KR20150100141A - Apparatus and method for analyzing behavior pattern - Google Patents

Abstract

Disclosed are an apparatus for analyzing a behavior pattern, and a method thereof. According to the present invention, the apparatus comprises: an image input unit for receiving image data collected through an image photographing apparatus; a background modeling unit for performing background modeling by using the image data inputted during a certain unit time to detect an object of interest from the image data; an object detection unit for detecting the object of interest by using a background model learned through the background modeling; a feature extraction unit for extracting the feature for a behavior pattern of the object of interest from the image data; a behavior pattern modeling unit for learning the behavior pattern of the object of interest by using the extracted feature, and modeling the same; and an analysis unit for analyzing the modeled behavior pattern to determine whether an unusual behavior event of the object of interest occurs or not.

Description

[0001] Apparatus and method for analyzing behavior pattern [

The present invention relates to a video surveillance system, and more particularly, to an apparatus and method for analyzing a behavior pattern for a video surveillance system.

Video surveillance systems are continuously increasing in importance due to various purposes such as personal safety, company security management, military use, and traffic situation analysis. Such a video surveillance system can observe the behavior of objects in an image and predict and detect the occurrence of a specific behavioral state which is a threat through the behavior pattern.

Conventional video surveillance systems require a lot of hardware resources for monitoring and manpower resources in order to operate continuously. In addition to reviewing the recorded images after the occurrence of the incident, the cases of risk situations and specific behaviors The search operates in a passive manner.

That is, the conventional video surveillance technology continuously observes a plurality of image data collected by a person using a video photographing apparatus such as a CCTV through a monitor and displays an object showing a specific behavior (a person approaching a prohibited area, Traffic signal violation cars, etc.) are mainly used. However, it is very difficult and inefficient to manually detect a specific behavior event in real time from a large number of image data having such a large scale. Thus, in some video surveillance systems, a restriction prohibited zone is set in advance, and a method of generating a real-time alarm when motion information of a person or an object is found in the set prohibited zone is used. However, such a method has a limitation that it is possible to monitor only a predetermined area, which is set in advance, and is sensitive to the change of the weather, so that it is difficult to operate in an outdoor environment. In addition, when the surveillance area is widened, a large number of camera networks and the like have constraint conditions necessary.

Accordingly, in order to automate the video surveillance system, it is necessary to accurately and efficiently detect the occurrence of a specific behavior event from a large amount of image data continuously input, including repeated changes of objects over a long period of time, Is required.

The present invention relates to an intelligent behavior pattern analyzing device that accurately and efficiently detects the occurrence of a specific behavior event by analyzing a behavior pattern from a large amount of continuously input image data including various environmental changes and repetitive movements of objects over a long time, .

According to an aspect of the present invention, a behavior pattern analyzing apparatus is disclosed.

A behavior pattern analyzing apparatus according to an embodiment of the present invention includes an image input unit for receiving image data collected through a video photographing apparatus, a background modeling unit for performing a background modeling using image data input for a predetermined unit of time to detect an object of interest from the image data, An object detection unit for detecting an object of interest using the background model learned through the background modeling, a feature extraction unit for extracting a feature of a behavior pattern of the object of interest from the image data, A behavior pattern modeling unit for learning and modeling a behavior pattern of the object of interest using the feature, and an analyzer for analyzing the modeled behavior pattern to determine whether a specific behavior event of the object of interest is generated.

The background modeling unit performs initial background modeling for an initial predetermined unit time, and if the object detecting unit detects the object of interest using the learned background model through initial or existing background modeling, the background modeling unit re- Background modeling is performed adaptively by learning the background model by updating the existing background model.

The background modeling unit performs background modeling using robust principal component analysis (RPCA) for background modeling robust against environmental changes and noise, and for detecting objects of interest using the background modeling.

The object detection unit separates the background and the moving foreground object using the difference between the learned background model and the currently inputted image data and detects the separated foreground object as the object of interest by removing the noise of the separated foreground object .

In order to learn and model the behavior pattern, the feature extraction unit extracts a motion pattern of a pixel unit pertaining to the interest object detected in the input image data, and extracts the feature.

