KR101081861B1 - Violence detection method by analyzing the motion image of moving peoples - Google Patents

Abstract

The present invention relates to a method for detecting an assault situation, and more particularly, detects a human object among moving objects in an input frame image, and detects a motion vector of each detected human object, a position vector between the human objects, and the frame image. In the assault situation detection method based on the video information of a moving object that can inform the monitor of occurrence of the assault situation by analyzing the audio information and determining the assault situation, and outputting the video message or the audio message to the outside when the assault situation occurs. It is about.

Assault, frame image, head movement vector, head position vector, object tracking

Description

Violence detection method by analyzing the motion image of moving peoples}

The present invention relates to a method for detecting an assault situation, and more particularly, detects a human object among moving objects in an input frame image, and detects a motion vector of each detected human object, a position vector between the human objects, and the frame image. In the assault situation detection method based on the video information of a moving object that can inform the monitor of occurrence of the assault situation by analyzing the audio information and determining the assault situation, and outputting the video message or the audio message to the outside when the assault situation occurs. It is about.

CCTV (closed-circuit television, CCTV) surveillance system is a video information that judges the abnormality of the surveillance area through video information in the area of crime investigation, security, traffic accident processing, etc. As one of the technologies, it has been used mainly for the purpose of preventing crime by being installed in public places or dangerous places, but recently, it is installed and used in various places such as companies or roads that require security management.

Conventional CCTV surveillance systems are managed through the monitoring of the majority of monitors, the monitoring neglect occurs frequently, there is a problem that the ability to cope with the real-time response due to the increase in the number of CCTV screens to monitor significantly falls.

In this case, the problem can be solved by analyzing the recorded CCTV image. However, the method of analyzing the recorded CCTV image cannot solve the emergency in real time, and the result is not correct according to the resolution or data storage state of the image. There is a problem that can be derived.

The present inventors have tried to determine the assault situation in real time by image processing of the CCTV image without relying on the monitoring of the monitor, and as a result of researching a method for recognizing the assault situation, the movement of a human object in the CCTV image. The present invention has been completed by developing a technical configuration that detects an assault situation in real time using a relationship between human objects and audio information of CCTV images and enables a monitor to recognize the assault situation.

Accordingly, an object of the present invention is to provide an assault situation detection method that can automatically determine the assault situation in real time by detecting the movement of each of the human objects, the movement between the human objects, such as CCTV images.

Another object of the present invention is to provide an assault situation detection method that provides a video message or a voice message to a supervisor at the time of an assault situation so that the supervisor can immediately recognize whether an assault situation occurs in the surveillance zone.

The objects of the present invention are not limited to the above-mentioned objects, and other objects that are not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, the present invention provides a head tracking box including a head region of each human object, the first step of receiving a current frame image, deleting a background without movement, and detecting human objects among moving objects. A second step of generating an image, and a third step of selecting the current frame image as an assault candidate image when the center coordinates of the head tracking boxes are located within a predetermined distance from each other, and the head from the previous frame images to the current frame image. A fourth step of obtaining first head movement vectors, which are motion vectors of each of the tracking boxes, a fifth step of obtaining second head movement vectors, which are motion vectors of each of the head tracking boxes, from the current frame image to a next frame image; A first angle that is an angle difference between the head movement vectors by comparing a first head movement vector and the second head movement vector; Calculate a difference, derive a first change vector using the acceleration information of the first angle difference and the second head movement vector, and then assault situation when the value of the first change vector changes to a value within a certain range. It provides a method for detecting an assault situation comprising a sixth step of determining.

In a preferred embodiment, after the third step and before the sixth step, steps 3-1 to obtain a first head position vector, which is a position vector between head tracking boxes in the current frame image, and the next frame image. And a third step of obtaining a second head position vector, which is a position vector between the head tracking boxes, wherein the sixth step compares the first head position vector and the second head position vector to the head position. After calculating a second angular difference, which is an angular difference between the vectors, deriving a second change vector using the acceleration information of the second angular difference and the second head position vector, the value of the second change vector is constant. In case of further change to the value within, it is determined as the assault situation.

In a preferred embodiment, the sixth step determines whether the voice information of the current frame image or the next frame image is an assault voice, and if the voice information is further determined as an assault voice, it is determined as the assault situation.

