KR101572366B1 - Kidnapping event detector for intelligent video surveillance system - Google Patents

Abstract

A method of recognizing a kidnapping situation for an intelligent video surveillance system is disclosed. A computer-implemented method for recognizing a crime scene comprising: extracting a plurality of regions of interest (ROI) corresponding to a person in each frame of a CCTV image; Recognizing a meeting situation in which different interest areas overlap in successive frames; Calculating a feature value according to the change of the ROI in consecutive frames based on the encounter situation when the encounter situation is recognized; And recognizing a crime occurrence event from a feature value according to the change of the region of interest.

Description

{KIDNAPPING EVENT DETECTOR FOR INTELLIGENT VIDEO SURVEILLANCE SYSTEM FOR INTELLIGENT VIDEO MONITORING SYSTEM}

Embodiments of the present invention relate to a technology capable of recognizing a crime scene based on a closed-circuit television (CCTV) image.

In general, CCTV is increasingly installed in residential areas, schools, and roads in urban areas in order to prevent and resolve various incidents / accidents quickly.

Recently, the CCTV system for security has been implemented by a remote monitoring system by a supervisor and an operator in a remote control center using a surveillance camera installed at a crime scene.

As a CCTV video surveillance technology, Korean Patent No. 10-0995569 (Nov. 15, 2010) entitled " Crime Prevention Intelligent CCTV System and Crime Prevention System & Method using the CCTV Video Surveillance Technology " It is possible to display information on the images taken from the camera and various crime prevention using the crime management kiosk (finding suspects, finding witnesses, finding a missing child, and accident scene image information), using real-time image information and crime prevention information Discloses a technique for providing various crime prevention services based on the results of the intelligent object recognition and for each type of crime.

In case of general video surveillance system, surveillance personnel monitoring CCTV monitor many monitors and recognize specific situation and respond accordingly. However, if the surveillance manpower is monitored through the monitor, active surveillance is difficult and the surveillance efficiency is low. When one monitoring person watches two monitors, it can not recognize 45% of events occurring after 10 minutes and 95% after 22 minutes. For this reason, it is not effective for surveillance personnel to monitor multiple CCTV images for 24 hours.

Therefore, the present invention proposes a technology for automatically detecting a crime situation by analyzing a CCTV image.

It provides a method of recognizing the situation of abduction by informing the surveillant automatically by detecting the crime situation through CCTV image analysis.

Provides a method of recognizing a kidnapping situation that automatically recognizes a kidnapping event from the pedestrian information appearing on the CCTV and informs the monitor of the occurrence of the kidnapping situation.

A computer-implemented method for recognizing a crime scene comprising: extracting a plurality of regions of interest (ROI) corresponding to a person in each frame of a CCTV image; Recognizing a meeting situation in which different interest areas overlap in successive frames; Calculating a feature value according to the change of the ROI in consecutive frames based on the encounter situation when the encounter situation is recognized; And recognizing a crime occurrence event from a feature value according to the change of the region of interest.

According to an aspect of the present invention, the extracting step may include extracting a position of a person using a haar-like feature or a histogram of oriented gradient feature, It is possible to assign a unique index to each region.

According to another aspect, the calculating step may calculate a feature value indicating at least one of a size change, a movement speed, and an optical flow change of the ROI.

According to another aspect of the present invention, the step of calculating includes calculating a rate of change of the transverse length of the ROI in the consecutive frame, a rate of change of the longitudinal length of the ROI in the consecutive frame, a sum of the transverse rate of change and the rate of longitudinal variation, A difference between X coordinate values of the center of the frame of interest, a difference between Y coordinate values of the center of the previous frame and the current frame, a sum of the X coordinate difference and the difference between the Y coordinate values, A change rate of the Y coordinate value of the center of the ROI in the consecutive frame, a sum of the X coordinate value change rate and the Y coordinate value change rate, a difference between the previous frame and the ROI in the current frame, (The width of the region of interest after the encounter situation is recognized) / (The sum of the maximum lengths of the respective regions of interest before the situation is recognized) and the sum of the maximum lengths of the regions of interest (before the encounter situation is recognized) The feature value according to the change of the ROI can be calculated using one of the feature values.

According to another aspect of the present invention, the recognizing step may recognize that a crime situation occurs when a feature value according to a change in the ROI exceeds a preset threshold using a crime situation recognition model learned about a crime situation have.

