CN102831617A - Method and system for detecting and tracking moving object - Google Patents

Abstract

The invention discloses a moving target detecting and tracking method based on a DSP (digital signal processor), and a system for realizing the method. The method comprises the following steps of: firstly, establishing a background model by using a method that an interframe difference is combined with a background difference aiming at a Y-component of an image; subsequently detecting a foreground moving target; establishing an initial target template; carrying out pyramid downsampling for two times on the target template and a video image to be tracked respectively so as to reduce resolution of the target template and the video image to be tracked. A particle group optimization algorithm is adopted on a top layer of a pyramid to position a tracked target; a diamond search method is adopted on a middle layer and a bottom layer of the pyramid to position the tracked target; and the target template is updated constantly along with change of the moving target during a process that the target is tracked, thereby realizing real-time continuous tracking to the target.

Description

A method and system for moving target detection and tracking

technical field

The invention belongs to the field of image processing and computer vision, relates to DSP technology, in particular to a method and a system for detecting and tracking a moving target of a pyramid multi-resolution algorithm, a particle swarm optimization algorithm and a diamond search algorithm.

Background technique

The detection and tracking technology of moving objects is an important branch of computer vision, and it is a hot issue in intelligent image processing and video processing at present. It has broad application prospects and long-term economic value. In terms of military affairs, moving target detection and tracking technology can be used to detect and track moving targets in the air or ground surveillance range; in terms of social life, moving target detection and tracking technology can be used for real-time monitoring of various units to detect danger in time , to predict possible dangers; in medicine, the detection and tracking technology of moving objects can enable doctors to better diagnose the disease and relieve the pain of patients.

The moving target detection technology refers to finding out the area where the moving target is located in the image through the operation between the image sequences, and preparing for the matching and tracking of the moving target. Commonly used moving target detection methods mainly include: background difference method, frame difference method and optical flow method. However, the background difference method cannot adapt to the changes in light in the actual environment; the motion area segmented by the frame difference method often has holes; the calculation amount of the optical flow method is huge, and the algorithm is complex and cannot meet the requirements of real-time tracking.

The moving object tracking technology refers to analyzing and detecting the obtained moving object, associating the same moving object in different frames to obtain its moving track. Commonly used moving target tracking methods are: template-based tracking, feature-based tracking, target area-based tracking and target contour-based tracking. But template-based tracking has a lot of calculations; feature-based tracking is difficult to determine the characteristics of the target; target area-based tracking is easy to lose the target when the target is occluded or deformed; tracking based on the target outline is difficult to track quickly or quickly in real time. Large-scale targets.

）和整个粒子群的最优解（全局极值）来更新自己的速度

和位置

，进而通过适应度评价函数计算对应的适应值，进而比较适应值选择出整体最优解，依据下式更新下一代的速度和位置，不断迭代从而寻找到系统的最优解。 The particle swarm optimization algorithm is an optimization method for solving complex problems proposed by Dr. Kennedy and Dr. Eberhart through the research on the foraging behavior of birds. When solving the optimal solution of the problem, each particle passes through its own optimal solution. (individual extremum

) and the optimal solution of the entire particle swarm (global extremum ) to update their speed

and location

, and then calculate the corresponding fitness value through the fitness evaluation function, and then compare the fitness value to select the overall optimal solution, update the speed and position of the next generation according to the following formula, and continuously iterate to find the optimal solution of the system.

              

                                 

                                 

为惯性权重，为粒子的速度，

为学习因子，其取值为

，

和

是（0,1）之间的随机数，

代表个体极值，

代表全局极值。粒子在空间中不断学习个体极值和全局极值的经验更新粒子速度和位置直到寻找到最优解。 in,

is the inertia weight, is the velocity of the particle,

is the learning factor, and its value is

,

and

is a random number between (0,1),

represents the individual extremum,

represents the global extremum. The particle continuously learns the experience of the individual extremum and the global extremum in the space to update the particle velocity and position until the optimal solution is found.

