A target tracking detection method, system, medium, device and terminal

Abstract

The application belongs to the technical field of moving target detection in static background, and discloses a target tracking detection method, system, medium, equipment and terminal, and an intelligent security system inputs a target tracking detection algorithm, and a monitoring camera selectively inputs an RGB or grayscale image to the system; the system transmits the RGB or grayscale image into a calculation characteristic difference value model, establishes a new model for the current input picture, thereby recording the extracted background characteristics; for each frame of the monitored picture, a characteristic difference value model is obtained; the system identifies the changed area according to the built-in threshold setting for the model, divides all the pixel points into background points and foreground points according to the output information of the model, and then draws the contour of the moving target in the grayscale image or tricolor image according to the requirement. The output result of the application is more consistent with the reality, so that the intelligent security detection algorithm is free from the influence of complex modeling, illumination change and low computing power of the edge equipment, and the intelligent security is realized.

Description

Technical Field

[0001] This invention belongs to the field of moving target detection technology under static background, and particularly relates to a target tracking and detection method, system, medium, device and terminal. Background Technology

[0002] Currently, there are many surveillance algorithms available for security purposes, including traditional inter-frame difference methods and background modeling methods, as well as methods that use deep learning to detect moving targets. However, the former requires the use of information from multiple frames or the establishment of a complex model that needs to be updated in real time, while the latter relies on the massive computing power of computers. These shortcomings are pain points in the field of intelligent security. The algorithm's output is not rapid, meaning it cannot simultaneously achieve both real-time performance and accuracy. While the latter can achieve the above, most edge devices often do not have the computing power to support it. In addition, traditional detection algorithms often cannot cope with changes in lighting, such as ghost areas that often appear in object detection.

[0003] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0004] (1) Traditional inter-frame difference method must utilize information from multiple frames, and it extracts the boundary of the target, which has a relatively coarse outline. When the target moves quickly, it will produce a ghosting effect, and when the target moves slowly, it will fail to detect the target due to overlap.

[0005] (2) Background modeling method, such as the ViBe algorithm, has the problem that the detected moving target is incomplete, that is, there are holes inside the target. This is usually caused by instantaneous changes in lighting that make it impossible for the model to update in time.

[0006] (3) Although deep learning-based target tracking algorithms are fast, they require computing power, which most edge devices often cannot support. Summary of the Invention

[0007] In view of the problems existing in the current intelligent security technology, the present invention provides a target tracking and detection method, system, medium, device and terminal, and particularly relates to a target tracking and detection method, system, medium, device and terminal based on the change of pixel neighborhood difference.

[0008] This invention is implemented as follows: a target tracking and detection method, comprising: after the intelligent security system accesses the target tracking and detection algorithm, the monitoring camera selectively inputs RGB or grayscale images to the system; the system inputs the RGB or grayscale images into a black box for feature difference model calculation, establishes a new model for the currently input image, and records the extracted background features; for each frame of the monitored image, a feature difference model is obtained; since the features between the background points and the internal points of the target and their neighboring pixels remain relatively unchanged during the target's movement, and only the features of the neighboring pixels at the boundary between the target's outline and the background change continuously with the target's movement, for each frame of the image, it is only necessary to determine whether its features have changed, and then identify the changed areas based on the threshold set built into the model, divide all pixels into background points and foreground points according to the model's output information, and then draw the outline of the moving target in grayscale or three-color image as needed.

[0009] Furthermore, the target tracking and detection method includes the following steps:

[0010] Step 1: In order to model the background, the system performs pixel-level learning on the current monitoring screen. Each point generates a set of feature differences based on its neighboring pixels. The feature differences are the correlation between pixels, thus obtaining the feature difference model. The feature difference model is used to record the background information under the current monitoring screen.

[0011] Step 2: In order to extract foreground points and eliminate the influence of noise, when a moving target enters the monitoring screen, the input of the algorithm changes. The system marks the pixels in the model that have changed and determines whether they are foreground points or noise based on the area size or context information.

[0012] Step 3: In order to output the contour information of the moving target, the computer preprocesses all the marked areas. After eliminating noise through preprocessing, all pixels are divided into background points and foreground points. Finally, the moving target is outlined on a colored background or a grayscale background as needed.