According to another aspect of the present invention, the method may further include providing a notification of a crime occurrence when the crime occurrence event is recognized.

A computer-implemented method for generating a crime scene recognition model, the method comprising: generating, from a training image including at least two images of a crime scene in a state where two pedestrians are encountered and an image representing a general encounter situation, Extracting feature data of a situation; And generating a rule for determining the crime situation based on a property difference between the crime scene characteristic data and the general encounter situation through a machine learning method using characteristic data of the general encounter situation, Provides a model generation method.

According to the embodiment of the present invention, by automatically recognizing the crime situation based on the CCTV image analysis and informing the monitor, it is possible to promptly recognize the crime and promptly respond to the crime.

According to the embodiment of the present invention, pedestrian information appearing in the CCTV is extracted and the abduction situation event is automatically recognized from the pedestrian information, thereby quickly recognizing the abduction situation, thereby promptly responding to the abduction crime, .

FIG. 1 is a flowchart illustrating a method of recognizing a kidnapping situation using a CCTV image according to an exemplary embodiment of the present invention. Referring to FIG.
2 is a flowchart for explaining a specific process of recognizing a kidnapping situation in an embodiment of the present invention.
3 shows an example image of a kidnapping situation.
FIG. 4 exemplarily shows a program screen for providing a notification of a kidnapping situation.
FIG. 5 is a flowchart illustrating a specific process of generating a kidnapping condition recognizer model according to an exemplary embodiment of the present invention. Referring to FIG.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

The embodiments of the present invention can be applied to a CCTV video surveillance system for a crime prevention purpose, and a method and apparatus for recognizing a kidnapping situation by automatically detecting a kidnapping behavior by analyzing a CCTV video image.

In order to prevent crime, demand for CCTV continues to increase, but there is a shortage of surveillance personnel. The monitoring efficiency is greatly reduced when the monitoring person directly monitors the monitor. To solve these problems, the development of an intelligent video surveillance system is progressing actively both domestically and abroad.

Surveillance targets through CCTV are events such as kidnapping, fighting, and theft. Among them, the abduction situation is very important. The kidnapping situation is likely to be missed in very short moments and if the monitor does not monitor the CCTV screen continuously. Therefore, the ability to recognize the kidnapping situation in intelligent CCTV is becoming very necessary. Therefore, if the kidnapping situation is automatically recognized and informed to the monitor, the monitoring efficiency will increase and the crime prevention rate will also improve.

Accordingly, the present invention proposes a kidnapping situation recognizer that analyzes the CCTV image and automatically recognizes the abduction behavior and informs the surveillance person. In detail, the information of the pedestrians appearing on the CCTV video is extracted and based on the information, the CCTV monitor can inform the occurrence of the kidnapping situation by automatically recognizing the kidnapping event. In order to recognize the abduction situation, characteristic information such as situation related static property information, situation related dynamic property information, appearance information of the pedestrian, and moving trajectory information are extracted from a video movie, and these characteristic information is input as learning data, To create a kidnapping situation recognizer model. By using the generated abduction situation recognizer, pedestrian feature information can be automatically extracted from the CCTV image and the abduction situation can be automatically recognized using the learned abduction model.

In the following embodiments, the abduction situation event is described as a representative example. However, the present invention is not limited to this example, but can be applied to event recognition of various situations that can be directly applied to the pedestrian, such as assault or theft.

Feature information for recognition of abduction situation can be defined as follows.

(1) Pedestrian extraction

The pedestrian extraction is to find the position of each pedestrian appearing in the video in each frame of the video. The pedestrian extraction can be performed using haar-like feature or Histogram of oriented gradient feature, which are widely used in existing pedestrian extraction systems. Since the name of the extracted pedestrian can not be known, it can be utilized as an identifier for distinguishing the pedestrian by assigning a unique index to each extracted pedestrian. The position of a pedestrian present in a video frame is represented by ROI (Region Of Interest).

(2) Abduction situation recognizer learning data

Learning data for machine learning of the abduction situation recognizer can be classified into static data and dynamic data.

First, the static data can extract the current state of the pedestrian as data that can be extracted from each frame of the video. Among the static data, the data that can be extracted using the computer vision algorithm include ROI size and ratio, silhouette, HOG, and color histogram. According to the data extraction algorithm, different data can be extracted and the information has information such as pedestrian color information, appearance, and size. Table 1 shows an example of the static data of the abduction status recognizer learning data.

