CN103152558A - Intrusion detection method based on scene recognition - Google Patents

Abstract

The invention provides an intrusion detection method based on scene recognition, which belongs to the technical field of intelligent video monitoring. This method effectively solves the problem of real-time intrusion detection in video areas under dynamic background. The invention includes the following steps: initialization: divide the whole video area into N*N image blocks, and calculate the mean value and standard deviation of each image block. Input the video image of the monitoring area: the input image is the video image collected in real time by the surveillance camera, or it can be an image sequence composed of multiple frames by decomposing the collected video file, and input the image one by one according to the time sequence; scene recognition and intrusion Area analysis and processing: According to the existing normal mode of the scene, the current scene is firstly effectively identified and matched, and then the intruded video area is obtained by calculating the mode deviation of each image block and comparing it with the threshold, so as to realize the surveillance video intrusion detection. Mainly used for intrusion detection.

Description

Intrusion Detection Method Based on Scene Recognition

technical field

The invention belongs to the technical field of intelligent video surveillance.

Background technique

Intrusion detection is an important part of the intelligent security system, and it is widely used in the monitoring and protection of key areas, such as military important areas, railways, museums, testing grounds, dangerous areas, warning areas, etc. Compared with installing special sensor equipment (such as infrared rays and voice-activated equipment), intrusion detection based on video images has the characteristics of large detection coverage, simple installation, convenient maintenance, low engineering cost, and wide application, so it has become the current intrusion detection technology. research hotspot.

Intrusion detection based on video image is to use computer vision technology to analyze the video image content of the monitoring scene, automatically detect abnormal conditions in the monitoring screen, and alarm and provide useful information, so as to more effectively remind security personnel to deal with illegal intrusion in a timely manner. At present, intrusion detection methods based on video images mainly include grayscale comparison method, background difference method, frame difference method and optical flow method. These methods realize intrusion detection and alarm functions by detecting moving targets from video sequences. The gray comparison method uses the gray statistical value of the background and the target to detect the moving target, but it is very sensitive to the change of the ambient light. The background difference method extracts the moving target by calculating the difference between the current input frame image and the background image, but the background image needs to be refreshed in real time, and its detection accuracy largely depends on the reliability of the background image. The inter-frame difference method is to subtract two adjacent frames or multiple frames to detect the retained moving target information. Although this method is less affected by changes in ambient light, when the shaking of the camera causes the corresponding "jitter" of the background points of two adjacent frames, this method cannot completely filter out the background, resulting in misjudgment; Or the target whose moving speed is too slow cannot be effectively detected. The optical flow method is to extract the moving target by analyzing the motion field of the image pixels, which is also difficult to deal with the target detection problem in the dynamic background. None of the above methods can effectively solve the problem of intrusion detection in dynamic backgrounds, such as moving cameras, scene switching, etc., and their adaptability is poor and their scalability is not strong.

In view of this, the present invention proposes an intrusion detection method based on scene recognition. This method divides the entire video scene into multiple sub-regions, that is, image blocks, and then establishes various normal modes (non-invasive modes) of the scene according to the mean and standard deviation of each image block, and then recognizes the mode of the current scene at runtime , that is, to match it with these normal patterns, and finally by calculating the deviation between the current pattern and each image block in the corresponding matching pattern and comparing the threshold, the invaded video area (composed of multiple image blocks) is obtained, so as to realize real-time intrusion detection .

Contents of the invention

The purpose of the present invention is to provide an intrusion detection method based on scene recognition, which can effectively realize real-time video area intrusion detection. As the video stream is input, the intrusion area will be dynamically updated according to the intrusion of each image block, so this process automatically realizes the motion tracking of the intrusion target. Therefore, the method of the present invention can be used for intrusion detection tasks of static or dynamic backgrounds, fixed cameras or mobile cameras, and not only has more accurate detection, strong adaptability and expansion capabilities, but also has a simple structure and is easy to implement.