[0013] Furthermore, in step one, pixel-level learning is performed on the current monitoring screen. Each point generates a set of feature differences based on its neighboring pixels. These feature differences represent the correlation between pixels, thus obtaining a feature difference model. The feature difference model is used to record background information under the current monitoring screen, including:

[0014] (1) For any pixel, let its eight surrounding pixels β be the neighborhood, and obtain a one-dimensional feature vector α:

[0015] α(β1,β2,β3,β4,β5,β6,β7,β8);

[0016] (2) Let G(x) and F(x) be the difference calculation function and the normalization function, respectively. The normalized feature difference S = F(G(α)) is obtained from the one-dimensional feature vector. The feature extraction process of deep learning is simulated by mathematical operation.

[0017] (3) Establish a feature difference model for the current frame, where h and w are the height and width, respectively;

[0018]

[0019] The feature difference model records the correlation between pixels mapped by the difference in neighboring pixel values.

[0020] Furthermore, the normalization function in step (2) is the sigmoid function.

[0021] Furthermore, in step two, when a moving target enters the monitoring screen, the algorithm's input changes. The system marks the pixels where the model changes, and determines whether they are foreground points or noise points based on their area or context information, including:

[0022] (1) The feature difference model M is obtained by performing the above processing on the images transmitted from the monitoring screen:

[0023] M:

[0024]

[0025] (2) Compare the information contained in the difference model and determine whether the features have changed according to the threshold function. Let the threshold function be V(x), then the output of the second step of the system is V(M). The threshold in the function V is adjusted according to the actual situation.

[0026] Furthermore, in step three, the computer preprocesses all marked areas. After eliminating noise through preprocessing, all pixels are divided into background pixels and foreground pixels. Finally, the moving target on a colored background or a grayscale background is outlined as needed, including:

[0027] (1) Preprocessing treats smaller areas as noise points, or judges them based on context information. If the feature difference model of pixels in neighboring frames has not changed, they are considered as background points.

[0028] (2) For all pixels in the current image, the function V(M) is regarded as a classifier - a set of fixed points A and a set of moving points B. Since the environment of the algorithm is a static background, the contour of the moving target is obtained after the pixel of the set of moving points B is assigned a value. The feature extraction operation of deep learning is simulated through various mathematical operations to obtain the foreground and background classifier, and then the target tracking and detection is realized.

[0029] Another object of the present invention is to provide a target tracking and detection system applying the aforementioned target tracking and detection method, the target tracking and detection system comprising:

[0030] The feature difference model construction module is used to perform pixel-level learning on the current monitoring screen. Each point generates a set of feature differences based on its neighboring pixels. The feature difference is the correlation between pixels, thus obtaining the feature difference model. The feature difference model is used to record the background information under the current monitoring screen.

[0031] The pixel marking module is used to mark pixels whose feature difference model changes when a moving target enters the monitoring screen, thus changing the input of the algorithm. The system then determines whether the area is a foreground point or noise based on the area size or context information.

[0032] The moving target delineation module uses a computer to preprocess all marked areas. After eliminating noise through preprocessing, all pixels are divided into background points and foreground points, and finally the moving target is delineated on a colored background or a grayscale background as needed.

[0033] Another object of the present invention is to provide a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the target tracking and detection method.

[0034] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the target tracking and detection method.

[0035] Another object of the present invention is to provide an information data processing terminal for implementing the target tracking and detection system described above.

[0036] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0037] First, addressing the technical problems existing in the prior art and the difficulty of solving them, this paper closely analyzes, in conjunction with the technical solution to be protected by this invention and the results and data obtained during the research and development process, how the technical solution of this invention solves the technical problems, and the inventive technical effects brought about by solving these problems. The specific description is as follows:

[0038] The target tracking and detection method based on pixel neighborhood difference changes provided by this invention utilizes the fact that pixels are the basic building blocks of a surveillance image. Therefore, starting with the pixels allows for maximum utilization of the information contained in the image, enabling the tracking of moving targets within the monitored area. Furthermore, the input of this invention is not limited to grayscale images; a three-channel RGB image can be directly input to obtain information about moving targets against a static real-world background. The output colors are more realistic, thus achieving intelligent security.

[0039] Second, considering the technical solution as a whole or from a product perspective, the technical effects and advantages of the technical solution to be protected by this invention are specifically described as follows:

[0040] The target tracking and detection method provided by this invention is based on the change of pixel neighborhood difference, which solves the problems of complex modeling, long update time and accuracy drop when the brightness of the monitoring screen changes in traditional target tracking algorithms in intelligent security scenarios. This allows intelligent security detection algorithms to get rid of the limitations of complex modeling, the influence of sudden changes in lighting and the low computing power of edge devices.