Attribute Name Explanation Width Width of pedestrian ROI Height Height of pedestrian ROI Center_x Horizontal center of ROI Center_y Vertical center of ROI HOG Histogram of pedestrian directional gradient Color Histogram Pedestrian color histogram

Next, the dynamic data is information on continuous movement of the pedestrian, which can be derived from the difference between the previous frames and the current frame of the moving picture. The computer vision algorithm can be used to extract ROI size change, movement speed, optical flow change, etc. in successive frames as dynamic data. These data can also be transformed into attributes for simple movements such as 'walk', 'run', 'sit', and 'stand up'. Table 2 shows an example of dynamic data of the abduction status recognizer learning data.

Attribute Name Explanation Width_rate Rate of change of transverse length of ROI of constant frame Height_rate Vertical rate of change of ROI of constant frame Width_rate + Height_rate Sum of Width_rate and Height_rate X_distance Difference in X value centered on ROI Y_distance ROI centered Y value difference All_distance Sum of X_distance, Y_distance X_rate The rate of change of X value centered on the ROI of a certain frame Y_rate Rate of change of Y value centered on ROI of certain frame X_rate + Y_rate Sum of X_rate and Y_rate Width_sub Difference in width from previous frame Height_sub Vertical length difference from previous frame Width_sub + Height_sub Width_sub, Height_sub Maxwidth_rate (Length of ROI after encounter) / (maximum length of ROI of two pedestrians before encounter) Maxheight_rate (Vertical length of ROI after encounter) / (ROI vertical maximum length value of two pedestrians before encounter) Optical Flow Map Map representing the change of optical flow Walk Whether you are walking or not Run Whether you are running Sit Whether you are sitting Stand Whether or not standing Handshake Whether you are shaking hands Hug Whether you are hugging

(3) Pedestrian appearance characteristics

The appearance characteristics of pedestrians are pedestrian information that can be recognized by computer vision technology in CCTV. The appearance characteristics of the pedestrian are used by using static data and dynamic data of the video image. For example, the clothes color of a pedestrian can be determined by finding the position of the clothes according to the pedestrian's body ratio, quantizing the color histogram of the clothes, and selecting the representative color as the color having the highest ratio. The appearance characteristics of the pedestrian can be very important clues in search of a kidnapper, a kidnapped pedestrian, a missing person, or an event recognition. The characteristics such as Table 3 can be extracted according to the image quality and appearance feature information extraction algorithm of CCTV.

Characteristic Explanation key Actual height value of pedestrian Color on top The colors of the tops of the pedestrians can be multiple colors Bottom color It can be multicolored by the color of the bottom of the pedestrian Pattern on top Patterns of tops can be extracted from patterns defined by developers such as non-patterned and striped patterns. Underlay Can be extracted from patterns defined by developers such as patterns, stripes, etc. Whether hat is worn Whether the hat is worn, expressed as true or false Whether you wear glasses Whether the glasses are worn as true or false Shoe color Pedestrians can be multi-colored with shoes color Skin color May be multiple colors due to pedestrian skin color

(4) Tracking movement of pedestrians

The movement tracking of a pedestrian means that the pedestrians found in the previous frame find out where they moved from the current frame and equally index the same person. As a pedestrian movement tracking method, optical flow using optical invariance, Kalman filter using probability, mean shift using histogram matching, and the like can be used. After extracting the pedestrian area (ROI) from the current frame, the index of the pedestrian that has been found is equally adjusted by tracking the movement of the pedestrian, and the newly introduced pedestrian gives the new index.

FIG. 1 is a flowchart illustrating a method of recognizing a kidnapping situation using a CCTV image according to an exemplary embodiment of the present invention. Referring to FIG.