The diamond search method refers to making full use of the central paranoid idea of the motion vector to find the best matching area between the target module and the area to be searched by searching for points near the center of the window. When using this method to search for the best matching area, choosing a small search template may fall into a local optimum, and choosing a large search template may not find the best matching area, so this method uses diamond templates of two shapes and sizes for matching search . The search process is as follows: First, start from the center of the search window, and repeatedly use the big diamond template to search. If the best matching point appears on the edge of the module, then use this point as the center to continue searching with the large diamond template; if the best matching point appears in the center of the template, then use the small diamond template to search for the surrounding four points , to find the best matching point, and finally end the search. During the search process, if the points that need to be searched in this step have been searched in the previous step, then these points are no longer considered. If some points in the template are beyond the scope of the search window, then these points are no longer considered. Although the diamond search method does not limit the number of searches, but due to the center offset of the motion vector, the best matching area will be searched quickly.

At present, more effective moving target detection and tracking methods need to be proposed.

Contents of the invention

In order to overcome the shortcomings of the above-mentioned existing technical problems, the present invention proposes a real-time DSP-based moving target detection and tracking method and an image processing system for realizing the method, thereby realizing real-time detection and tracking of moving targets.

The DSP-based moving target detection and tracking system provided by the present invention includes a video acquisition module, a video processing module and a display module; the video acquisition module collects video signals through a CCD camera, and transmits the collected analog video signals to SEED VPM642 video The processing module converts the analog video signal into a video signal in BT.656 format through the high-performance video decoder TVP5150 in VPM642, and transmits the signal to the video interface of DSP; the video interface of DSP combines the EDMA channel to transmit the video signal to In the buffer area of SDRAM, after the data is processed by DM642, the video encoder SAA7121H converts the data into an analog signal and sends it to the display of the display module, and displays the foreground moving target through the display.

The present invention also provides an improved background modeling method for the Y component in the collected image, comprising the following steps:

，其差分图像分别记为

，其中

，其三帧差分图像记为

，此处&代表逻辑与运算； Step 1: First input a frame of image, determine whether the number of frames is greater than 3, and when the number of frames is greater than 3, perform three consecutive frames of difference on the Y component of the input image, and binarize the image after the difference; assuming that the input of three The frame images are

, and their difference images are denoted as

,in

, and its three-frame differential image is denoted as

, where & stands for logical AND operation;

进行二值化处理，生成二值化图像

； Step 2: Put

Perform binarization processing to generate a binarized image

;

分割为

的模块，记为

； Step 3: Binarize the image

divided into

module, denoted as

;

进行形态学连通区域检测和滤波，去除噪声；如果该区域内0的个数超过整体像素数的85%，则此块图像的背景稳定，进行步骤5；否则，进行步骤6； Step 4: Put each piece

Perform morphological connected region detection and filtering to remove noise; if the number of 0s in this region exceeds 85% of the overall number of pixels, then the background of this block image is stable, and proceed to step 5; otherwise, proceed to step 6;

Step 5: generate a background image model of the block image;

，否则，进行步骤1。 Step 6: Determine whether the number of generated background modules is greater than 90% of the total number of modules. If it is greater than 90%, the initialization of the background model is completed and the background image model is generated.

, otherwise, go to step 1.

The DSP-based moving target detection and tracking method that the present invention provides comprises the following steps:

A. For the Y component of the image, the foreground moving target is detected by a method combining inter-frame difference and background difference;

B. Establish a target template through the foreground moving target and determine the area to be tracked in the next frame;

C. Pyramid sampling the target template and the area to be tracked;

D. Track moving targets through particle swarm optimization algorithm and diamond search algorithm;

E. Constantly update the target template to achieve real-time tracking of moving targets.

The preferred solution of the aforementioned moving target detection and tracking method is that the specific steps of the step A are as follows:

A1, for the Y component of the image, the background model is established by the method of inter-frame difference and background difference modeling;

A2, subtracting the Y component of the current frame image from the Y component of the background image to obtain a difference image of the Y component of the image;

A3, performing binarization processing on the difference image of the Y component obtained by A2;

A4, the binarized image of the Y component that A3 obtains carries out connected region detection and filtering, removes noise;

A5. Project the binarized image of the Y component obtained in A4 to the x-axis and the y-axis respectively, and find out the foreground moving target according to the set threshold.

的区域作为目标模板，在下一帧图像的Y分量中以上一帧图像测得的运动目标的形心为中心，选取

的区域作为待跟踪区域。 The aforementioned moving target detection and tracking method, the preferred solution is that the specific steps of the step B are as follows: take the centroid of the moving target as the center in the Y component of the current frame image, and select

The region of the image is used as the target template, and in the Y component of the next frame image, the centroid of the moving target measured in the previous frame image is taken as the center, and the

area as the area to be tracked.

