is represented by x₀Weight of the N-th order histogram as the center, N represents the number of pixels in the region, k (.) is the Epanechnikov kernel function, x_iIs any point in the region, | | x_i-x₀| is x_iTo x₀Is equal to (d).]Is a unit impulse function, h (x)_i) Is x_iThe corresponding order in the three-dimensional feature histogram.

Selecting a common Bhattacharyya coefficient in the aspect of similarity measurement to calculate the similarity:

wherein

Is represented by x₀Using the built weighted feature histogram as the target model

And a pre-established template q (n), the greater p,the higher the degree of similarity; m represents the order of the histogram.

Mean shift tracking: the target tracking comprises three parts of position prediction, mean shift search and feature update.

The position prediction of the target is realized by adopting a gray template matching method, the approximate position of the target in the current frame can be found through the position prediction, and the accurate position of the target is obtained through mean shift search.

The mean shift location search is shown as equation (26):

where W and H represent the width and height of the target template,

a geometric center coordinate point, x, representing the current target_iFor the sample point, g (.) is the derivative of the kernel function, h represents the kernel bandwidth, ω_iAre weighting coefficients.

The updating of the target model is necessary for realizing stable and accurate target tracking, and blind updating may cause external interference to be doped into the model, so that the model cannot completely describe the characteristics of the target, and as time increases, the model deviates from the real condition of the target more and more, resulting in reduced tracking accuracy.

The strategy for model update is shown in equation (28):

if | ρ_k-ρ_k-1|>ρ_k-1×0.9ANDρ_k>0.9

q_i=q_i-1×0.95+p_k×(1-0.95)（28）

Where rho_kBhattacharyya coefficient, q, representing the best position of the k-th frame_iRepresenting the target color model after the i-th update, p_kA model representing the acquired image object of the k-th frame.

FIG. 5 is a schematic diagram of a software program flow chart of a video alarm, which describes a control method of the intelligent video perimeter fence system, and comprises the following steps:

5.1) after the system is powered on or reset, initializing a video acquisition module, initializing a storage module, loading the system and an application program from respective FLASH by a video alarm and an automatic tracker, completing initialization of a chip and configuration of peripheral hardware, and entering a normal working state; 5.2) creating a video acquisition thread; 5.3) judging whether waiting audio and video are input, if so, entering the next step, and if not, cycling the LOOP to the previous step; 5.4) the video analysis module judges whether a target intrusion event exists according to the key information in the video source, if so, the next step is carried out, and if not, the LOOP LOOPs to the previous step; 5.5) alarming, sending to an automatic tracker, instructing the tracker to track a target, starting an alarm video and a clear picture snapshot at the same time, and sending an early warning signal, an alarm picture and the alarm video to a monitoring center; 5.6) audio and video coding, compression and local storage are carried out, and an alarm signal is sent to a monitoring center; 5.7) creating a video acquisition thread, starting a client service thread and starting a watchdog program at the same time when the video analysis thread is started; 5.8) the client service thread is always in a waiting state, when the client sends a connection request, the client immediately responds, and simultaneously, the XML analysis module is called to analyze the client connection request name and the command processing module is called to process the client connection request name; and 5.9) waiting whether the audio and video data are sent or not, and if so, sending the audio and video data.

The working process is that the embedded operating system starts to start the application program after being started, video acquisition initialization and storage management initialization are sequentially carried out, a video acquisition thread, a video analysis thread and a client service thread are created, and meanwhile, a watchdog program is started.

When audio and video data enter, the acquisition thread immediately calls the video analysis module to detect whether an abnormal target exists in the perimeter fence, if so, the acquisition thread sequentially calls the coding (compression) processing module and the storage processing module to complete coding and compression of the audio and video data and storage on the SD card.

The client service thread is always in a waiting state, when a client sends a connection request, the client immediately responds, meanwhile, the XML analysis module is called to analyze the name of the client connection request, and the command processing module is called to process the name.

FIG. 6 is a schematic diagram showing a software flow chart of an automatic tracker, i.e., the steps of starting an intrusion detection algorithm, confirming the intrusion of a target, and performing response tracking on the target are as follows:

6.1) acquiring the number, the position parameter and the focal length parameter of the monitoring area; 6.2) waiting whether a target tracking instruction comes or not, and if not, looping to the previous step by the LOOP; if a target tracking instruction exists, entering the next step; 6.3 rotating the holder and positioning the target; 6.4) video analysis; 6.5) judging whether the target invasion exists, if not, circulating the LOOP to the previous step; if the target is invaded, entering the next step; 6.6) recording, capturing, tracking and sending the images of the intrusion target.

After the target response tracking module is started, the serial numbers of all set areas in the perimeter defense area and the corresponding position parameters and focal length parameters are firstly obtained. After receiving a tracking instruction sent by a video alarm, searching corresponding position parameters and focal length parameters according to a target position area code in the instruction, controlling a holder to rotate through a control port, adjusting the focal length, zooming a lens, locking a target, simultaneously performing behavior analysis on the target, immediately starting video recording, tracking and snapshotting the target if the target is an invasive target, and transmitting the shot image to an alarm management center. It should be noted that the operating system of the auto-tracker software module still adopts the embedded Linux operating system. Except for the newly added target tracking response module, the other modules of the application software are completely the same as the video alarm application software module.