Referring to FIG. 1, the abduction situation recognition apparatus can perform background modeling on each frame obtained from a CCTV image, and then extract a pedestrian appearing in the frame (S101 to S103). In the case of a fixed camera, the background can be modeled through a Gaussian Mixture Model due to the invariance of the background. Using the modeled background makes it possible to extract pedestrians more accurately and quickly. However, background modeling can be omitted if the camera moves like a PTZ (Pan-Tilt-Zoom) camera. At this time, the abduction situation recognizing device extracts the pedestrian by finding the position of each pedestrian by using the Har-like characteristic or the HOG characteristic for each frame, and then assigns a unique index as an identifier for identifying the extracted pedestrian have. Since objects other than the pedestrian are included even if the object is extracted through the background modeling, only the pedestrian should be extracted through the HOG algorithm or the like in order to accurately recognize the abduction situation. In addition, if the object extraction is performed through the background modeling, there is a possibility that when two pedestrians overlap, the object may be extracted as one object. Therefore, the pedestrian overlapping in the pedestrian extraction process must be extracted separately.

Then, the abduction situation recognizing device judges whether or not the pedestrian extracted from each frame is a pedestrian appearing in the previous frame, and extracts feature information such as static data, dynamic data, appearance characteristics, and moving trajectory of the pedestrian from the corresponding frame S104 to S108). In this case, the feature information excluding the dynamic data is extracted for the first-time pedestrian, and if the present pedestrian is a pedestrian presently present, the previously provided index should be maintained while extracting the feature information including the dynamic data. It maintains the index through the movement tracking algorithm such as the optical flow of the pedestrian surface that existed in the previous frame, maintains the index of the pedestrian through the histogram comparison when the pedestrian appeared after disappearing from the CCTV, and the like again. For example, the color histogram and HOG of the emergent people are stored in the database, and when the new pedestrian appears, it maintains the existing index in comparison with the information of the pedestrians who have already appeared and disappeared. The index maintenance method may vary depending on the performance of the video and the position of the camera.

The feature information may be continuously updated in the database for the kidnapping situation recognition for each of the frames. That is, the database includes pedestrian appearance feature data that can quickly search for pedestrians, as well as static data and dynamic data extracted from moving images for use in abduction recognition. At this time, the static data and the dynamic data can be extracted for each frame and used for recognition in the recognizer that learned the kidnapping situation. The appearance characteristic data of the pedestrian can be extracted by processing the static data or the dynamic data, and a database including the appearance characteristic data of the pedestrian can be constructed so that quick search for the kidnapped or abducted pedestrian can be performed.

Finally, the abduction situation recognition device can recognize the abduction situation of pedestrian characteristic information by using the learned abduction situation recognizer model, and when the abduction situation occurs, it can notify the observer about the situation (S109 to S111).

In the present invention, the CCTV image automatically recognizes the abduction situation and informs the CCTV observer that the observer can quickly recognize and cope with the abduction situation, thereby preventing the development of a major accident in advance.

2 is a flowchart for explaining an example of a specific process of recognizing a kidnapping situation in an embodiment of the present invention.

Referring to FIG. 2, when a plurality of pedestrians are recognized in one frame by receiving each frame of the CCTV image, the abduction situation recognizing device can recognize a first meeting condition between pedestrians while tracking each pedestrian S201 to S203). As an example, the first encounter situation between pedestrians can be recognized as the first time the ROI, which is the pedestrian's position, overlaps (see the third frame in FIG. 3). At this time, before the first encounters between pedestrians occur, pedestrian characteristic information can be managed by collecting characteristic information of each pedestrian and updating the database (S204). On the other hand, when the first encounters between pedestrians occur, the abduction situation recognition apparatus manages the pedestrian characteristic information and calculates a function (f) capable of recognizing the abduction situation. The calculation result of the function (f) (T), it can be recognized that a kidnapping situation has occurred (S205 to S207).

The abduction situation recognition device can be a different recognizer depending on the environment in which the CCTV is installed and the performance of the CCTV. In other words, the reference threshold T value can be set differently according to the purpose, environment, and expected performance of CCTV.

In order for the kidnapping situation recognizer model to be used, a rule must be created through learning. In this process, the threshold value T can be determined. In case of pedestrians encountered in the training data where the general coexistence video and the abduction image are captured, it can be classified as the abduction situation if it exceeds the allowable threshold T based on the difference property of the situation after the encounter.

For example, if the momentum of the pedestrians before the encounter is k, if there are many frames in which the amount of exercise is 2k or more in the abduction situation, and there are many frames of less than 2k in the general companion, this rule first identifies the abduction situation and the accompanying situation And to use the momentum as a rule for the kidnapping recognizer. In order to use the generated rule, it is necessary to calculate the momentum before and after the jumping time. Therefore, we collect the data needed by the rule such as the momentum of the pedestrians before the encounter, recognize the joke frame, and calculate the threshold value (T) to be used in the abduction rule with the jumping data. The final rule to judge the kidnapping situation for the people who appeared in the CCTV image with the calculated threshold value is created, and the data is collected after the encounter.

