Robust target tracking method based on compressed sensing
By combining particle filter and Kalman filtering, using the compressed perception method based on sparse projection matrix and machine learning mechanism, the tracking failure problem of the existing target tracking algorithm under the target scale changes and occlusion is solved, and efficient and robust Target tracking effect.
Patent Information
- Application Number
- CN202411072324.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing target tracking algorithms are prone to failure when the target scale changes, with large storage volume, slow computing speed, high hardware requirements, and difficult to effectively track under occlusion and blur.
A robust target tracking method based on compression perception is adopted, and a particle filter and Kalman filter are combined, and a sparse projection matrix and machine learning mechanism are used to reduce the calculation amount, improve the anti-occlusion capability, and update the target model online.
It realizes stable tracking under target scale changes and occlusion, reduces computing volume and storage requirements, and improves tracking robustness and real-timeness.
Smart Images

Figure CN118918147A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of electronic information technology, and in particular to a robust target tracking method based on compressed sensing. Background Art
[0002] The application scope of target tracking has become increasingly wide. It is not only widely used in daily life, such as home cameras, video cameras, games, etc., but also widely used in security monitoring equipment, especially in airports, large shopping malls, parking lots, etc., where target tracking plays a very important role. In addition, from the perspective of scientific research applications, target tracking methods are also an important basis for some other applications, such as face pose estimation and expression analysis. The pros and cons of target tracking methods have become a major bottleneck restricting such applications.
[0003] TLD is an efficient target tracking algorithm that only requires less prior information to achieve long-term online tracking of the target. It has a fast computing speed and high real-time performance, and can be effectively applied to situations where the target is obscured. It is of great significance for target tracking of drones. The original tracking algorithms have the disadvantage that tracking will fail when the target scale changes. The TLD algorithm requires a relatively large amount of storage, has a relatively slow computing speed, and has relatively high hardware requirements.
[0004] Therefore, those skilled in the art provide a robust target tracking method based on compressed sensing to solve the problems raised in the above background technology. Summary of the invention
[0005] The purpose of the present invention is to provide a robust target tracking method based on compressed sensing to solve the problems raised in the above background technology.
[0006] In order to achieve the above-mentioned invention object, the present invention adopts the following technical scheme:
[0007] The present invention provides a robust target tracking method based on compressed sensing, comprising the steps of:
[0008] S1: Get the first frame of video image from the video stream, mark the target object with the mouse, pre-process the image, extract the target object by using the image of the first frame, and establish a subspace;
[0009] For subsequent frames, the particle filter method is used to add Gaussian noise to generate particles based on the position of the target object in the previous frame. For each particle, its image is extracted, vectorized, and then projected into the subspace. The projection coefficient is obtained using PCA. If the image is occluded or blurred, an occlusion processing strategy is adopted.
[0010] Then the weight of each particle is obtained according to the reconstruction error;
[0011] Finally, the predicted position of the final target object is obtained according to the sampling method;
[0012] S2: Generate a sparse projection matrix for the target area where the target position is located;
[0013] S3: Initialize the detection module and Kalman prediction module, and perform foreground detection on the image;
[0014] S4: Load the next frame of video image, perform image foreground prediction and image preprocessing;
[0015] S5: The detection module detects the target in the image, and the tracking module tracks the target;
[0016] S6: The tracking module and the detection module are integrated, and the situation of the target in the image is determined to generate a system tracking pre-frame;
[0017] S7: Determine whether the object tracking fails. If so, perform Kalman prediction; if successful, proceed to the next step;
[0018] S8: Determine the occlusion state of the target in the image. If the occlusion is serious, perform Kalman prediction and display the target motion trajectory. If the occlusion is not serious, learn the target model online, update the target model in real time, and correct the errors of the tracker in the tracking module and the detector in the detection module.
[0019] S9: Repeat steps S3-S8 until the video stream ends.
[0020] Furthermore, the sampling method refers to: using the shoulder-hair rule in the image, selecting a part of the pixel points to calculate the size of the projection coefficient, and then calculating the particle weight according to the size of the projection coefficient.
[0021] Furthermore, the long column vector is compressed using the sparse projection matrix to obtain a low-dimensional feature, wherein the low-dimensional feature has a feature dimension of 300.
[0022] Furthermore, the low-dimensional features are retrained to obtain a feature projection matrix, through which high-discriminative features of positive samples and high-discriminative features of negative samples are selected.
[0023] Furthermore, the positive sample high discriminative features and the negative sample high discriminative features are positive sample high discriminative features and negative sample high discriminative features with a dimension of 70 dimensions.
[0024] Compared with the prior art, one or more of the above technical solutions have the following beneficial effects:
[0025] The present invention provides a robust target tracking method based on compressed sensing, which integrates a detection module, reduces the amount of calculation by setting a foreground detection module, improves the anti-occlusion ability of target detection by means of Kalman filtering, and utilizes a machine learning mechanism to enhance the tracking robustness of the detection module. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0027] Figure 1 It is a flow chart of the robust target tracking method based on compressed sensing of the present invention. DETAILED DESCRIPTION
[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.
[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0030] In this application, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inside", "outside", "center", "vertical", "horizontal", "lateral", "longitudinal" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.
[0031] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to specific circumstances.