[0041] Third, as supplementary evidence of the inventive step of the claims of this invention, it is also reflected in the following important aspects:

[0042] (1) The technical solution of this invention fills a technical gap in the industry both domestically and internationally:

[0043] This invention combines the advantages of inter-frame difference method and ViBe background modeling algorithm. For inter-frame difference method, this algorithm breaks through the barrier that the difference method requires subtraction between two or more frames, and only requires one frame as input. This is because this algorithm constructs a feature difference model for the neighborhood of the pixels in the current frame. According to the core theory of this algorithm, during the movement of the target, the correlation between the background points and the internal points of the target and their neighborhood pixels remains relatively unchanged. Only the features of the neighborhood pixels at the boundary between the target contour and the background change with the movement of the target. Therefore, for each frame of image, it is only necessary to determine whether its features have changed, without comparing with the previous and next frames, to obtain the contour information of the target, thereby completing the target detection.

[0044] For the ViBe background modeling algorithm, the model established by this algorithm is based on the difference between neighboring pixels, rather than the simple mean. This is because neighboring pixels have spatial characteristics, that is, the pixel values ​​of nearby pixels are close and increase or decrease synchronously under the influence of illumination. The difference operation can cancel out the influence of illumination on the image, because the increase or decrease of their pixel values ​​is synchronous and the difference is not affected by illumination. Therefore, the feature difference model of this invention is simple to model and updates quickly.

[0045] (2) The technical solution of the present invention solves a technical problem that people have long wanted to solve but have never been able to solve successfully:

[0046] In the field of traditional image processing, noise caused by changes in illumination has always been a pain point. This invention uses the difference between neighboring pixels as the input for modeling, and takes advantage of the fact that the difference in pixel values ​​between adjacent pixels is not affected by changes in illumination to overcome the impact on the accuracy of the algorithm under conditions such as local exposure. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the target tracking and detection method provided in an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of system control analysis provided in an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of the feature difference model generation process provided in the embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] To address the problems existing in the prior art, the present invention provides a target tracking and detection method, system, medium, device, and terminal. The present invention will be described in detail below with reference to the accompanying drawings.

[0053] I. Explanatory and Illustrative Embodiments. To enable those skilled in the art to fully understand how the present invention is specifically implemented, this section provides an explanatory and illustrative description of the embodiments described in the claims.

[0054] The target tracking and detection method provided in this invention embodiment is implemented as follows: After the intelligent security system accesses the target tracking and detection algorithm, the monitoring camera selectively inputs RGB (three-channel color image) or grayscale image to the system. The system inputs it into the feature difference model, which establishes a simple model for the currently input image that does not require real-time updates, thereby recording the extracted background features. In this way, for each frame of the monitored image, this invention will obtain unique features, called the feature difference model. The system will identify the changed areas based on the built-in threshold setting of the model. Since the mathematical calculation is pixel-based, the accuracy is at the pixel level. Finally, based on the output information of the model, all pixels are divided into two categories, namely background points and foreground points, and the outline of the moving target under the grayscale image or three-color image is drawn as needed.

[0055] like Figure 1 As shown, the target tracking and detection method provided in this embodiment of the invention includes the following steps:

[0056] S101, after the intelligent security system integrates the target tracking and detection algorithm, the surveillance camera selectively inputs RGB or grayscale images to the system;

[0057] S102, the system inputs the RGB or grayscale image into the feature difference model to build a new model for the current input image, thereby recording the extracted background features; for each frame of the monitored image, a feature difference model is obtained;

[0058] S103, the system identifies the changed areas of the model based on the built-in threshold settings, divides all pixels into background points and foreground points according to the model's output information, and then draws the outline of the moving target in grayscale or tri-color image as needed.

[0059] As a preferred embodiment, the target tracking and detection method provided by this invention specifically includes the following steps:

[0060] Step 1: Perform pixel-level learning on the current monitoring screen. Each point generates a set of feature differences based on its neighboring pixels. These feature differences represent the correlation between pixels, thus obtaining a feature difference model. This model records the background information of the current monitoring screen and does not need to be updated in real time. Since the model is built based on the difference calculation between pixels, even if the pixels change, such as when the image is overexposed, this difference calculation can cancel out the impact.