The system work flow is as follows:

1. after the system is powered on or reset, the video alarm and the automatic tracker load the system and the application program from respective FLASH to complete the initialization of the chip and the configuration of peripheral hardware, and enter a normal working state.

2. The video alarm continuously collects video images in the perimeter defense area through the video collection module and sends the video images to the video analysis module for analysis and processing.

3. The video analysis module judges an abnormal event according to key information in a video source, and immediately acquires target position information and motion track information and sends the information to the automatic tracker according to whether target motion and behavior characteristics violate rules set by an alarm or not, the target is instructed to track the target by the automatic tracker, an alarm video and a clear picture snapshot are started at the same time, and an early warning signal, an alarm picture and the alarm video are sent to a monitoring center.

4. After receiving a target tracking instruction, the automatic tracker controls a self holder to carry out omnibearing rotation and self-adaptive zoom control according to target position information and motion track information in the instruction, and zooms a lens to lock a target; and starting an intrusion detection algorithm, confirming that the target invades, and tracking the target. And simultaneously starting video recording, snapshotting clear pictures, and sending an alarm signal, an alarm picture and an alarm video to a monitoring center.

5. After receiving the early warning or alarm, the monitoring center automatically switches the current picture into an alarm picture and simultaneously sends out alarms (warning tone, sound and light alarm number, horn, short message and multimedia information), and the monitoring personnel confirms the alarm condition and takes corresponding measures. And if necessary, the automatic tracking device of the defense area can be manually controlled to manually search or track the target.

Claims

1. An intelligent video perimeter fence system comprises a front terminal system, an intermediate transmission subsystem and a rear end central monitoring subsystem; the system is characterized in that the front terminal system comprises at least one automatic tracker and at least one or more fixed video alarms, the middle transmission subsystem comprises a wireless router, a switch and a transmission cable, and the rear-end central monitoring subsystem comprises a central server, a monitoring host, a client and a UPS (uninterrupted power supply); the front terminal system equipment and the front terminal system and the rear end central monitoring subsystem are connected in a wired or wireless or mixed transmission mode.

2. The intelligent video perimeter fence system as claimed in claim 1, wherein said video alarm comprises an "ARM + DSP" dual core architecture mode core processor, wherein the dual cores communicate with each other via a PCIE bus, and a digital camera with built-in WIFI, 3G and RJ45 network interface modules and an embedded intelligent video analysis module.

3. The intelligent video perimeter fence system of claim 1 wherein said auto-tracker is comprised of a pan-tilt system, a communications system, and a camera system, the camera system being a high-definition digital camera with an embedded intelligent video analysis module.

4. The intelligent video perimeter fence system of claim 1 wherein the video alarm, the automatic tracker, and the monitoring center are wirelessly transmitted, and the specific communication method is as follows: 3G, WIFI, Bluetooth, COFDM, FSK, Zigbee, wired, any one or combination thereof.

5. The method of controlling an intelligent video perimeter fencing system as claimed in claim 1, further comprising the steps of:

5.1) after the system is powered on or reset, initializing a video acquisition module, initializing a storage module, loading the system and an application program from respective FLASH by a video alarm and an automatic tracker, completing initialization of a chip and configuration of peripheral hardware, and entering a normal working state;

5.2) creating a video acquisition thread;

5.3) judging whether waiting audio and video are input, if so, entering the next step, and if not, cycling the LOOP to the previous step;

5.4) the video analysis module judges whether a target intrusion event exists according to the key information in the video source, if so, the next step is carried out, and if not, the LOOP LOOPs to the previous step;

5.5) alarming, if the video alarm is operated, sending a target tracking instruction to the automatic tracker, and instructing the automatic tracker to track the target; simultaneously starting an alarm video and a sharp picture snapshot;

5.6) audio and video coding, compression and local storage are carried out, and an alarm signal, an alarm picture and an alarm video can be sent to a monitoring center;

5.7) creating a video acquisition thread, starting a client service thread and starting a watchdog program at the same time when the video analysis thread is started;

5.8) the client service thread is always in a waiting state, when the client sends a connection request, the client immediately responds, and simultaneously, the XML analysis module is called to analyze the client connection request name and the command processing module is called to process the client connection request name;

and 5.9) waiting whether the audio and video data are sent or not, and if so, sending the audio and video data.

6. The method for controlling the intelligent video perimeter fence system according to claim 1, wherein the steps of starting an intrusion detection algorithm, confirming the intrusion of the target, and tracking the target are as follows:

6.1) acquiring the number, the position parameter and the focal length parameter of the monitoring area;

6.2) waiting whether a target tracking instruction comes or not, and if not, looping to the previous step by the LOOP; if a target tracking instruction exists, entering the next step;

6.3) rotating the holder and positioning the target;

6.4) video analysis;

6.5) judging whether the target invasion exists, if not, circulating the LOOP to the previous step; if the target is invaded, entering the next step;

6.6) recording, capturing, tracking and sending the images of the intrusion target.