The purpose of the present invention is achieved through the following technical solutions: said technical solution comprises the steps:

(1) Initialization

Divide the entire video area into N×N image blocks, and the size of N can be set according to the actual area covered by the video, such as N=30. According to the pixel brightness value of the image, calculate the mean and standard deviation of each image block. Suppose the mean and standard deviation of the i-th image block are μ _i and σ _i respectively, then the mode Z of the entire scene is a vector composed of the mean and standard deviation corresponding to the N×N image blocks, namely:

Z=(μ ₁ ,σ ₁ ,μ ₂ ,σ ₂ ,…,μ _i ,σ _i ,…,μ _N×N ,σ _N×N ).

For dynamic scenes, individual image patches will have different means and standard deviations in different situations, so dynamic scenes will have multiple different modes. Let Z _k represent the Kth mode of the scene, and the mean value and standard deviation of the i-th image block corresponding to this mode are denoted as μ _k,i and σ _k,i respectively, then:

Z _k = (μ _k,1 ,σ _k,1 ,μ _k,2 ,σ _k,2 ,…,μ _{k,i ,} σ k _{,i ,} …,μ _k,N×N ,σ _{k,N× N} ).

For a fixed camera, the scene changes mainly come from lighting changes (such as day, night, lights) and background motion (such as tree swaying); while for a moving camera, the scene changes will be more, that is, the scene will have more modes . Get the possible modes of the scene before monitoring starts. Specifically, for a fixed camera, representative video images are collected at multiple times and under various weather conditions; for a mobile camera, a group of video images are collected every time the camera moves by a certain angle (such as every 1 degree), This set of images refers to representative video images collected at multiple times and under various weather conditions. According to the images obtained above, calculate various modes of the scene, and these modes are all normal modes of the scene when no intrusion has occurred.

(2) Input the video image of the monitoring area

The video image input for intrusion detection, the input image is the video image collected in real time by the surveillance camera, or it can be an image sequence composed of multiple frames from the collected video file, and the image is input one by one according to the time sequence. If the input image is empty, the whole process is aborted.

(3) Scene recognition

According to the same method as in the initialization, calculate the scene mode Z ^t of the monitoring image at the current moment t, namely:

${\mathrm{<span\; class="notranslate">\; Z</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

和

分别表示场景模式Z^t的第i个图像块的均值和标准差。设

表示当前场景模式Z^t与场景第K个正常模式的距离，则

计算为：in,

and

Represent the mean value and standard deviation of the i-th image block of the scene mode Z ^t , respectively. set up

Indicates the distance between the current scene mode Z ^t and the Kth normal mode in the scene, then

Calculated as:

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; K</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; .</span>$

Calculate and compare the distance between the current scene mode Z ^t and all normal modes of the scene, set is the serial number of the scene normal mode corresponding to the smallest distance among all distances, that is:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; K</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; min</span>}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; K</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}$

和

分别表示场景模式

的第i个图像块的均值和标准差。Among them, H is the set of all normal mode serial numbers of the scene. Therefore, the result of scene recognition is that the first normal mode $Z_{\hat{K}}=({\mathrm{\&mu;}}_{\hat{k},1},{\mathrm{\&sigma;}}_{\hat{k},1}\text{,}{\mathrm{\&mu;}}_{\hat{k},2},{\mathrm{\&sigma;}}_{\hat{k},2},...,{\mathrm{\&mu;}}_{\hat{k},i}\text{,}{\mathrm{\&sigma;}}_{\hat{k},i},...,{\mathrm{\&mu;}}_{\hat{k},N\&times;N},{\mathrm{\&sigma;}}_{\hat{k},N\&times;N})$ As the mode to which the current scene Z ^t belongs, where

and

Indicates the scene mode respectively

The mean and standard deviation of the i-th image block of .

(4) Analysis and treatment of intrusion areas

，计算每个图像块对应的模式偏差。设e_i为第i个图像块对应的模式偏差，则e_i计算为：According to the current scene mode Z ^t and the mode it belongs to

, to calculate the mode deviation corresponding to each image patch. Let e _i be the mode deviation corresponding to the i-th image block, then e _i is calculated as:

${\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; .</span>$

For all N×N image blocks, if the deviation value is greater than the threshold θ _e , the image block is marked as intrusion, otherwise it is not intrusion. The value of θ _e can be selected and set according to the application test results according to the specific situation.