The preferred solution of the aforementioned moving target detection and tracking method is that the specific steps of step C are as follows:

C1, carry out down-sampling processing to target template, obtain the middle layer pyramid image of target template; Then carry out down-sampling processing to the middle layer pyramid of target template, obtain the top layer pyramid of target template;

C2. Perform down-sampling processing on the area to be tracked to obtain a middle-level pyramid of the area to be tracked, and then perform down-sampling processing on the middle-level pyramid of the area to be tracked to obtain a top-level pyramid of the area to be tracked.

The preferred solution of the aforementioned moving target detection and tracking method is that the specific steps of the step D are as follows:

D1, carry out matching operation to target template and the top pyramid of area to be tracked, combine PSO algorithm in the process of searching, obtain the position of moving target in the top pyramid of area to be tracked by the interaction between particles, correlation matching used in the present invention The function is the minimum mean absolute difference function (MAD), using this function can effectively reduce the amount of calculation;

D2, according to the pixel position of the matching area obtained by D1, search and match in the middle layer image of the area to be tracked, the searched area is near the position where D1 is matched and positioned, and the search method used in the search process is the diamond search method;

D3, according to the position determined by D2, search and match in the original layer image of the area to be tracked, the method used in the search process is the diamond search method, through this step, search for the position of the moving target in this frame;

D4. According to the position determined in D3, mark the target position on the image frame, and mark the search range of the next frame at the same time.

The preferred solution of the aforementioned moving target detection and tracking method is that the specific steps of the step E are as follows:

E1, according to the position determined by D4, update the scope of the target template and the area to be tracked in the next frame image;

E2, judging whether the tracking is over, if not, returning to B1 to track the next frame.

The method for detecting and tracking a moving target provided by the present invention includes: collecting a video image sequence, and combining the Y component of the image sequence with inter-frame difference and inter-frame difference to establish a background image under the Y component; background difference between the Y component of the background image and the Y component of the background image to obtain the difference image of the Y component of the current frame image; then binarize the Y component of the obtained image, and then perform connected region detection and filtering on it to remove noise, and then Project the image to the x-axis and y-axis respectively, and find out the foreground moving target according to the set threshold.

的区域作为目标模板，在下一帧图像的Y分量中以上一帧图像测得的运动目标的形心为中心，选取的区域作为待跟踪区域。 The moving object detection and tracking method provided by the present invention includes: taking the centroid of the foreground moving object found as the center, and selecting

The area of the image is used as the target template, and in the Y component of the next frame image, the centroid of the moving target measured in the previous frame image is the center, and the area as the area to be tracked.

Then carry out pyramid sampling on the target template and the area to be tracked, and the specific steps are as follows: down-sampling the target template to obtain the pyramid image of the middle layer of the target template; then down-sampling the middle layer pyramid of the target template to obtain the Top-level pyramid: perform down-sampling processing on the area to be tracked to obtain the middle-level pyramid of the area to be tracked, and then perform down-sampling processing on the middle-level pyramid of the area to be tracked to obtain the top-level pyramid of the area to be tracked.

The matching operation is performed on the target template and the top pyramid of the area to be tracked, and the PSO algorithm is combined in the search process to obtain the position of the moving target in the top pyramid of the area to be tracked through the interaction between particles. Using the PSO algorithm can optimize the search process, and effectively find the best matching area of the target template in the area to be tracked.

为惯性权重，

为粒子的速度，为学习因子，其取值为

，

和

是（0,1）之间的随机数，

代表个体极值，

代表全局极值。粒子在空间中不断学习个体极值和全局极值的经验更新粒子速度和位置。知道寻找到最优解。在顶层金子塔匹配中，粒子的搜索区域就是待搜索区域，在粒子搜过的过程中给他定义的粒子的速度为粒子的位移和方向，用

表示，其中

 表示行的位移大小和方向，表示列的位移大小和方向。本发明中使用的适应度评价函数是最小平均绝对差值函数（MAD），根据适应度评价函数更新个体极值和群体极值，通过该方法可以有效快速的在待跟踪区域的顶层金字塔中寻找到运动目标。 in,

is the inertia weight,

is the velocity of the particle, is the learning factor, and its value is

,

and

is a random number between (0,1),

represents the individual extremum,

represents the global extremum. Particles continuously learn the experience of individual extremum and global extremum in space to update particle velocity and position. know to find the optimal solution. In the top-level pyramid matching, the search area of the particle is the area to be searched, and the velocity of the particle defined for it during the search process of the particle is the displacement and direction of the particle.