[0061] Step 2: When a moving target enters the monitoring screen, the input of the algorithm will change, so the output of its feature difference model will also be different. That is, the feature difference model of some pixels will no longer be uniform. The system will mark the pixels whose models have changed. At this time, these areas may be foreground points or noise points. This invention can make a judgment based on their area size or their context information. Pixels whose models have not changed can be considered as background points or internal points of moving targets.

[0062] Step 3: The computer performs a series of preprocessing steps on all marked areas. After eliminating noise through these preprocessing steps, the algorithm divides all pixels into three categories: foreground points, target interior points, and background points. According to the core theory of this algorithm, the points where the correlation between pixels has not changed can only be background points or target interior points. As the target moves, the pixel correlation changes only at the outline of the target. Finally, the moving target is outlined on a colored background or a grayscale background as needed.

[0063] Step 1 provided in this embodiment of the invention specifically includes:

[0064] (1) For any pixel, let its eight surrounding pixels β (the pixel value of a pixel in the image, ranging from 0 to 255) be its neighborhood, thus obtaining a one-dimensional feature vector α:

[0065] α(β1,β2,β3,β4,β5,β6,β7,β8)

[0066] (2) Let G(x) and F(x) be the difference calculation function and the normalization function, such as the sigmoid function, respectively. Therefore, based on the above one-dimensional feature vector, the present invention can obtain the normalized feature difference S = F(G(α)), thereby simulating the feature extraction process of deep learning through this mathematical operation.

[0067] (3) Establish a feature difference model for the current frame, where h and w are the height and width, respectively;

[0068]

[0069] This model records information mapped by the difference in neighboring pixel values, i.e., the correlation between pixels. It can not only overcome the limitations brought about by changes in lighting, but also has unique features for each frame of the image, i.e., pixel-level "image fingerprints".

[0070] Step 2 provided in this embodiment of the invention specifically includes:

[0071] (1) The feature difference model M is obtained by performing the above processing on the images transmitted from the monitoring screen:

[0072] M:

[0073]

[0074] (2) Make a judgment based on the information contained in the difference model. Let the built-in threshold function of the algorithm be V(x), then the input of the second step of the system is V(M). At this time, the threshold function regards the pixels within a certain range of change as unchanged. According to the above theory, these unchanged points are either background points or internal points of the target and are not drawn. It is only necessary to input the set of changed pixels into step 3 for drawing. The threshold in function V can be adjusted according to the actual situation.

[0075] Step 3 provided in this embodiment of the invention specifically includes:

[0076] (1) Preprocessing treats smaller areas as noise points, or makes judgments based on contextual information, for example, if the feature difference model of the pixel does not change in the neighboring frames, then there is reason to consider it as a background point.

[0077] (2) Based on the above steps, for all pixels in the current image, the present invention can regard the function V(M) as a classifier, namely, the set of fixed points A and the set of moving points B. Since the environment of this algorithm is a static background, the contour of the moving target can be obtained after assigning a value (e.g., 255) to the pixels of the set of moving points B. Thus, the "feature extraction" operation of deep learning is simulated through the above mathematical operations, thereby obtaining a foreground and background classifier, and realizing target tracking and detection in a simple and efficient manner.

[0078] The target tracking and detection system provided in this embodiment of the invention includes:

[0079] The feature difference model construction module is used to perform pixel-level learning on the current monitoring screen. Each point generates a set of feature differences based on its neighboring pixels. The feature difference is the correlation between pixels, thus obtaining the feature difference model. The feature difference model is used to record the background information under the current monitoring screen.

[0080] The pixel marking module is used to mark the pixels in the model that have changed when a moving target enters the monitoring screen, and to determine whether the area is a foreground point or noise based on the area size or context information.

[0081] The moving target delineation module uses a computer to preprocess all marked areas. After eliminating noise through preprocessing, all pixels are divided into background points and foreground points, and finally the moving target is delineated on a colored background or a grayscale background as needed.

[0082] II. Application Examples. To demonstrate the inventiveness and technical value of the technical solution of this invention, this section provides application examples of the technical solution of the claims on specific products or related technologies.

