the hand pattern starting from the distance between the persons, and if abnormal behavior is recognized at each step, A determination module configured to notify, via the output unit, the result of the determination, if it is determined that the abnormal behavior is deviated from a predetermined reference pattern at each step,
Based on the behavior pattern of the user. The method of claim 1 wherein the skeletal coordinates are selected from the group consisting of head, shoulders, spine, center heap, right shoulder, left shoulder, right elbow, left elbow, right wrist, left wrist, right hand, left hand, The left knee, the right ankle, the left ankle, the right foot, and the left foot. The omni-vision pattern recognition module according to claim 2, wherein the omni-vision pattern recognition module determines that the pattern of the body is an abnormal behavior when the arm of the person moves as a whole, and when it is determined that the arm performs a third person And determines that the pattern is an abnormal behavior. The apparatus of claim 4, wherein the hand pattern recognition module is configured to determine a risk based on the number of fingers analyzed by the person. [Claim 4] The behavior pattern-based crime predicting apparatus of claim 4, wherein the decision module determines that an abnormal behavior occurs when the abnormal behavior of the body pattern, the arm pattern, and the hand pattern according to the distance recognized by the distance recognition module coincides with each other. The method of claim 5, wherein the determination module determines that the body pattern of the person does not exhibit abnormal behavior in the omni view pattern recognition module when the distance between the persons recognized by the distance sensing module is within a social distance or an intimate distance range Wherein the controller is configured to continuously analyze the body pattern of the person when the distance between the persons is out of the social distance range.

Images