The feature extraction unit may divide the input image data into grid areas of a cell unit, extract motion information of foreground objects in the divided areas, accumulate the extracted motion information in units of a specific time, A feature vector histogram is generated by accumulating the feature vectors in a video unit composed of a plurality of image data, and a matrix of feature vectors composed of the number of frames of the input video is generated.

The behavior pattern modeling unit learns and models the behavior pattern of the object of interest using the motion pattern of the pixel unit extracted from the input image data.

The behavior pattern modeling unit may include a first probability of the feature vector extracted from the input video data and the input video data, a second probability of the feature vector of the specific behavior event, the specific behavior event And learns the second probability and the third probability.

The analysis unit determines a specific action event having the largest value of the third probability among a plurality of defined specific action events as a specific action event occurrence or representative action pattern for the input video data.

If the maximum value of the third probability for a plurality of specific action events is smaller than a preset threshold value, the extracted action pattern is smaller than the probability value for the defined normal action event. Therefore, It is judged as an event.

The behavior pattern modeling unit and the analysis unit simultaneously perform learning of the behavior pattern and detection of the singular behavior event by simultaneously taking into consideration long-term and short-term behavior patterns.

According to another aspect of the present invention, a behavior pattern analysis method performed by a behavior pattern analysis apparatus is disclosed.

A method for analyzing a behavior pattern according to an exemplary embodiment of the present invention includes receiving image data collected through an image capturing apparatus, performing background modeling using image data input for a predetermined unit of time to detect an object of interest from the image data, Extracting a feature of a behavior pattern of the object of interest from the image data using a background model learned through the background modeling, extracting a feature of a behavior pattern of the object of interest from the image data, Learning and modeling a behavior pattern of the object, and analyzing the modeled behavior pattern to determine whether a specific behavior event of the interest object is generated.

The step of performing the background modeling may include performing initial background modeling for an initial predetermined unit time, detecting the object of interest using the learned background model through initial or existing background modeling, And performing background modeling adaptively by learning the background model by updating the initial or existing background model.

Wherein the step of detecting the object of interest comprises the steps of: separating the background and the moving foreground object using the difference between the learned background model and the currently input image data; and removing the foreground object, And detecting with the object of interest.

The step of extracting the feature may include extracting a motion pattern of a pixel unit belonging to the interest object detected in the input image data and extracting the feature to learn and model the behavior pattern.

The step of extracting the motion pattern by extracting the motion pattern of the pixel unit may include dividing the input image data into a grid area of a cell unit, extracting motion information of foreground objects in each of the divided areas, Accumulating the extracted motion information in a specific time unit to generate a feature vector in the form of a histogram vector; accumulating the feature vectors in a video unit composed of a plurality of image data to generate a feature vector histogram; And generating a matrix of feature vectors consisting of the number of frames.

The step of learning and modeling the behavior pattern of the object of interest includes learning and modeling a behavior pattern of the object of interest using the motion pattern of the pixel unit extracted from the input image data.

Wherein learning and modeling a behavior pattern of the object of interest using the motion pattern of the pixel unit comprises: inputting the input video data and a first probability for a feature vector extracted from the input video data, Calculating a third probability of the singular behavior event for the input video data, and learning the second probability and the third probability.

Wherein the step of determining whether or not a specific behavior event of the object of interest generates a specific action event includes generating a specific action event having the largest value of the third probability among a plurality of defined specific action events, And determining a representative behavior pattern.

Wherein the step of determining whether a specific behavior event of the object of interest is generated comprises:

If the maximum value of the third probability for a plurality of specific behavioral events is smaller than a predetermined threshold value, the extracted behavioral pattern is smaller than the probability value for the defined normal behavioral event, so that the extracted behavioral pattern is determined as an abnormal behavioral event .

The present invention can accurately and efficiently detect the occurrence of a specific behavior event by analyzing a behavior pattern from a vast amount of image data continuously input including various environmental changes and repetitive movements of objects over a long period of time.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 schematically illustrates a configuration of an intelligent behavior pattern analyzing apparatus. FIG.
FIGS. 2 to 10 are diagrams for explaining a behavior pattern analyzing apparatus. FIG.
11 is a flowchart showing a behavior pattern analysis method.