said, among them

Indicates the displacement magnitude and direction of the row, Indicates the displacement magnitude and direction of the column. The fitness evaluation function used in the present invention is the minimum mean absolute difference function (MAD), and the individual extremum and the group extremum are updated according to the fitness evaluation function. Through this method, the top pyramid of the area to be tracked can be effectively and quickly searched for to exercise goals.

According to the pixel position of the matching area obtained, search and match are carried out in the middle layer image of the area to be tracked, and the top layer of the searched area is located near the matching location. The search method used in the search process is the diamond search method; Search and match in the original layer image of the area. The method used in the search process is the diamond search method. After this step, the position of the moving target in this frame is searched; the target position is marked on the frame image, and at the same time the The search range for the next frame.

According to the marked target position, update the target template and the scope of the area to be tracked in the next frame image, and then judge whether the tracking is over, if not, start the tracking of the next frame.

The invention adopts a template-based tracking method, and in order to reduce the complexity of calculation, proposes a template matching method based on pyramid multi-resolution combined with particle swarm optimization algorithm and diamond search algorithm. Pyramid multi-resolution refers to shrinking the image by reducing the number of pixels in the image, and then matching and searching the reduced target template with the target search area. Secondary image reduction, the number of image reductions depends on the size of the target template image and the image of the region to be searched. If the reduced images are arranged up and down according to the resolution, a pyramid structure is formed. In this pyramid, the lower the resolution of the image at the upper layer, the fewer the number of pixels of the image.

The method provided by the present invention first establishes a background model for the Y component of the image using the method of combining inter-frame difference and background difference, and then detects the foreground moving target; The image is subjected to pyramid downsampling twice to reduce the resolution of the target template and the video image to be tracked. The particle swarm optimization algorithm is used to locate the tracking target on the top pyramid, and the diamond search method is used to locate the tracking target on the middle and bottom pyramids. In the process of target tracking, the target template is continuously updated with the change of the moving target, realizing real-time continuous tracking of the target.

Compared with the prior art, the technical advantages of the present invention are also reflected in:

1. The improved background modeling method for the Y component in the input image proposed by the present invention can establish a background model on the Y component of the image through 9 frames. If the background image is generated after calculating the mean value, it needs 30 frames to generate better background mockup.

2. The present invention introduces pyramid multi-resolution into the detection and tracking method of moving targets, effectively reducing the resolution of target templates and video images to be tracked.

3, the present invention uses particle swarm optimization algorithm in the moving target detection and tracking system based on DSP for the first time, uses this method to locate the moving target quickly and effectively in the top pyramid, and its calculation amount is ordinary in the specific implementation of the present invention 90% of the algorithm, if the area to be tracked is larger, the calculation amount will be reduced more obviously.

4. The present invention uses the diamond search method in the DSP-based moving target detection and tracking system. This method can be used to quickly locate and track the target in the middle layer and the bottom pyramid. In the specific implementation of the present invention, the overall calculation amount is a common template 30% of the matching method, the correct rate can reach 92% under stable light conditions.

Description of drawings

Figure 1: Structural diagram of the image processing system.

Figure 2: Initializing the background model.

Figure 3: Flow chart of moving target detection and tracking algorithm.

Detailed ways

The technical solutions of the present invention will be described in detail below in conjunction with the embodiments and drawings, but the scope of protection is not limited thereto.

Embodiments of a real-time DSP-based detection and tracking method for a moving target and an image processing system for realizing the method, thereby realizing real-time detection and tracking of a moving target.

Figure 1 is a structural diagram of a DSP-based moving target detection and tracking system (image processing system). As shown in Figure 1, the image processing system includes three parts: video acquisition module, video processing module and display module. In the present invention, the video signal is collected by a CCD camera, the analog video signal collected is transmitted to the SEED VPM642 video processing module, and the analog video signal is converted into the video signal of the BT.656 format by the high-performance video decoder TVP5150 in the VPM642, and Transmit this signal to the video interface of DSP. The video interface of the DSP combines the EDMA channel to transmit the video signal to the SDRAM buffer area. After the data is processed by the DM642, the video encoder SAA7121H converts the data into an analog signal and sends it to the display. The effect of target tracking can be seen through the display.