Therefore, if one or more image blocks are marked as being intruded, the whole scene is considered to be intruded, otherwise the scene is considered not to be intruded. For a fixed camera situation, when the scene is invaded, the intrusion area (composed of multiple image blocks) can be directly highlighted and an alarm prompt; for a mobile camera, when the scene is invaded, the camera movement should be stopped first, and then the intrusion area should be highlighted And call the police. With the input of the video stream, the intrusion area will be dynamically updated according to the intrusion of each image block, so the process automatically realizes the motion tracking of the intrusion target.

If the intrusion detection continues, jump to (2), otherwise the whole process stops.

In the method of the present invention, after the above (1)-(4) processing, according to the existing normal mode of the scene, the current scene is firstly effectively identified and matched, and then by calculating the mode deviation of each image block and comparing it with the threshold value, The intruded video area is obtained, so as to realize the intrusion detection of the surveillance video.

Advantages and positive effects of the present invention compared with the prior art: the method establishes a corresponding pattern for the video scene, then recognizes and matches the scene pattern at runtime, and then extracts the invaded video area, thereby realizing real-time intrusion detection and tracking. Therefore, the method of the present invention can be used for the intrusion detection task of static or dynamic background, fixed camera or mobile camera, not only the detection is more accurate, the adaptability and expansion ability are strong, but also the structure is simple and easy to implement.

Description of drawings

Fig. 1 is the schematic diagram that the video area of the present invention is divided into image blocks

Fig. 2 is technical flowchart of the present invention

Detailed ways

The present invention will be further described below according to accompanying drawing:

The present invention can be implemented according to the logical process of the technical flow chart of the present invention and the steps described in the summary of the invention. The method of the invention can be used in various occasions of intrusion detection under video images. Before being used for intrusion detection, firstly, by processing representative scene images under different conditions at different times, various normal modes of the scene may be established or the video area is divided into image blocks; and the video surveillance camera is installed in a suitable position, The video range can cover the required monitoring area; then adopt suitable video transmission means, such as wired or wireless mode, extract the video images collected by the camera in real time during the intrusion detection process, and then perform these images according to the method of the present invention Scene recognition and matching, and finally through the intrusion analysis of each image block in the scene, the regional intrusion detection results are obtained. According to the method of the present invention, it can be used for the intrusion detection tasks of static or dynamic background, fixed camera or mobile camera, not only the detection is more accurate, the adaptability and expansion ability are strong, but also the structure is simple and easy to realize.

Technical scheme of the present invention comprises the steps:

(1) Initialization

Divide the entire video area into N×N image blocks, and the size of N can be set according to the actual area covered by the video, such as N=30. According to the pixel brightness value of the image, calculate the mean and standard deviation of each image block. Suppose the mean and standard deviation of the i-th image block are μ _i and σ _i respectively, then the mode Z of the entire scene is a vector composed of the mean and standard deviation corresponding to the N×N image blocks, namely:

Z=(μ ₁ ,σ ₁ ,μ ₂ ,σ ₂ ,…,μ _i ,σ _i ,…,μ _N×N ,σ _N×N ).

For dynamic scenes, individual image patches will have different means and standard deviations in different situations, so dynamic scenes will have multiple different modes. Let Z _k represent the Kth mode of the scene, and the mean value and standard deviation of the i-th image block corresponding to this mode are denoted as μ _k,i and σ _k,i respectively, then:

Z _k = (μ _k,1 ,σ _k,1 ,μ _k,2 ,σ _k,2 ,…,μ _{k,i ,} σ k _{,i ,} …,μ _k,N×N ,σ _{k,N× N} ).

For a fixed camera, the scene changes mainly come from lighting changes (such as day, night, lights) and background motion (such as tree swaying); while for a moving camera, the scene changes will be more, that is, the scene will have more modes . Get the possible modes of the scene before monitoring starts. Specifically, for a fixed camera, representative video images are collected at multiple times and under various weather conditions; for a mobile camera, a group of video images are collected every time the camera moves by a certain angle (such as every 1 degree), This set of images refers to representative video images collected at multiple times and under various weather conditions. According to the images obtained above, calculate various modes of the scene, and these modes are all normal modes of the scene when no intrusion has occurred.