While the present invention has been described in connection with certain exemplary 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 similarities. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings. In addition, numerals used in the description of the present invention are merely an identifier for distinguishing one component from another.

Also, in this specification, when an element is referred to as being "connected" or "connected" with another element, the element may be directly connected or directly connected to the other element, It should be understood that, unless an opposite description is present, it may be connected or connected via another element in the middle.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In order to facilitate a thorough understanding of the present invention, the same reference numerals are used for the same means regardless of the number of the drawings.

FIG. 1 schematically illustrates a configuration of an intelligent behavior pattern analyzing apparatus, and FIGS. 2 to 10 are diagrams for explaining a behavior pattern analyzing apparatus. Hereinafter, a behavior pattern analyzer will be described with reference to Fig. 1, with reference to Figs. 2 to 10. Fig.

1, the behavior pattern analyzing apparatus includes an image input unit 10, a background modeling unit 20, a background database 21, an object detecting unit 30, a feature extracting unit 40, a behavior pattern modeling unit 50, A behavior pattern database 51, and an analysis unit 60. [

The image input unit 10 receives image data collected through the image photographing apparatus. For example, the image input unit 10 may receive image data from a CCTV installed in a specific area and perform surveillance, or may receive image data photographed by a moving image.

The background modeling unit 20 performs background modeling using image data input for a predetermined unit of time to detect an object of interest. That is, the background modeling unit 20 performs initial background modeling for a predetermined unit of time, and when the object detecting unit 30 detects an object of interest using the learned background model through initial or existing background modeling, Background modeling can be performed adaptively by reworking the model and updating the initial or existing background model to learn the background model. At this time, the background model calculated through the background modeling may be stored in the background database 21.

For example, the background modeling unit 20 may perform background modeling using robust principal component analysis (RPCA), as shown in FIG.

The task of accurately detecting the background from the input image data is an image processing technique that must be preceded for video surveillance and detection of the object of interest. Basically, the foreground object is detected by the difference between the input image data and the pre- Is mainly used. However, such a method is susceptible to noise, and is susceptible to environmental changes such as weather changes.

Therefore, the background modeling unit 20 can perform background modeling using RPCA for background modeling robust against noise and environment change such as weather change, and detection of an object of interest using the background modeling. Background modeling using RPCA is a type of matrix factorization technique that divides the input image data set M into low-rank matrix L components and sparse matrix S components, and can be performed by optimizing Equation 1 below. Here, the calculated L component becomes the background, and the S component can be regarded as the foreground object or an object other than the background.

[Equation 1]

minimize ∥L∥ + λ∥S∥

subject to L + S = M

Such background modeling overcomes the problem of needing image data of the same state at a specific time without motion change for background modeling in many existing background modeling methods, and can effectively model the background even in an environment where the object is continuously moving There are advantages.

The object detection unit 30 detects an object of interest using the learned background model through initial or existing background modeling. For example, as shown in FIG. 3, the object detecting unit 30 separates the background and the moving foreground object using the difference between the learned background model and the currently input image data, and removes the noise of the separated foreground object So that the separated foreground object can be detected as an object of interest.

The feature extraction unit 40 extracts a feature of a behavior pattern of the object of interest from the input image data in order to learn and model the behavior pattern. That is, the feature extracting unit 40 may extract a motion pattern of a pixel unit belonging to the interest object detected in the input image data, and extract the feature.

For example, the feature extracting unit 40 may extract a motion pattern on a pixel-by-pixel basis from the input image data to extract a feature, as illustrated in FIG.

In order to detect a specific behavior by modeling a behavior pattern, locus information through motion detection and tracking of the objects of interest is basically required. A behavior pattern is modeled on the basis of the motion trajectory, and a behavior pattern of the objects of interest is analyzed for new input image data based on the behavior pattern, thereby determining whether a specific behavior event occurs. Generally, there are two methods for extracting features for learning and modeling behavior patterns. The first is a method of modeling motion trajectory information generated while detecting and tracking an object of interest, and the second is a method of extracting and modeling a motion pattern of a pixel unit such as an optical flow. In the case of using the feature of the object unit, as shown in FIG. 4, a method of separating the foreground in the background and tracking each interest object in the foreground is mainly used in order to extract motion trajectory information on an object basis. In this method, detailed motion definition and detection of each object unit can be performed. However, occlusion occurs between objects in a congested environment in which a plurality of objects appear at the same time. Since the background changes frequently in a moving automobile environment, There is a problem that it is not easy to separate the foreground from the background. Therefore, the feature extracting unit 40 can extract the feature using a method of extracting the motion pattern of the pixel unit pertaining to the interest object detected from the input image data.