[0083] The reason this invention uses neighboring pixels for interpolation is that when the image's illumination changes, due to the spatial characteristics of pixels—neighborhood pixels change synchronously—the interpolation operation can offset the increase or decrease in pixel values ​​caused by illumination, unlike other modeling algorithms that use mean calculation. This effectively reduces the limitations of illumination on the algorithm. Furthermore, the foreground / background point separation method used in this invention determines whether the feature difference of a pixel has changed. This is because in an image with a static background, the characteristics between background points remain relatively unchanged. Similarly, the features between pixels also remain relatively unchanged for the internal points of the target, since the pixels inside the target move with the target. Therefore, the only pixels whose features change in the image are those near the target's outline. During the target's movement, the internal edge pixels of the target and the background points are constantly changing. By fully utilizing this characteristic of target movement, the target's outline information can be obtained and drawn to detect the target.

[0084] III. Evidence of the Relevant Effects of the Embodiments. The embodiments of the present invention have achieved some positive effects during research and development or use, and indeed possess significant advantages compared to existing technologies. The following description, in conjunction with data, charts, and other materials from the experimental process, illustrates these advantages.

[0085] To address the problems existing in current intelligent security technologies, this invention proposes a target tracking and detection method based on pixel neighborhood difference changes. In each frame of the monitored image, every object is related to its surrounding objects. For example, a person is surrounded by trees, and trees are surrounded by buildings. Macroscopically, if the relative positions of the person, trees, and buildings do not change significantly, they appear motionless to the human eye. Only when the relative positions of these three objects change, such as a person moving from the left to the right of a tree, can movement be detected. Otherwise, this invention has reason to believe that the monitored frame is the same as the previous frame. Similarly, microscopically, each pixel is related to its surrounding pixels, i.e., neighboring pixels. If this relationship between a pixel and its neighboring pixels does not change... For example, background points: no matter how the target moves, the correlation between background points, such as the difference or mean, will not change. Therefore, this invention has reason to believe that the image of this frame is the same as the image of the previous frame. In a static background, the points whose correlation does not change are background points. Or, when the area of ​​the moving target is large, the correlation between the pixels inside it is also unchanged. The neighboring pixels whose correlation changes have a common property: they divide all pixels into two categories, foreground points and background points. When these pixels are connected, it is found that they form the outline of the moving target. In this way, this invention can find the basis for foreground and background separation, thereby achieving foreground and background separation in a static background.

[0086] After iterating through the pixels of each frame of the input image using the above operations, this embodiment of the invention can determine whether it contains a moving target. This embodiment of the invention fully utilizes the data contained in the basic unit of an image—the pixel—and simulates a series of operations in deep learning, such as convolution and pooling, through more efficient mathematical operations, thereby realizing a feature extractor. However, it does not require a large number of parameter variables and has no expensive learning costs. Therefore, it not only possesses the simplicity and efficiency advantages of traditional algorithms but also the accuracy advantages of deep learning algorithms, while overcoming the limitations of device computing power. Therefore, it can be described as a target tracking and detection algorithm that integrates traditional detection methods with deep learning.

[0087] Figure 2 The operation flow of the target tracking and detection method provided in the embodiments of the present invention is described.

[0088] Initially, when the system starts running, the algorithm waits for input from the monitoring system, which can be a grayscale image or an RGB three-color image. Once the algorithm receives input, it feeds it into the black box—the feature difference calculation model. The specific process of the black box is as follows: Figure 3In detail, after the algorithm calculates the feature difference model of the image, it will determine whether the feature has changed. According to the above theory, this feature is unchanged inside the background and inside the target. It only changes around the outline of the moving target. Therefore, when it changes, the point can be considered as a foreground point. Otherwise, it is a background point. If the change range in the set of foreground points exceeds a certain threshold, further processing will be carried out. After excluding noise, the remaining pixels are considered as foreground points, i.e., moving targets. Otherwise, they are background points. Finally, the drawing output result is performed.

[0089] Figure 3 Described Figure 2 The black-box operation process in the process.

[0090] When the black box receives input, whether it is an RGB three-color image or a grayscale image, it will first determine the neighborhood range of each pixel to obtain the feature vector of the neighborhood of any pixel. Then, it will calculate the feature difference according to the specific operation method mentioned above. For example, it will select the neighboring points with the farthest relative distance as a pair to calculate the difference. After obtaining the feature difference, it is necessary to normalize the feature difference for the convenience of subsequent calculations, such as using the sigmoid function. Finally, the generated feature difference model is output as the result.