The part where the generated rule is used in the recognizer is f (Kidnap)> T. f (Kidnap) is a function that recognizes the kidnapping situation. It calculates the number or percentage of frames that are considered to be kidnapped. T is a threshold value for classifying the kidnapping situation. If the result of the kidnapping recognition function is T or more, it is recognized as a kidnapping situation. For example, if the number of frames that the abduction recognition function judges as the abduction situation is calculated, T will be the number of the minimum frames judged as the abduction situation. Accordingly, if the number of frames judged as kidnapping after the encounter is greater than or equal to the threshold value, it is determined that kidnapping is performed.

Fig. 3 shows an example image of a kidnapping situation. The property data of the pedestrian shown in Table 4 is extracted to automatically determine the abduction situation and the general traveling situation.

Attribute Name Explanation Width_rate Rate of change of transverse length of ROI of constant frame Height_rate Vertical rate of change of ROI of constant frame Width_rate + Height_rate Sum of Width_rate and Height_rate X_distance Difference in X value centered on ROI Y_distance ROI centered Y value difference All_distance Sum of X_distance, Y_distance X_rate The rate of change of X value centered on the ROI of a certain frame Y_rate Rate of change of Y value centered on ROI of certain frame X_rate + Y_rate Sum of X_rate and Y_rate Width_sub Difference in width from previous frame Height_sub Vertical length difference from previous frame Width_sub + Height_sub Width_sub, Height_sub Maxwidth_rate (Length of ROI after encounter) / (maximum length of ROI of two pedestrians before encounter) Maxheight_rate (Vertical length of ROI after encounter) / (ROI vertical maximum length value of two pedestrians before encounter)

The attributes of Table 4 can be extracted from both the existing captured situation and the general coexistence situation, and a rule can be made to distinguish between abduction and cohabitation after learning the abduction and accompanying class respectively. At this time, the amount of exercise of the pedestrian used to distinguish the frame of the abduction situation from the frame of the traveling situation can be calculated using at least one of the attributes of Table 4. [ It can use the well-known algorithms used to calculate the momentum of an object between frames.

The data extracted using the pedestrian information extraction algorithm can be inputted into the abduction recognition rule to determine the abduction situation and the abduction situation notification can be provided through the program as shown in Fig. In the kidnapping situation notification program, a video of a predetermined section judged as a kidnapping situation, feature data of a pedestrian extracted from the video, and a notification message of a kidnapping status can be displayed.

FIG. 5 is a flowchart illustrating a specific process of generating a kidnapping condition recognizer model according to an exemplary embodiment of the present invention. Referring to FIG.

As shown in FIG. 5, there are three types of CCTV image data used for generating the kidnap recognizer: a learning CCTV image, a test CCTV image, and an actual CCTV image. Each CCTV image can generate a more accurate model when it is captured in fixed CCTV. In order to create a model of the abduction situation recognizer using CCTV images, the following three steps must be taken.

The first step (S501) is the generation of a kidnap recognition rule using learning CCTV data. The first step is to process the training data and select data and models for more accurate abduction recognition. Since it contains different information for each data, it is necessary to process and extract necessary attributes to create more accurate kidnapping recognizer. In addition, each data mining model has different characteristics. Although it is possible to recognize accurately through a very complicated model, since the kidnapping situation should be recognized in real time, the model satisfying the real-time characteristic is selected.

Data necessary for learning are extracted from actual CCTV images. The data to be used for learning comes from two pedestrians, and the data is extracted by dividing into the image that is kidnapped and the image which is not kidnapped. Static, dynamic data of abduction situation and static and dynamic data of non-abduction situation are all extracted from CCTV image. The data is transformed into the desired data through processing and removes data judged as noise. Then, learning is performed using the processed data.

And a rule for judging abduction through the data mining learning device is generated for the extracted learning data. Among various learning methods, we select and learn effective methods for abduction recognition. In data mining, the more data, the higher the accuracy. In some cases, multiple models are combined to create more powerful rules. However, if the rules are too complicated, it is impossible to recognize the kidnapping in real time. Therefore, the model with high accuracy and high accuracy is selected. According to the angle and position of CCTV, CCTV images can be extracted in different aspects. Therefore, if the model is created for each CCTV, more accurate kidnapping will be possible.