The method of extracting a motion pattern on a pixel basis is mainly used in an unsupervised learning method such as topic modeling. Topic modeling is an algorithm that is mainly used in natural language processing. It assigns small movements occurring in a scene as words, assigns clusters of these unit movements to a document, . The subject detected in this way can be classified as a meaningful region on the image screen, and there is an advantage that it can be learned by a comparative learning method without prior information on a large amount of input data.

5, the feature extraction unit 40 divides the input image data into a grid area of a specific cell and extracts motion information of the foreground objects in each of the divided areas, Can be accumulated in a specific time unit to generate a feature vector in the form of a histogram vector. The feature extracting unit 40 accumulates feature vectors in units of video composed of a plurality of video data to generate a feature vector histogram, and generates a matrix of feature vectors composed of the number of frames (i.e., the number of video data) Lt; / RTI >

The behavior pattern modeling unit 50 learns and models a behavior pattern of the object of interest using the features extracted from the input image data. That is, the behavior pattern modeling unit 50 learns and models the behavior pattern of the object of interest using the extracted motion pattern of the pixel unit. At this time, the learned behavior pattern is stored in the behavior pattern database 51. [

For example, referring to FIG. 6, p (W | D) is the probability of the input video data D and the feature vector W extracted therefrom, p (W | Z) (Z | D) may be a probability of a specific action event for the input video data D. The behavior pattern modeling unit 50 can learn p (W | Z) and p (Z | D) using an EM learning algorithm.

The analysis unit 60 analyzes the modeled behavior pattern to determine whether a specific behavior event of the interest object occurs.

For example, the analysis unit 60 calculates p (W | Z) and p (Z | D) using the probability learning method described above with reference to FIG. 6, The specific action event Z can be determined as shown in FIG. That is, the analysis section 60 is the k in the defined specific behavioral events (Z _k), p (Z _k | D) generating unusual behavior event of the Z _k the value is the largest singular behavior events to the video data D _i or It can be judged based on the representative behavior pattern.

For example, the analyzer 60 may determine whether a specific behavior event is generated by dividing the normal behavior event into the abnormal behavior event using the dictionary definition of the normal behavior event. Unusual behaviors are characterized by a rare occurrence of frequency of occurrence, or behavior patterns of objects of interest (vehicles, people, etc.) deviate from behavior patterns defined as normal behavior events. Referring to the feature distribution of normal behavior and abnormal behavior shown in FIG. 8, the feature vectors extracted in a specific unit time in a similar scene have a similar shape. However, referring to the left example, , It can be seen that the fire engine has blocked the vertical traffic flow and the abnormal behavior event occurred across the road horizontally. Therefore, behavior patterns of long-term and short-term are considered at the same time, so that learning of a behavior pattern and detection of a specific behavior event can be accurately performed. At this time, the number k of predefined behaviors can be actively increased depending on the operation of the system.

For example, when the value of max p (Z _k | D) for _k specific action events (Z _k ) is smaller than a predetermined threshold value, the analyzer 60 extracts the extracted normal action Is smaller than the probability value for the event, the extracted behavior pattern can be determined as an abnormal behavior event. 9, the analyzing unit 60 determines a normal behavior event (horizontal traffic flow, vertical traffic flow, and the like) with respect to the video input, and generates an abnormal behavior event (such as fire truck traffic) Situation, collision risk, illegal U-turn, emergency situation, etc.). In addition, the analyzer 60 can detect an abnormal behavior event occurring in a surveillance area or the like, as shown in the example shown in FIG.

11 is a flowchart showing a behavior pattern analysis method.