[0091] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0092] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A target tracking and detection method, characterized in that, The target tracking and detection method includes: after the intelligent security system accesses the target tracking and detection algorithm, the monitoring camera selectively inputs RGB or grayscale images into the system; the system inputs the RGB or grayscale images into a black box to calculate the feature difference model and establishes a new model for the currently input image; for each frame of the monitored image, a feature difference model is obtained; for each frame of the image, it is only necessary to determine whether its features have changed, and then identify the changed areas based on the threshold set built into the model, divide all pixels into background points and foreground points according to the output information of the model, and then draw the outline of the moving target under the grayscale image or the three-color image as needed; The target tracking and detection method includes the following steps: Step 1: Perform pixel-level learning on the current monitoring screen. Each point generates a set of feature differences based on its neighboring pixels. The feature differences are the correlation between pixels, thus obtaining the feature difference model. The feature difference model is used to record the background information under the current monitoring screen. Step 2: When a moving target enters the monitoring screen, the input of the algorithm changes. The system marks the pixels in the model that have changed and determines whether they are foreground points or noise points based on their area or context information. Step 3: The computer preprocesses all marked areas, removes noise through preprocessing, divides all pixels into background and foreground pixels, and finally outlines the moving target against a colored or grayscale background as needed. In step one, pixel-level learning is performed on the current monitoring image. Each point generates a set of feature differences based on its neighboring pixels. The feature differences represent the correlation between pixels, thus obtaining a feature difference model. The feature difference model is used to record the background information of the current monitoring image, including: (1) For any pixel, let its eight surrounding pixels β be the neighborhood, and we get a one-dimensional feature vector: α(β1,β2,β3,β4,β5,β6,β7,β8)； (2) Let G(x) and F(x) be the difference calculation function and the normalization function, respectively. The normalized feature difference S = F(G(α)) is obtained from the one-dimensional feature vector. The feature extraction process of deep learning is simulated by mathematical operation. (3) Establish a feature difference model for the current frame, where h and w are the height and width, respectively; The feature difference model records the correlation between pixels mapped by the difference in neighboring pixel values.

2. The target tracking and detection method as described in claim 1, characterized in that, The normalization function in step (2) is the sigmoid function.

3. The target tracking and detection method as described in claim 1, characterized in that, In step two, when a moving target enters the monitoring screen, the algorithm's input changes. The system marks the pixels where the model changes, and determines whether the region is a foreground point or noise based on its area or context information. (1) The feature difference model M is obtained by processing the images transmitted from the monitoring screen: M： (2) Compare the information contained in the difference model. Let the function be V(x), then the output of the second step of the system is V(M). The threshold in the function V is adjusted according to the actual situation.

4. The target tracking and detection method as described in claim 1, characterized in that, In step three, the computer preprocesses all marked areas, eliminating noise through preprocessing, and then divides all pixels into background and foreground pixels. Finally, it outlines moving targets against a colored or grayscale background as needed, including: (1) Preprocessing treats smaller areas as noise, or judges based on context information. If the feature difference model of pixels in neighboring frames has not changed, it is considered as noise. (2) For all pixels in the current image, the function V(M) is regarded as a classifier - a set of fixed points A and a set of moving points B. Since the environment of the algorithm is a static background, the contour of the moving target is obtained after the pixel of the set of moving points B is assigned a value. The feature extraction operation of deep learning is simulated through various mathematical operations to obtain the foreground and background classifier, and then the target tracking and detection is realized.

5. A target tracking and detection system applying the target tracking and detection method as described in any one of claims 1 to 4, characterized in that, The target tracking and detection system includes: The feature difference model construction module is used to perform pixel-level learning on the current monitoring screen. Each point generates a set of feature differences based on its neighboring pixels. The feature difference is the correlation between pixels, thus obtaining the feature difference model. The feature difference model is used to record the background information under the current monitoring screen. The pixel marking module is used to mark the pixels in the model that have changed when a moving target enters the monitoring screen, and to determine whether the area is a foreground point or noise based on the area size or context information. The moving target delineation module uses a computer to preprocess all marked areas. After eliminating noise through preprocessing, all pixels are divided into background points and foreground points, and finally the moving target is delineated on a colored background or a grayscale background as needed.

6. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the target tracking and detection method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the target tracking and detection method as described in any one of claims 1 to 4.

Images