(2) Input the video image of the monitoring area

The video image input for intrusion detection, the input image is the video image collected in real time by the surveillance camera, or it can be an image sequence composed of multiple frames from the collected video file, and the image is input one by one according to the time sequence. If the input image is empty, the whole process is aborted.

(3) Scene recognition

According to the same method as in the initialization, calculate the scene mode Z ^t of the monitoring image at the current moment t, namely:

${\mathrm{<span\; class="notranslate">\; Z</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

和

分别表示场景模式Z^t的第i个图像块的均值和标准差。设表示当前场景模式Z^t与场景第K个正常模式的距离，则

计算为：in,

and

Represent the mean value and standard deviation of the i-th image block of the scene mode Z ^t , respectively. set up Indicates the distance between the current scene mode Z ^t and the Kth normal mode in the scene, then

Calculated as:

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; K</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; .</span>$

为所有距离中最小距离对应的场景正常模式的序号，即：Calculate and compare the distance between the current scene mode Z ^t and all normal modes of the scene, set

is the serial number of the scene normal mode corresponding to the smallest distance among all distances, that is:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; K</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; min</span>}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; K</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}$

个正常模式 $Z_{\hat{K}}=({\mathrm{\&mu;}}_{\hat{k},1},{\mathrm{\&sigma;}}_{\hat{k},1}\text{,}{\mathrm{\&mu;}}_{\hat{k},2},{\mathrm{\&sigma;}}_{\hat{k},2},...,{\mathrm{\&mu;}}_{\hat{k},i}\text{,}{\mathrm{\&sigma;}}_{\hat{k},i},...,{\mathrm{\&mu;}}_{\hat{k},N\&times;N},{\mathrm{\&sigma;}}_{\hat{k},N\&times;N})$ 作为当前场景Z^t所属的模式，其中

和分别表示场景模式

的第i个图像块的均值和标准差。Among them, H is the set of all normal mode serial numbers of the scene. Therefore, the result of scene recognition is that the first

normal mode $Z_{\hat{K}}=({\mathrm{\&mu;}}_{\hat{k},1},{\mathrm{\&sigma;}}_{\hat{k},1}\text{,}{\mathrm{\&mu;}}_{\hat{k},2},{\mathrm{\&sigma;}}_{\hat{k},2},...,{\mathrm{\&mu;}}_{\hat{k},i}\text{,}{\mathrm{\&sigma;}}_{\hat{k},i},...,{\mathrm{\&mu;}}_{\hat{k},N\&times;N},{\mathrm{\&sigma;}}_{\hat{k},N\&times;N})$ As the mode to which the current scene Z ^t belongs, where

and Indicates the scene mode respectively

The mean and standard deviation of the i-th image block of .

(4) Analysis and treatment of intrusion areas

According to the current scene mode Z ^t and the mode it belongs to Calculate the pattern deviation corresponding to each image patch. Let e _i be the mode deviation corresponding to the i-th image block, then e _i is calculated as:

${\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; .</span>$

For all N×N image blocks, if the deviation value is greater than the threshold θ _e , the image block is marked as intrusion, otherwise it is not intrusion. The value of θ _e can be selected and set according to the application test results according to the specific situation.

Therefore, if one or more image blocks are marked as being intruded, the whole scene is considered to be intruded, otherwise the scene is considered not to be intruded. For a fixed camera situation, when the scene is invaded, the intrusion area (composed of multiple image blocks) can be directly highlighted and an alarm prompt; for a mobile camera, when the scene is invaded, the camera movement should be stopped first, and then the intrusion area should be highlighted And call the police. With the input of the video stream, the intrusion area will be dynamically updated according to the intrusion of each image block, so the process automatically realizes the motion tracking of the intrusion target.

If the intrusion detection continues, jump to (2), otherwise the whole process stops.

In the method of the present invention, after the above (1)-(4) processing, according to the existing normal mode of the scene, the current scene is firstly effectively identified and matched, and then by calculating the mode deviation of each image block and comparing it with the threshold value, The intruded video area is obtained, so as to realize the intrusion detection of the surveillance video.