The working process of the DSP-based moving target detection and tracking system proposed by the present invention mainly includes three parts: DM642 initialization, system driver initialization, and moving target detection and tracking. DM642 initialization mainly includes chip internal memory interface initialization, peripheral device initialization, interrupt initialization, etc.; system driver initialization includes video codec initialization, DM642 video port initialization, EDMA channel initialization, etc. After the system initialization is completed, DM642 no longer intervenes in the input and output of video signals, and the system enters an infinite loop stage. In this stage, DM642 is mainly used to realize the detection and tracking of moving targets.

The present invention proposes an improved background modeling method aimed at the Y component in the collected image, and its algorithm steps are as shown in Figure 2:

，其差分图像分别记为，其中，其三帧差分图像记为

（此处&代表逻辑与运算）。 Step 1: First input a frame of image, determine whether the number of frames is greater than 3, and when the number of frames is greater than 3, perform three consecutive frames of difference on the Y component of the input image, and binarize the image after the difference. Assume that the input three frames of images are

, and their difference images are denoted as ,in , and its three-frame differential image is denoted as

(Here & stands for logical AND operation).

进行二值化处理，生成二值化图像

。 Step 2: Put

Perform binarization processing to generate a binarized image

.

分割为的模块，记为

。 Step 3: Binarize the image

divided into module, denoted as

.

进行形态学连通区域检测和滤波，去除噪声。如果该区域内0的个数超过整体像素数的85%，则此块图像的背景稳定转入5 )，否则，转入6 )。 Step 4: Put each piece

Perform morphological connected region detection and filtering to remove noise. If the number of 0 in this area exceeds 85% of the overall pixel number, then the background of this block image is stable and turns to 5), otherwise, turns to 6).

Step 6: Generate a background image model of the block image.

，否则，转入1 )。 Step 7: Determine whether the number of generated background modules is greater than 90% of the total number of modules. If it is greater than 90%, the initialization of the background model is completed and the background image model is generated.

, otherwise, go to 1).

Fig. 3 is the moving target detection and tracking method based on DSP that the present invention proposes, and this scheme comprises the steps:

与背景模型

进行差分得到灰度图像，然后将得到的灰度图像按一定的阈值二值化生成二值化图像，对该图像进行形态学连通区域检测和滤波，去除噪声。即

，其中

为阈值)。 Step A: For the Y component of the image, the foreground moving target is detected by combining the inter-frame difference and the background difference (by combining the current frame

with the background model

The gray-scale image is obtained by difference, and then the obtained gray-scale image is binarized according to a certain threshold to generate a binary image, and the morphological connected region detection and filtering are performed on the image to remove noise. Right now

,in

is the threshold).

The obtained binarized image is projected to the x-axis and y-axis respectively, and the foreground moving target is found by the gray histogram method.

In summary, the specific steps of step A are as follows:

A1, for the Y component of the image, the background model is established by the method of inter-frame difference and background difference modeling;

A2, subtracting the Y component of the current frame image from the Y component of the background image to obtain a difference image of the Y component of the image;

A3, performing binarization processing on the difference image of the Y component obtained by A2;

A4, the binarized image of the Y component that A3 obtains carries out connected region detection and filtering, removes noise;

A5. Project the binarized image of the Y component obtained in A4 to the x-axis and the y-axis respectively, and find out the foreground moving target according to the set threshold.

的区域作为待跟踪区域)。 Step B: Establish a target template through the foreground moving target and determine the area to be tracked in the next frame (taking the centroid of the foreground moving target found as the center, select The region of the image is used as the target template, and in the Y component of the next frame image, the centroid of the moving target measured in the previous frame image is taken as the center, and the

area as the area to be tracked).

的区域作为待跟踪区域。 To sum up, the specific steps of step B are as follows: take the centroid of the moving target as the center in the Y component of the current frame image, and select The region of the image is used as the target template, and in the Y component of the next frame image, the centroid of the moving target measured in the previous frame image is taken as the center, and the

area as the area to be tracked.