In step S1110, the behavior pattern analyzing apparatus receives image data collected through the image capturing apparatus. For example, the behavior pattern analyzer may receive image data from a CCTV installed in a specific area and perform surveillance, or receive image data photographed as a moving image.

In step S1120, the behavior pattern analyzing apparatus performs background modeling using image data input for a predetermined unit of time to detect an object of interest. That is, the behavior pattern analyzing apparatus performs initial background modeling for a predetermined unit of unit time, detects the object of interest using the learned background model through the initial or existing background modeling, re-executes the background modeling, Background modeling can be adaptively performed by learning the background model by updating the background model.

In step S1130, the behavior pattern analyzing apparatus detects an object of interest using the learned background model through initial or existing background modeling. For example, the object behavior pattern analyzer uses the difference between the learned background model and the currently inputted image data to separate the background and the moving foreground object, removes the noise of the separated foreground object, Can be detected.

In step S1140, the behavior pattern analyzing apparatus extracts a behavior pattern characteristic of the object of interest from the input image data to learn and model the behavior pattern of the object of interest. That is, the behavior pattern analyzing apparatus extracts a motion pattern of a pixel unit belonging to the interest object detected in the input image data, and extracts the feature.

In step S1150, the behavior pattern analyzing apparatus learns and models the behavior pattern of the object of interest using the features extracted from the input image data. That is, the behavior pattern analyzing apparatus learns and models the behavior pattern of the object of interest using the extracted motion pattern of the pixel unit.

In step S1160, the behavior pattern analyzing apparatus analyzes the modeled behavior pattern by learning and determines whether a specific behavior event of the interest object occurs.

Meanwhile, the behavior pattern analysis method according to an embodiment of the present invention may be implemented in a form of a program command that can be executed through a variety of means for processing information electronically and recorded in a storage medium. The storage medium may include program instructions, data files, data structures, and the like, alone or in combination.

Program instructions to be recorded on the storage medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of software. Examples of storage media include magnetic media such as hard disks, floppy disks and magnetic tape, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, magneto-optical media and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. The above-mentioned medium may also be a transmission medium such as a light or metal wire, wave guide, etc., including a carrier wave for transmitting a signal designating a program command, a data structure and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as devices for processing information electronically using an interpreter or the like, for example, a high-level language code that can be executed by a computer.

The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention as defined in the appended claims. It will be understood that the invention may be varied and varied without departing from the scope of the invention.

10:
20: background modeling unit
21: Background Database
30: Object detection unit
40: Feature extraction unit
50: behavior pattern modeling part
51: behavior pattern database
60: Analytical Department

Claims (20)