In a preferred embodiment, the first step is a first-first step of removing a background of the current frame image and detecting the moving object by using a Gaussian Mixture Model, and median filtering. Step 1-2 of removing image noise of the moving object using masking and masking, and generating a human object box including the moving object, obtaining a histogram of the human object box, Comparing with the histogram, steps 1-3 are determined to determine whether the moving object is a human object.

In a preferred embodiment, the first to third steps further determine whether the moving object is the human object using the ratio of the width and the length of the human object box.

In a preferred embodiment, the second step divides the human object box into three boxes by dividing the human object box into three boxes, obtains a Y-axis histogram of the highest box among the three boxes, and obtains the highest point of the derivative of the value. The head tracking boxes may be generated by setting the neck coordinates of the head region to determine the head region from the neck coordinates to the highest coordinate of the human object.

In a preferred embodiment, in the third step, a candidate image detection area including the head tracking boxes is set, and when the candidate image detection areas cross each other, the current frame image is selected as an assault candidate image. do.

In a preferred embodiment, the first and second head movement vectors and the first and second head position vectors are calculated as motion vectors of the center coordinates of the head tracking boxes.

In a preferred embodiment, if it is determined that the assault situation, a seventh step of notifying the assault situation by outputting a video message or a voice message.

In a preferred embodiment, the first change vector is derived by Equation 1 below.

Here, V ₁ is the first change vector, θ ₁ is the first angle difference, α ₁ is the acceleration information of the second head movement vector.

In a preferred embodiment, when the value of the first change vector is a constant value between 40 and 100 is determined as the assault situation.

In a preferred embodiment, the second change vector is derived by Equation 2 below.

Here, V ₂ is the second change vector, θ ₂ is the second angle difference, α ₂ is the acceleration information of the second head position vector.

In a preferred embodiment, the frame images are input from a closed-circuit television (CCTV).

The present invention has the following excellent effects.

First, according to the assault situation detection method of the present invention, a change vector for the movement of each human object in the frame image input from the CCTV, a change vector for the position between the human objects in the frame image, and audio information of the frame image Complex judgments have the effect of accurately determining the assault situation in real time.

In addition, the assault situation detection method of the present invention has an effect of promptly responding to the assault situation in the surveillance zone by delivering the video or audio message to the monitor immediately when the assault situation occurs.

The terms used in the present invention were selected as general terms as widely used as possible, but in some cases, the terms arbitrarily selected by the applicant are included. In this case, the meanings described or used in the detailed description of the present invention are considered, rather than simply the names of the terms. The meaning should be grasped.

Hereinafter, with reference to the accompanying drawings and preferred embodiments will be described in detail the technical configuration of the present invention.

However, the present invention is not limited to the embodiments described herein but may be embodied in other forms. Like reference numerals designate like elements throughout the specification.

1 is a view showing an assault situation detection system for performing an assault situation detection method according to an embodiment of the present invention, Figure 2 is a head tracking box for explaining the assault situation detection method according to an embodiment of the present invention 3 is a view showing frame images of the assault situation detection method according to an embodiment of the present invention, Figure 4 is a view for explaining a method of selecting an assault candidate image of the assault situation detection method according to an embodiment of the present invention 5 is a view for explaining the head movement vector of the assault situation detection method according to an embodiment of the present invention, Figure 6 is a view for explaining the head position vector of the assault situation detection method according to an embodiment of the present invention. 7 is a flowchart illustrating a method of detecting an assault situation according to an embodiment of the present invention.

Referring to FIG. 1, an assault situation detection method according to an embodiment of the present invention is performed using an assault situation detection system 100, and the assault situation detection system 100 includes a background deletion module 110 and object tracking. The module 120, the voice analysis module 130, the assault situation determination module 140, the assault situation alarm module 150, the display means 160, and the speaker means 170 are included.

In addition, each module of the assault situation detection system 100 is programmed and stored in a storage medium of a computer to perform a function.

In addition, the functions of each component of the assault situation detection system 100 will be described with reference to FIG. 7.

Referring to FIG. 7, in the assault situation detection method according to an embodiment of the present invention, first, the background deletion module 110 receives images in real time from the CCTV 10, and displays a current frame image that is a currently input image ( 11, the background without motion is deleted, and the human objects 11a and 11b of the moving objects are detected (S1000).