Step C: Pyramid sampling the target template and the area to be tracked, the specific steps are as follows:

Downsampling is performed on the target template to obtain the middle layer pyramid image of the target template; then downsampling is performed on the middle layer pyramid of the target template to obtain the top pyramid of the target template; downsampling is performed on the area to be tracked to obtain the image of the area to be tracked The middle layer pyramid, and then the middle layer pyramid of the area to be tracked is down-sampled to obtain the top layer pyramid of the area to be tracked.

Step D: track the moving target through the particle swarm optimization algorithm and the diamond search algorithm (matching the target template and the top pyramid of the area to be tracked, combining the PSO algorithm during the search process, and obtaining the top layer of the area to be tracked through the interaction between particles The position of the moving target in the pyramid. Using the PSO algorithm can optimize the search process and effectively find the best matching area of the target template in the area to be tracked).

为惯性权重，

为粒子的速度，

为学习因子，其取值为

，

和是（0,1）之间的随机数，

代表个体极值，

代表全局极值。粒子在空间中不断学习个体极值和全局极值的经验更新粒子速度和位置。知道寻找到最优解。在顶层金子塔匹配中，粒子的搜索区域就是待搜索区域，在粒子搜过的过程中给他定义的粒子的速度为粒子的位移和方向，用

表示，其中

 表示行的位移大小和方向，表示列的位移大小和方向。本发明中使用的适应度评价函数是最小平均绝对差值函数（MAD），根据适应度评价函数更新个体极值和群体极值，通过该方法可以有效快速的在待跟踪区域的顶层金字塔中寻找到运动目标。 in,

is the inertia weight,

is the velocity of the particle,

is the learning factor, and its value is

,

and is a random number between (0,1),

represents the individual extremum,

represents the global extremum. Particles continuously learn the experience of individual extremum and global extremum in space to update particle velocity and position. know to find the optimal solution. In the top-level pyramid matching, the search area of the particle is the area to be searched, and the velocity of the particle defined for it during the search process of the particle is the displacement and direction of the particle.

said, among them

Indicates the displacement magnitude and direction of the row, Indicates the displacement magnitude and direction of the column. The fitness evaluation function used in the present invention is the minimum mean absolute difference function (MAD), and the individual extremum and the group extremum are updated according to the fitness evaluation function. Through this method, the top pyramid of the area to be tracked can be effectively and quickly searched for to exercise goals.

According to the pixel position of the matching area obtained, search and match are carried out in the middle layer image of the area to be tracked, and the top layer of the searched area is located near the matching location. The search method used in the search process is the diamond search method; Search and match in the original layer image of the area. The method used in the search process is the diamond search method. After this step, the position of the moving target in this frame is searched; the target position is marked on the frame image, and at the same time the The search range for the next frame.

Generally speaking, the specific steps of the step D are as follows:

D1, carry out matching operation to target template and the top pyramid of area to be tracked, combine PSO algorithm in the process of searching, obtain the position of moving target in the top pyramid of area to be tracked by the interaction between particles, correlation matching used in the present invention The function is the minimum mean absolute difference function (MAD), using this function can effectively reduce the amount of calculation;

D2, according to the pixel position of the matching area obtained by D1, search and match in the middle layer image of the area to be tracked, the searched area is near the position where D1 is matched and positioned, and the search method used in the search process is the diamond search method;

D3, according to the position determined by D2, search and match in the original layer image of the area to be tracked, the method used in the search process is the diamond search method, through this step, search for the position of the moving target in this frame;

D4. According to the position determined in D3, mark the target position on the image frame, and mark the search range of the next frame at the same time.

Step: E: Constantly update the target template to achieve real-time tracking of the moving target (according to the marked target position, update the target template and the range of the area to be tracked in the next frame of image, then judge whether the tracking is over, if not, return to The second step is to track the next frame).

Generally speaking, the specific steps of the step E are as follows:

E1, according to the position determined by D4, update the scope of the target template and the area to be tracked in the next frame image;

E2, judging whether the tracking is over, if not, returning to B1 to track the next frame.

Claims (8)

1. The moving target detection and tracking system based on the DSP is characterized by comprising a video acquisition module, a video processing module and a display module; the video acquisition module acquires video signals through one path of CCD camera, transmits the acquired analog video signals to the SEED VPM642 video processing module, converts the analog video signals into video signals in a BT.656 format through a high-performance video decoder TVP5150 in the VPM642 and transmits the signals to a video interface of the DSP; the video interface of the DSP combines with an EDMA channel to transmit the video signal to a buffer area of SDRAM, after the data is processed by DM642, the data is converted into an analog signal by a video encoder SAA7121H and transmitted to a display of a display module, and the foreground moving object is displayed by the display.