The method of the present invention can be realized by programming any computer programming language (such as C language), and the system software realized based on the method of the present invention can realize real-time area intrusion detection application in any PC or embedded system.

Claims (2)

1. An intrusion detection method based on scene recognition, comprising the steps of: (1) Initialization Divide the entire video area into N×N image blocks, calculate the mean value and standard deviation of each image block according to the pixel brightness value of the image, set the mean value and standard deviation of the i-th image block to be μ _i and σ _i respectively, Then the mode Z of the entire scene is a vector composed of the mean and standard deviation corresponding to the N×N image blocks, namely: Z=(μ ₁ ,σ ₁ ,μ ₂ ,σ ₂ ,…,μ _i ,σ _i ,…,μ _N×N ,σ _N×N ). For dynamic scenes, each image block will have different means and standard deviations in different situations, so dynamic scenes will have multiple different modes, let Z _k represent the Kth mode of the scene, and the i-th image corresponding to this mode The mean and standard deviation of the block are denoted as μ _k,i and σ _k,i respectively, then: Z _k = (μ _k,1 ,σ _k,1 ,μ _k,2 ,σ _k,2 ,…,μ _{k,i ,} σ k _{,i ,} …,μ _k,N×N ,σ _{k,N× N} ). Obtain possible patterns of the scene before monitoring begins, specifically, for fixed cameras, capture representative video images at multiple times and under various weather conditions; (2) Input the video image of the monitoring area The video image input for intrusion detection, the input image is the video image collected in real time by the surveillance camera, or it can be an image sequence composed of multiple frames from the captured video file, and the images are input one by one in time order; if the input If the image is empty, the entire process is aborted; (3) Scene recognition According to the same method as in the initialization, calculate the scene mode Z ^t of the monitoring image at the current moment t, namely: ${\mathrm{<span\; class="notranslate">\; Z</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$ in,

and Represent the mean value and standard deviation of the i-th image block of the scene mode Z ^t respectively; set

Indicates the distance between the current scene mode Z ^t and the Kth normal mode in the scene, then

Calculated as: ${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; K</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; .</span>$ Calculate and compare the distance between the current scene mode Z ^t and all normal modes of the scene, set

is the serial number of the scene normal mode corresponding to the smallest distance among all distances, that is: $\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; K</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; min</span>}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; K</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}$ Among them, H is the set of all normal mode serial numbers of the scene; therefore, the result of scene recognition is that the first scene of the scene

normal mode $Z_{\hat{K}}=({\mathrm{\&mu;}}_{\hat{k},1},{\mathrm{\&sigma;}}_{\hat{k},1}\text{,}{\mathrm{\&mu;}}_{\hat{k},2},{\mathrm{\&sigma;}}_{\hat{k},2},...,{\mathrm{\&mu;}}_{\hat{k},i}\text{,}{\mathrm{\&sigma;}}_{\hat{k},i},...,{\mathrm{\&mu;}}_{\hat{k},N\&times;N},{\mathrm{\&sigma;}}_{\hat{k},N\&times;N})$ As the mode to which the current scene Z ^t belongs, where

and

Indicates the scene mode respectively

The mean and standard deviation of the i-th image block of ; (4) Analysis and treatment of intrusion areas According to the current scene mode Z ^t and the mode it belongs to Calculate the mode deviation corresponding to each image block, let e _i be the mode deviation corresponding to the i-th image block, then e _i is calculated as: ${\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; .</span>$ For all N×N image blocks, if the deviation value is greater than the threshold θ _e , the image block is marked as intrusion, otherwise it is not intrusion, and the value of θ _e can be selected and set according to the application test results according to the specific situation; After the above (1)-(4) processing, according to the existing normal mode of the scene, the current scene is firstly recognized and matched effectively, and then by calculating the mode deviation of each image block and comparing it with the threshold, the intruded image is obtained. Video area, so as to realize the intrusion detection of surveillance video. 2. A kind of intrusion detection method based on scene recognition according to claim 1, characterized in that: for a mobile camera, when the camera moves a certain angle, a group of video images are collected, and the images are at multiple moments and Representative video images collected under various weather conditions; based on the images obtained above, calculate the various modes of the scene, which are the normal modes of the scene when no intrusion has occurred.

Images