Referring to FIG. 2, the background deleting module 110 deletes the background of the current frame image 11 without moving using the Gaussian Mixture Model (GMM). Detect objects Next, image noise of the moving objects is removed through median filtering and masking, and the human object boxes 11a 'and 11b' including the respective moving objects are generated to generate the image noise. The images of the human object boxes 11a 'and 11b' are analyzed and it is determined whether the moving objects are the human objects 11a and 11b, respectively.

In addition, the method of determining the human objects (11a, 11b), first calculates the image histogram (image histogram) of each of the human object boxes (11a ', 11b'), and then calculated each human object box (11a '). 11b ') and the image histogram of a sample of a human object, which is a sample of a general person, to determine whether a moving object in each of the human object boxes 11a' and 11b 'is a human object 11a or 11b. do.

Also, by using the image histogram, a ratio of the horizontal (x) and the vertical (y) of each of the human object boxes 11a 'and 11b' determined as the human objects 11a and 11b is calculated to have a general human object sample. The similarity is determined by comparing the ratio of the width (x) and the length (y), and then the moving object is determined whether the moving object is a real human object (11a, 11b).

That is, in the detecting of the human object in the current frame image 11 (S1000), the human object among the moving objects using the histogram and the horizontal and vertical ratios of each of the human object boxes 11a 'and 11b' are combined. Only (11a, 11b) can be judged correctly.

Next, the object tracking module 120 generates the head tracking boxes 121a and 121b including the head regions of the human objects 11a and 11b among the human object boxes 11a 'and 11b' ( S2000).

In addition, the head tracking boxes 121a and 121b divide the human object boxes 11a 'and 11b' into three boxes by dividing the human object boxes 11a 'and 11b' into three boxes and then y-axis the highest box among the three boxes. A histogram is obtained and the highest point of the derivative of the value, that is, the narrowest coordinate of the object on the y-axis is set as the neck coordinate of each of the human objects 11a and 11b, and the respective human objects 11a, The area up to the highest coordinate of 11b) is generated.

Here, the reason for generating and tracking the head tracking boxes 121a and 121b among the human object boxes 11a 'and 11b' is that the head region of the human body is each of the frame images 11-2, 11-1, and 11b. (11 + 1) is the smallest change, so tracking is easy and tracking accuracy is high.

In addition, the head tracking boxes 121a and 121b are stored in a cache memory (not shown) as RGB data and center coordinates H ₁ and H ₂ , respectively.

Next, the object tracking module 120 determines whether the head tracking boxes 121a and 121b are located at a constant distance from each other, and the current frame image when the head tracking boxes 121a and 121b are located within a predetermined distance. 11 is selected as the assault candidate image (S3000).

3 and 4, the object tracking module 120 includes candidate image detection areas 122a and 122b having a predetermined area including respective head tracking boxes 121a and 121b in the current frame image 11. ), And if the candidate image detection regions 122a and 122b of the current frame image 11 intersect with each other, the current frame image 11 is selected as the assault candidate image.

In other words, when the center coordinates H ₁ and H _{2 of the} head tracking boxes 121a and 121b are located at a constant distance from each other, the current frame image 11 is selected as the assault candidate image.

In addition, the area range of the candidate image detection areas 122a and 122b can be adjusted according to the supervisor's request. When the area range is increased, the detection sensitivity is high, but the calculation amount is increased for the detection of the assault situation, and the area range is increased. If it is made small, the sensitivity of detection is low, but the amount of calculation becomes small.

In addition, although the candidate image detection regions 122a and 122b are illustrated as rectangular regions, the candidate image detection regions 122a and 122b may be selected as circular regions.

Next, the object tracking module 120 uses the RGB data of each of the head tracking boxes 121a and 121b and the center coordinates H ₁ and H ₂ to display previous frame images of the current frame image 11. 11-1) (11-2) is obtained for each of the coordinates _{(H 1 -1, H 2 -1} ) (H 1 -2, H 2 -2).

In this case, the identity of the head tracking boxes between the frame images 11, 11-1, and 11-2 is similar to the RGB data and the center coordinates H ₁ , H ₂ (H ₁ -1, H). _2-1) in the same center ₍₁ H -2, is determined by the distance H ₂ -2), the RGB data is different from each other and each of the frame images 11 (11-1) (11-2) If the coordinates _{(H 1, H 2) (} H 1 -1, H 2 -1) (H 1 -2, H 2 -2) the quality suddenly away from each other, it is determined that this low identity is treated as a background.