[0032] In addition, the terms "installed", "set", "provided with", "connected", "connected", and "socketed" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, elements, or components. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0033] As shown in the attached figure of the instruction manual Figure 1 As shown, the present invention provides a robust target tracking method based on compressed sensing, comprising the steps of
[0034] S1: Get the first frame of video image from the video stream, mark the target object with the mouse, pre-process the image, extract the target object by using the image of the first frame, and establish a subspace;
[0035] First, assume that the current frame is the t-th frame, and the obtained subspace is U. Use the particle filter to select some candidate regions in the current image to be predicted, set as x1, x2, ..., xN, where N is the total number of particles. For each particle x (assuming xi is the value of its i-th pixel), the reconstruction error can be calculated;
[0036] For subsequent frames, the particle filter method is used to add Gaussian noise to generate particles based on the position of the target object in the previous frame. For each particle, its image is extracted, vectorized, and then projected into the subspace. The projection coefficient is obtained using PCA. If the image is occluded or blurred, an occlusion processing strategy is adopted.
[0037] Then the weight of each particle is obtained according to the reconstruction error;
[0038] Finally, the predicted position of the final target object is obtained according to the sampling method;
[0039] S2: Generate a sparse projection matrix for the target area where the target position is located;
[0040] S3: Initialize the detection module and Kalman prediction module, and perform foreground detection on the image;
[0041] Select the video frame containing the target object with the mouse on the PC, pre-process the video frame, initialize the tracking module and detection module of the system, perform foreground detection on the video frame, and preliminarily determine whether the video frame contains the target;
[0042] S4: Load the next frame of video image, perform image foreground prediction and image preprocessing;
[0043] S5: The detection module detects the target in the image, and the tracking module tracks the target;
[0044] The tracking module uses the pyramid tracking method and adds a new tracking method for tracking failure detection algorithm. According to the target frame of the previous frame, several pixel points are selected as feature points, and the feature points of the previous frame corresponding to the position in the current frame are found in the next frame. Then, the position changes of these feature points between two adjacent frames are sorted, and the pyramid optical flow method is used to find the position of several feature points of the previous frame in the current frame in two adjacent video frames;
[0045] S6: The tracking module and the detection module are integrated, and the situation of the target in the image is determined to generate a system tracking pre-frame;
[0046] S7: Determine whether the object tracking fails. If so, perform Kalman prediction; if successful, proceed to the next step;
[0047] S8: Determine the occlusion state of the target in the image. If the occlusion is serious, perform Kalman prediction and display the target motion trajectory. If the occlusion is not serious, learn the target model online, update the target model in real time, and correct the errors of the tracker in the tracking module and the detector in the detection module.
[0048] S9: Repeat steps S3-S8 until the video stream ends.
[0049] The sampling method refers to: using the shoulder-hair rule in the image, selecting a part of the pixels to calculate the size of the projection coefficient, and then finding the particle weight according to the size of the projection coefficient.
[0050] The long column vector is compressed using a sparse projection matrix to obtain low-dimensional features, where the low-dimensional features have a feature dimension of 300.
[0051] The low-dimensional features are retrained to obtain a feature projection matrix, through which high discriminant features of positive samples and high discriminant features of negative samples are selected.
[0052] The positive sample high discriminant features and the negative sample high discriminant features are positive sample high discriminant features and negative sample high discriminant features with a dimension of 70 dimensions.
[0053] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A robust target tracking method based on compressed sensing, characterized in that: Included steps S1: Get the first frame of video image from the video stream, mark the target object with the mouse, pre-process the image, extract the target object by using the image of the first frame, and establish a subspace; For subsequent frames, the particle filter method is used to add Gaussian noise to generate particles based on the position of the target object in the previous frame. For each particle, its image is extracted, vectorized, and then projected into the subspace. The projection coefficient is obtained using PCA. If the image is occluded or blurred, an occlusion processing strategy is adopted. Then the weight of each particle is obtained according to the reconstruction error; Finally, the predicted position of the final target object is obtained according to the sampling method; S2: Generate a sparse projection matrix for the target area where the target position is located; S3: Initialize the detection module and Kalman prediction module, and perform foreground detection on the image; S4: Load the next frame of video image, perform image foreground prediction and image preprocessing; S5: The detection module detects the target in the image, and the tracking module tracks the target; S6: The tracking module and the detection module are integrated, and the situation of the target in the image is determined to generate a system tracking pre-frame; S7: Determine whether object tracking fails. If so, perform Kalman prediction. If successful, proceed to the next step; S8: Determine the occlusion state of the target in the image. If the occlusion is serious, perform Kalman prediction and display the target motion trajectory; If the occlusion is not serious, the target model is learned online, updated in real time, and the errors of the tracker in the tracking module and the detector in the detection module are corrected; S9: Repeat steps S3-S8 until the video stream ends.
2. The method for robust target tracking based on compressed sensing according to claim 1, characterized in that: The sampling method is to select a part of pixel points in the image using the rule of shoulder-hair style to calculate the size of the projection coefficient, and then calculate the particle weight according to the size of the projection coefficient.
3. The method for robust target tracking based on compressed sensing according to claim 1, characterized in that: The long column vector is compressed using the sparse projection matrix to obtain a low-dimensional feature, wherein the low-dimensional feature is a low-dimensional feature with a feature dimension of 300.
4. The method for robust target tracking based on compressed sensing according to claim 3, characterized in that: The low-dimensional features are retrained to obtain a feature projection matrix, through which high-discriminative features of positive samples and high-discriminative features of negative samples are selected.
5. The method for robust target tracking based on compressed sensing according to claim 1, characterized in that: The positive sample high discriminative features and the negative sample high discriminative features are positive sample high discriminative features and negative sample high discriminative features with a dimension of 70 dimensions.