2. An improved background modeling method for a Y component in an acquired image, comprising the steps of:

step 1: firstly, inputting a frame image, judging whether the frame number is greater than 3, carrying out continuous three-frame difference on a Y component of the input image when the frame number is greater than 3, and carrying out binarization on the image after difference; suppose that the input three frames of images are respectively

Their difference images are respectively notedWherein

And the three-frame differential image is recorded as

Here, the&Represents a logical AND operation;

step 2: will be provided withPerforming binarization processing to generate a binarized image

；

And step 3: will binarize the image

Is divided into

Module of

；

And 4, step 4: each block is divided into

Performing morphological connected region detection and filtering to remove noise; if the number of 0 in the area exceeds 85 percent of the whole pixel number, the background of the block image is stable, and the step 5 is carried out; otherwise, go to step 6;

and 5: generating a background image model of the block image;

step 6: judging whether the number of the generated background modules is more than 90% of the number of the whole modules, if so, finishing the initialization of the background model, and generating a background image model

Otherwise, performing step 1.

3. A moving target detection and tracking method based on DSP is characterized by comprising the following steps:

A. detecting a foreground moving target by combining interframe difference and background difference aiming at the Y component of the image;

B. establishing a target template through the foreground moving target and simultaneously determining the area to be tracked of the next frame;

C. carrying out pyramid sampling on the target template and the area to be tracked;

D. tracking a moving target by a particle swarm optimization algorithm and a diamond search algorithm;

E. and continuously updating the target template to realize real-time tracking of the moving target.

4. The moving object detecting and tracking method according to claim 3, wherein the step A specifically comprises the following steps:

a1, establishing a background model by an interframe difference and background difference modeling method aiming at the Y component of the image;

a2, subtracting the Y component of the current frame image from the Y component of the background image to obtain a difference image of the Y component of the image;

a3, performing binarization processing on the difference image of the Y component obtained by the A2;

a4, detecting and filtering a connected region of the binary image of the Y component obtained by A3, and removing noise;

and A5, projecting the binary image of the Y component obtained by the A4 to the x axis and the Y axis respectively, and finding out the foreground moving target according to a set threshold value.

5. The moving object detecting and tracking method according to claim 3, wherein the step B comprises the following steps: selecting the Y component of the current frame image by taking the centroid of the moving object as the center

The area of the image is used as a target template, the centroid of the moving target measured by the previous frame of image in the Y component of the next frame of image is taken as the center, and the centroid of the moving target is selected

The area of (2) is used as the area to be tracked.

6. The moving object detecting and tracking method according to claim 3, wherein the step C comprises the following steps:

c1, performing down-sampling processing on the target template to obtain a pyramid image in the middle layer of the target template; then, performing down-sampling processing on the middle pyramid of the target template to obtain a top pyramid of the target template;

and C2, performing down-sampling processing on the region to be tracked to obtain a middle-layer pyramid of the region to be tracked, and then performing down-sampling processing on the middle-layer pyramid of the region to be tracked to obtain a top-layer pyramid of the region to be tracked.

7. The moving object detecting and tracking method according to claim 3, wherein the step D comprises the following specific steps:

d1, matching the target template and the top pyramid of the region to be tracked, and combining a PSO algorithm in the searching process to obtain the position of the moving target in the top pyramid of the region to be tracked through the interaction between particles;

d2, searching and matching in the middle layer image of the region to be tracked according to the pixel position of the matching region obtained by the D1, wherein the searched region is near the position where the D1 is matched and positioned, and the searching method used in the searching process is a diamond searching method;

d3, searching and matching in the original layer image of the area to be tracked according to the position determined by D2, wherein the method used in the searching process is a diamond searching method, and the position of the moving object in the frame is searched through the step;

and D4, marking the target position on the image of the frame according to the position determined by the D3, and marking the search range of the next frame.

8. The moving object detecting and tracking method according to claim 3, wherein the step E comprises the following specific steps:

e1, updating the range of the target template and the area to be tracked in the next frame image according to the position determined by the D4;

e2, judging whether the tracking is finished or not, if not, returning to B1 to track the next frame.

Images