Next, the object tracking module 120 coordinates the center coordinates H ₁ and H ₂ of the current frame image 11 from the previous frame images 11-1 and 11-2 to the current frame image 11. The first head movement vectors V ₁ ′, V ₁ ″, which are the motion vectors of the respective center coordinates H ₁ and H ₂ , are calculated using the movement trajectory.

That is, the first head movement vectors V ₁ ′ and V ₁ ″ are motion vectors obtained by moving one human object from the past to the present, and provide motion information before an assault situation occurs.

In addition, the first head movement vectors V ₁ ′ and V ₁ ″ each include angle information, which is information about a direction of movement in the xy plane, and acceleration information, which is information about how quickly it moves.

Next, the object tracking module 120 extracts each of the center coordinates H ₁ + 1 and H ₂ + 1 in the next frame image 11 + 1 of the current frame image 11, and the current frame image. Each of the center coordinates (H ₁ , H ₂ ) of (11) represents second head movement vectors V ₁ '+ 1, V ₁ "+1, which are motion vectors moved in the next frame image 11 + 1. Derived (S4000a).

Also, the second head movement vectors V ₁ '+ 1, V ₁ "+1 also include angle information and acceleration information, respectively, like the first head movement vectors V ₁ ', V ₁ ". do.

Next, a first angle difference θ that is an angle difference between the second head movement vectors V ₁ '+ 1, V ₁ "+1) and the first head movement vectors V ₁ ', V ₁ ". ₁ ', θ ₁ ″ and acceleration information of each of the second head movement vectors V ₁ ′ +1 and V ₁ ″ +1, respectively, for each center coordinate H ₁ of the current frame image 11. H ₂ ) and first change vectors V ₁ , which are motion vectors between the center coordinates H ₁ +1 and H ₂ +1 of the next frame image 11 + ₁ , are derived (S4000b).

In addition, each of the first change vectors V ₁ is derived by Equation 3 below.

Here, V ₁ is the first change vector, θ ₁ is the first angular difference, α ₁ is acceleration information of each of the second head moving vector (V ₁ '+ 1, V ₁ "+ 1).

That is, the first change vector V ₁ is derived for each center coordinate H ₁ +1, H ₂ +1 in the next frame image 11 + 1.

Meanwhile, each of the first change vectors V ₁ of the next frame image 11 + 2 of the next frame image 11 + 1 is a first angle difference θ ₁ of the next frame image 11 + 1. ', θ ₁ ′) and the change amount of the first angular difference between the next frame image 11 + 2 and the acceleration information α ₁ of the next frame image 11 + 1 and the next frame image 11+. It is derived by the amount of change of acceleration information in 2).

Therefore, the first change vectors V ₁ from the next frame image 11 + 2 are derived by Equation 4 below.

Here, Δθ ₁ is a change amount of a first angle difference θ ₁ ', θ ₁ ″ of the next frame image 11 + 1 and a first angle difference of the next frame image 11 + 2, Δα ₁ is a change amount of acceleration information of second head movement vectors of the next frame image 11 + 1 and the next frame image 11 + 1.

Next, the assault situation determination module 140 analyzes the first change vectors V ₁ to determine whether the next frame image 11 + 1 is an assault situation (S5000).

In more detail, the assault situation determination module 140 may determine which of the first change vectors V ₁ for each center coordinate H ₁ + 1, H ₂ + 1 in the next frame image 11 + 1. When the value of one first change vector V ₁ is a constant value between 40 and 100, the next frame image 11 + 1 is determined as an assault situation.

For reference, when the human objects 11a and 11b move, the first change vector V ₁ has a value of 10 to 30.

That is, the assault situation determination module 140 determines the assault situation when the value of the first change vector V ₁ suddenly increases.

Next, when it is determined that the next frame image 11 + 1 is an assault situation, the assault situation alarm module 150 generates a video message and an audio message and outputs the same to the display means 160 and the speaker means 170, respectively. Do it.

For example, the video message may use a method of periodically overlapping a red image with frame images input from the CCTV 10, and the voice message may be an alarm voice such as a siren voice.