A video input unit for receiving video data collected through a video photographing apparatus;
A background modeling unit for performing background modeling using image data input for a predetermined unit of time to detect an object of interest from the image data;
An object detection unit for detecting an object of interest using the background model learned through the background modeling;
A feature extraction unit for extracting a feature of a behavior pattern of the object of interest from the image data;
A behavior pattern modeling unit for learning and modeling a behavior pattern of the object of interest using the extracted features; And
And analyzing the modeled behavior pattern to determine whether a specific behavior event of the interest object is generated.
The method according to claim 1,
The background modeling unit performs initial background modeling for an initial predetermined unit time, and if the object detecting unit detects the object of interest using the learned background model through initial or existing background modeling, the background modeling unit re- And the background model is adaptively performed by learning the background model by updating the existing background model.
3. The method of claim 2,
Wherein the background modeling unit performs background modeling using robust principal component analysis (RPCA) for background modeling robust against environmental changes and noise, and for detecting an object of interest using the background modeling.
The method according to claim 1,
The object detection unit separates the background and the moving foreground object using the difference between the learned background model and the currently input image data, and detects the separated foreground object as the object of interest by removing the noise of the separated foreground object And a behavior pattern analyzing device.
The method according to claim 1,
Wherein the feature extracting unit extracts a motion pattern of a pixel unit belonging to the interest object detected in the input image data and extracts a feature for learning and modeling the behavior pattern.
6. The method of claim 5,
The feature extraction unit may divide the input image data into grid areas of a cell unit, extract motion information of foreground objects in the divided areas, accumulate the extracted motion information in units of a specific time, And generating a feature vector histogram by accumulating the feature vectors in video units composed of a plurality of image data and generating a matrix of feature vectors composed of the number of frames of the input video Pattern analyzer.
The method according to claim 1,
Wherein the behavior pattern modeling unit learns and models the behavior pattern of the object of interest using the pixel-by-pixel motion pattern extracted from the input image data.
8. The method of claim 7,
The behavior pattern modeling unit may include a first probability of the feature vector extracted from the input video data and the input video data, a second probability of the feature vector of the specific behavior event, the specific behavior event Calculates a third probability of the second probability, and learns the second probability and the third probability.
9. The method of claim 8,
Wherein the analysis unit determines a specific behavior event having the largest value of the third probability among a plurality of defined specific behavior events as a specific action event occurrence or representative action pattern for the input video data, Device.
9. The method of claim 8,
If the maximum value of the third probability for a plurality of specific action events is smaller than a preset threshold value, the extracted action pattern is smaller than the probability value for the defined normal action event. Therefore, And determines that the event is an event.
The method according to claim 1,
Wherein the behavior pattern modeling unit and the analyzing unit simultaneously perform the learning of the behavior pattern and the detection of the specific behavior event by simultaneously considering the behavior patterns of the long-term and the short-term. Pattern analyzer.
A behavior pattern analyzing method performed by a behavior pattern analyzing apparatus,
Receiving image data collected through a video photographing apparatus;
Performing background modeling using image data input for a predetermined unit time to detect an object of interest from the image data;
Detecting an object of interest using a background model learned through the background modeling;
Extracting a characteristic of a behavior pattern of the object of interest from the image data;
Learning and modeling a behavior pattern of the object of interest using the extracted features; And
And analyzing the modeled behavior pattern to determine whether a specific behavior event of the interest object is generated.
13. The method of claim 12,
Wherein performing the background modeling comprises:
Performing initial background modeling during an initial predetermined unit time;
If the object of interest is detected using the background model learned through the initial or existing background modeling, re-executing the background modeling; And
And performing background modeling adaptively by learning the background model by updating the initial or existing background model.
13. The method of claim 12,
Wherein detecting the object of interest comprises:
Separating the background and the moving foreground object using the difference between the learned background model and the currently input image data; And
Detecting a separated foreground object as the object of interest by removing noise of the separated foreground object.
13. The method of claim 12,
Wherein the extracting of the feature comprises:
Extracting a motion pattern of a pixel unit belonging to the object of interest detected in the input image data and extracting a feature for learning and modeling the behavior pattern.
16. The method of claim 15,
Wherein the step of extracting the motion pattern of the pixel unit and extracting the feature comprises:
Dividing the input image data into a grid area of a cell unit;
Extracting motion information of foreground objects in each of the divided regions;
Accumulating the extracted motion information by a specific time unit to generate a feature vector in the form of a histogram vector;
Accumulating the feature vectors in units of video composed of a plurality of video data to generate a feature vector histogram; And
And generating a matrix of feature vectors consisting of the number of frames of the input video.
13. The method of claim 12,
Learning and modeling a behavior pattern of the object of interest comprises:
And learning and modeling a behavior pattern of the object of interest using the pixel-by-pixel motion pattern extracted from the input image data.
18. The method of claim 17,
Learning and modeling a behavior pattern of the object of interest using the motion pattern of the pixel unit,
A first probability of a feature vector extracted from the input video data and a second probability of the feature vector with respect to a specific behavior event, a third probability of the specific behavior event with respect to the input video data, Calculating; And
And learning the second probability and the third probability.
19. The method of claim 18,
Wherein the step of determining whether a specific behavior event of the object of interest is generated comprises:
Determining a specific action event having the largest value of the third probability among a plurality of defined specific action events as a specific action event occurrence or representative action pattern for the input video data, Analysis method.
19. The method of claim 18,
Wherein the step of determining whether a specific behavior event of the object of interest is generated comprises:
If the maximum value of the third probability for a plurality of specific behavioral events is smaller than a predetermined threshold value, the extracted behavioral pattern is smaller than the probability value for the defined normal behavioral event, so that the extracted behavioral pattern is determined as an abnormal behavioral event The method comprising the steps of:

Images