The second step (S502) is a hijacked recognizer test using CCTV data for test. In this step, the accuracy of the recognizer generated through learning is evaluated. The kidnapping situation recognizer model tests the kidnapping recognition rate as an important point because the recognition rate of the kidnapping situation is important.

Test the performance of the kidnapping recognizer by using unused data in the learning model. The data used for the test are independent of the data used for the learning. In reality, it is more dangerous for a recognizer not to recognize a kidnapping situation than to recognize a non-kidnapping situation as a kidnapping, so it is necessary to test whether the recognizer correctly recognizes the kidnapping situation.

The third step (S503) is the actual CCTV-based abduction recognition. The model created through the above two steps (S501 to S502) is combined with the actual CCTV so that the kidnapping situation can be recognized. The speed of recognition depends on the model implementation method and hardware performance. Since the real-time recognition of the kidnapping situation recognizer model is important, it is implemented to realize the abduction recognition as much as possible in real time.

As described above, according to the embodiment of the present invention, by automatically recognizing the crime situation based on the CCTV image analysis and informing the monitor, it is possible to promptly recognize the crime and promptly respond to the crime. Therefore, by extracting pedestrian information from CCTV and automatically recognizing a kidnapping event from pedestrian information, it is possible to quickly respond to a kidnapping crime by increasing awareness of the kidnapping situation, thereby increasing the kidnapping prevention rate.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the apparatus and components described in the embodiments may be implemented within a computer system, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA) , A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (7)

A computer-implemented crime scene recognition method,
Extracting a plurality of regions of interest (ROI) corresponding to a person in each frame of the CCTV image;
Recognizing a meeting situation in which different interest areas overlap in successive frames;
Calculating a feature value according to the change of the ROI in consecutive frames based on the encounter situation when the encounter situation is recognized; And
Recognizing a crime occurrence event from a feature value according to the change of the region of interest
Lt; / RTI >
Wherein the extracting comprises:
Extracting facial feature data including color information, appearance information, and size information related to a facial feature of a person in each frame of interest in each frame before the encounters are recognized,
Storing and maintaining the appearance characteristic data of the ROI extracted from each frame in a database,
A unique index is assigned to each region of interest representing the position of a person as an identifier for identifying each person in each frame,
Comparing the appearance characteristic data of the ROI with the data stored in the database with respect to the ROI extracted in the current frame, maintaining the existing index or giving a new index according to the comparison result,
Wherein the calculating step comprises:
A motion speed, and an optical flow change by comparing consecutive frames before and after the encounters are recognized based on the encounters to recognize the crime occurrence event Compute feature values for dynamic data
A crime scene recognition method. The method according to claim 1,
Wherein the extracting comprises:
Extracting a person's position using a haar-like feature or a Histogram of oriented gradient feature
A crime scene recognition method. delete The method according to claim 1,
Wherein the calculating step comprises:
A ratio of a vertical length change rate of the ROI in the consecutive frame, a sum of the horizontal length change rate and the vertical length change rate, a difference between X coordinate values of the center of the ROI of the previous frame and the current frame, A difference between the X coordinate value difference of the frame and the current frame center of interest, a sum of the difference between the X coordinate value and the Y coordinate value, a change rate of the X coordinate value of the center of the interest region in the consecutive frame, A sum of the X coordinate value change rate and the Y coordinate value change rate, a difference between a previous frame and a current frame in a current frame, a difference between a previous frame and a current frame in a current frame, Sum of differences, (the width of the area of interest after the encounter situation is recognized) / (the maximum width of each area of interest Length of the area of interest after the encounters are recognized) / (sum of the maximum lengths of the respective areas of interest before the encounters are recognized) To calculate characteristic values
A crime scene recognition method. The method according to claim 1,
Wherein the recognizing comprises:
A recognition that a crime situation has occurred when a feature value according to a change in the area of interest exceeds a preset threshold value using a crime situation recognition model learned about a crime situation
A crime scene recognition method. The method according to claim 1,
If the crime occurrence event is recognized, providing a notification of occurrence of a crime situation
The method comprising the steps of: delete

Images