However, the assault situation alarm module 150 may notify the occurrence of the assault situation to the watcher's mobile phone using a short message service (SMS) or a multimedia messaging service (MMS).

Meanwhile, referring to FIG. 6, when the current frame image 11 is selected as an assault candidate image, the object tracking module 120 selects between center coordinates H ₁ and H _{2 in} the current frame image 11. A first head position vector V ₂ ′, which is a position vector, is further derived.

In addition, the first derivation of the head position vector (V ₂ ') may be formed at the same time and derivation of the first change vector (V ₁₎ of the first change vector (V ₁₎ may be, derived after deriving the.

Next, the object tracking module 120 derives a second head position vector V ₂ ″, which is a position vector between the center coordinates H ₁ + 1 and H ₂ + 1 in the next frame image 11 + 1. (S4000c).

Next, the object tracking module 120 derives a second change vector V ₂ , which is a change vector of the first head position vector V ₂ ′ and the second head position vector V ₂ ″.

More specifically, the second change vector V ₂ compares the first head position vector V ₂ ′ and the second head position vector V ₂ ″ to compare the head position vectors V ₂ ′, V. ₂ ") an angle a difference between the second angle difference (θ ₂₎ by using the acceleration information of the second angle difference (θ ₂₎ and the second head position vector calculation, and the (α _2), the second change between vector ( V ₂ ) is derived (S4000d).

In addition, the second change vector V ₂ is derived by Equation 4 below.

Here, V ₂ is the second change vector, θ ₂ is the second angle difference, α ₂ is the acceleration information of the second head position vector (V ₂ ").

Meanwhile, the second change vector V ₂ of the next frame image 11 + 2 is the second angle difference θ ₂ of the next frame image 11 + 1 and the next frame image 11 + 2. Change amount of the second angle difference θ ₂ and acceleration information α ₁ of the second head position vector V ₂ ″ of the next frame image 11 + ₁ and the next frame image 11 + 2. Is derived from the amount of change in the acceleration information of the second head position vector V ₂ ′ '.

Therefore, the second change vectors V ₂ from the next frame image 11 + 2 are derived by Equation 6 below.

Here, Δθ ₂ is a change amount between the second angular difference between the next frame image 11 + 1 and the next frame image 11 + 2, and Δα ₂ is the next frame image 11+. The change amount of the acceleration information α ₂ between 1) and the second change vectors V ₂ of the next frame image 11 + 2.

That is, the first change vector (V ₁ ) is derived by motion information in which one center coordinate is moved in several frame images, and the second change vector (V ₂ ) is between several center coordinates of one frame image. It is derived from the distance information.

In addition, the assault situation determination module 140 determines that the assault situation occurs in the next frame image 11 + 1 when the value of the second change vector V ₂ is calculated to be a constant value.

That is, the assault situation determination module 140 determines that the next frame image 11 + 1 is an assault situation when both the first change vector V ₁ and the second change vector V ₂ are determined to be an assault situation. To judge.

However, the assault situation determination module 140 may determine the next frame image 11 + 1 even if only one vector of the first change vector V ₁ and the second change vector V ₂ is determined as the assault situation. Can be judged as an assault situation.

In addition, the voice analysis module 130 receives voice information of the current frame image 11 or the next frame image 11 + 1 from the CCTV 10 and determines whether the voice information is an assault voice. (S4000e).

In addition, the determination of the assault sound may be made after the derivation process of the first change vector V ₁ and the second change vector V ₂ , or the first change vector V ₁ and the second change vector V. It can be done simultaneously with the derivation process of V ₂ ).

For example, the voice analysis module 130 may determine the voice information as an assault voice when the voice information has a decibel (dB) value within a predetermined range, and the assault information such as “save” the syllable information of the voice information. In the case of syllables occurring in a situation, the voice information may be determined to be a flood situation.

That is, the assault situation determination module 140 determines the next frame image 11 + 1 when the first change vector V ₁ , the second change vector V ₂ , and the audio information are all determined to be an assault situation. ) Is judged as an assault situation.

However, when the assault situation determination module 140 determines that any one of the first change vector V ₁ , the second change vector V ₂ , and the voice information is an assault situation, the next frame image 11 You can also judge +1) as an assault situation.

Therefore, the assault situation can be accurately detected by using the video and audio information of the CCTV in combination.

As described above, the present invention has been illustrated and described with reference to preferred embodiments, but is not limited to the above-described embodiments, and is provided to those skilled in the art without departing from the spirit of the present invention. Various changes and modifications will be possible.

1 is a view showing an assault situation detection system for performing an assault situation detection method according to an embodiment of the present invention;

2 is a view for explaining the head tracking box of the assault situation detection method according to an embodiment of the present invention,

3 is a view showing frame images of an assault situation detection method according to an embodiment of the present invention;

4 is a view for explaining a method for selecting an assault candidate image of the assault situation detection method according to an embodiment of the present invention;

5 is a view for explaining the head movement vector of the assault situation detection method according to an embodiment of the present invention,

6 is a view for explaining the head position vector of the assault situation detection method according to an embodiment of the present invention;

7 is a flowchart illustrating a method of detecting an assault situation according to an embodiment of the present invention.

In the drawings according to the present invention, the same reference numerals are used for components having substantially the same configuration and function.

<Description of the symbols for the main parts of the drawings>

100: assault situation detection system 110: background deletion module

120: object tracking module 130: voice analysis module

140: assault situation determination module 150: assault situation alarm module

160: display means 170: speaker means

11-2, 11-1, 11, 11 + 1, 11 + 2: Frame video 11a, 11b: Human object

11a ', 11b': Person Object Box 121a, 121b: Head Tracking Box

Claims (13)

A first step of receiving a current frame image and deleting a background having no movement and detecting human objects among moving objects; A second step of generating head tracking boxes including head areas of the respective human objects; A third step of selecting the current frame image as an assault candidate image when the center coordinates of the head tracking boxes are located within a predetermined distance from each other; Obtaining first head movement vectors, which are motion vectors of the head tracking boxes, from previous frame images to the current frame image; A fifth step of obtaining second head movement vectors, which are motion vectors of the head tracking boxes, from the current frame image to a next frame image; And Comparing the first head movement vector and the second head movement vector, a first angle difference, which is an angular difference between the head movement vectors, is calculated, and acceleration information of the first angle difference and the second head movement vector is used. And a sixth step of deciding a first change vector and determining an assault situation when the value of the first change vector changes to a value within a predetermined range. The first step, Generating a human object box including the moving object, obtaining a histogram of the human object box, and comparing it with a histogram of a sample of the human object to determine whether the moving object is a human object. The method of claim 1, After the third step and before the sixth step, Obtaining a first head position vector which is a position vector between the head tracking boxes in the current frame image; And And calculating a second head position vector, which is a position vector between the head tracking boxes, in the next frame image. The sixth step may be performed by comparing the first head position vector and the second head position vector to calculate a second angle difference that is an angle difference between the head position vectors, and comparing the second head position vector with the second head position vector. After deriving the second change vector using the acceleration information, if the value of the second change vector changes to a value within a predetermined range assault situation detection method characterized in that it is determined as the assault situation. delete The method according to claim 1 or 2, The first step: A step 1-1 of removing a background of the current frame image and detecting the moving object by using a Gaussian Mixture Model; And And a first to second step of removing image noise of the moving object by using median filtering and masking. A first step of receiving a current frame image and deleting a background having no movement and detecting human objects among moving objects; A second step of generating head tracking boxes including head areas of the respective human objects; A third step of selecting the current frame image as an assault candidate image when the center coordinates of the head tracking boxes are located within a predetermined distance from each other; Obtaining first head movement vectors, which are motion vectors of the head tracking boxes, from previous frame images to the current frame image; A fifth step of obtaining second head movement vectors, which are motion vectors of the head tracking boxes, from the current frame image to a next frame image; And Comparing the first head movement vector and the second head movement vector, a first angle difference, which is an angular difference between the head movement vectors, is calculated, and acceleration information of the first angle difference and the second head movement vector is used. And a sixth step of deciding a first change vector and determining an assault situation when the value of the first change vector changes to a value within a predetermined range. The first step, Generating a human object box including the moving object, and determining whether the moving object is the human object by using a ratio between the width and the length of the human object box. A first step of receiving a current frame image and deleting a background having no movement and detecting human objects among moving objects; A second step of generating head tracking boxes including head areas of the respective human objects; A third step of selecting the current frame image as an assault candidate image when the center coordinates of the head tracking boxes are located within a predetermined distance from each other; Obtaining first head movement vectors, which are motion vectors of the head tracking boxes, from previous frame images to the current frame image; A fifth step of obtaining second head movement vectors, which are motion vectors of the head tracking boxes, from the current frame image to a next frame image; And Comparing the first head movement vector and the second head movement vector, a first angle difference, which is an angular difference between the head movement vectors, is calculated, and acceleration information of the first angle difference and the second head movement vector is used. And a sixth step of deciding a first change vector and determining an assault situation when the value of the first change vector changes to a value within a predetermined range. In the second step, each human object is divided into three boxes by dividing the human object into Y-axis, and the Y-axis histogram of the highest box among the three boxes is obtained, and the highest point of the derivative of the value is set as the human neck coordinate. And determining the head tracking box from the neck coordinate to the highest coordinate of the human object as the head region to generate the head tracking boxes. The method according to claim 1, 5 or 6, In the third step, a candidate image detection region including the head tracking boxes is set, and when the candidate image detection regions cross each other, the assault situation is characterized in that the current frame image is selected as an assault candidate image. Detection method. A first step of receiving a current frame image and deleting a background having no movement and detecting human objects among moving objects; A second step of generating head tracking boxes including head areas of the respective human objects; A third step of selecting the current frame image as an assault candidate image when the center coordinates of the head tracking boxes are located within a predetermined distance from each other; Obtaining first head movement vectors, which are motion vectors of the head tracking boxes, from previous frame images to the current frame image; A fifth step of obtaining second head movement vectors, which are motion vectors of the head tracking boxes, from the current frame image to a next frame image; And Comparing the first head movement vector and the second head movement vector, a first angle difference, which is an angular difference between the head movement vectors, is calculated, and acceleration information of the first angle difference and the second head movement vector is used. And a sixth step of deciding a first change vector and determining an assault situation when the value of the first change vector changes to a value within a predetermined range. And the first and second head movement vectors and the first and second head position vectors are calculated as a motion vector of a center coordinate of each head tracking box. A first step of receiving a current frame image and deleting a background having no movement and detecting human objects among moving objects; A second step of generating head tracking boxes including head areas of the respective human objects; A third step of selecting the current frame image as an assault candidate image when the center coordinates of the head tracking boxes are located within a predetermined distance from each other; Obtaining first head movement vectors, which are motion vectors of the head tracking boxes, from previous frame images to the current frame image; A fifth step of obtaining second head movement vectors, which are motion vectors of the head tracking boxes, from the current frame image to a next frame image; Comparing the first head movement vector and the second head movement vector, a first angle difference, which is an angular difference between the head movement vectors, is calculated, and acceleration information of the first angle difference and the second head movement vector is used. A sixth step of deriving a first change vector and determining an assault situation when the value of the first change vector changes to a value within a predetermined range; And And a seventh step of notifying the assault situation by outputting a video message or an audio message when it is determined as the assault situation. A first step of receiving a current frame image and deleting a background having no movement and detecting human objects among moving objects; A second step of generating head tracking boxes including head areas of the respective human objects; A third step of selecting the current frame image as an assault candidate image when the center coordinates of the head tracking boxes are located within a predetermined distance from each other; Obtaining first head movement vectors, which are motion vectors of the head tracking boxes, from previous frame images to the current frame image; A fifth step of obtaining second head movement vectors, which are motion vectors of the head tracking boxes, from the current frame image to a next frame image; And Comparing the first head movement vector and the second head movement vector, a first angle difference, which is an angular difference between the head movement vectors, is calculated, and acceleration information of the first angle difference and the second head movement vector is used. And a sixth step of deciding a first change vector and determining an assault situation when the value of the first change vector changes to a value within a predetermined range. The assault condition detection method of claim 1, wherein the first change vector is derived by Equation 1 below. [Equation 1]

Here, V ₁ is the first change vector, θ ₁ is the first angle difference, α ₁ is the acceleration information of the second head movement vector. 11. The method of claim 10, The assault situation detection method, characterized in that the determination as the assault situation when the value of the first change vector is a constant value between 40 and 100. The method of claim 2, The second change vector is an assault situation detection method, characterized in that derived by the following equation (2). [Equation 2]

Here, V ₂ is the second change vector, θ ₂ is the second angle difference, α ₂ is the acceleration information of the second head position vector. delete

Images