Real-time continuous tracking method, system and device under complex background and medium

Abstract

The invention discloses a real-time continuous tracking method, system and device under a complex background and a medium, and the method comprises the steps: judging a tracking state of a tracking target, calling a sub-template in a sub-template cache pool when the tracking target is lost, and calculating a current target tracking state; and executing an LET-NET network loss recapture mechanism according to the sub-template to recapture the lost target, the method, the system, the equipment and the medium can realize rapid capture of the target, the processing speed is relatively high, and the application requirements are met.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and relates to a real-time continuous tracking method, system, device and medium under complex background. Background Art

[0002] In the field of computer vision, target tracking technology is widely used in video surveillance, autonomous driving, robot navigation and other scenarios. However, in complex environments, target tracking systems still face the challenge of continuous and stable high-precision tracking tasks. This challenge usually faces the problem of loss and recapture during the tracking process. Tracking loss usually occurs in complex situations such as rapid movement, deformation or occlusion of the target. How to quickly and accurately recapture the target after tracking loss directly determines the continuous robustness of the tracking task.

[0003] In the process of target tracking, in order to solve the problem of loss caused by rapid movement, existing technologies usually use motion model prediction and multi-scale search, but these methods have limited prediction accuracy and take a long time when facing fast and irregular motion. In order to improve the effect of target recapture, modern methods usually combine high frame rate interpolation technology with accurate motion prediction mechanism to ensure that the latest position of the target can be captured in time when the target moves at high speed. At the same time, continuous tracking tasks should have flexible feature description and model update strategies. In particular, when the target is deformed, the system can quickly adapt to the new target appearance through deep learning models (such as long short-term memory networks) and adaptive feature extraction methods. In addition, when dealing with target occlusion problems, target recapture technology needs to rely on adaptive appearance models and trajectory prediction mechanisms, combined with local feature tracking and trajectory association technology, dynamically adjust feature descriptors to reduce background interference, and infer the possible motion path of the target through trajectory information, so as to quickly relocate the target after the occlusion ends. The above comprehensive measures are aimed at improving the robustness of the tracking system and ensuring the ability to recapture targets in complex environments.

[0004] Although the above solutions significantly improve the robustness of the target tracking system in special scenarios and provide a feasible practical method for continuous tracking in complex environments, these optimization mechanisms are usually accompanied by high computational complexity, resulting in a significant decrease in processing speed. In addition, they lack a clear distinction between target tracking states and are difficult to meet the needs of real-time applications. Therefore, although these technologies provide effective support for continuous tracking tasks in theory and practice, they still face challenges in real-time performance in practical applications. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a real-time continuous tracking method, system, device and medium under complex backgrounds. The method, system, device and medium can achieve rapid capture of the target with a fast processing speed to meet the needs of real-time applications.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme:

[0007] In one aspect, the present invention provides a real-time continuous tracking method under complex background, comprising:

[0008] Determine whether the tracking target is lost. If the tracking target is lost, call the sub-template in the sub-template cache pool to calculate and determine the target status;

[0009] The LET-NET network loss recapture mechanism is executed according to the sub-template to recapture the lost target.

[0010] The real-time continuous tracking method under complex backgrounds of the present invention is further improved in that:

[0011] Further, in the process of executing the LET-NET network loss recapture mechanism to recapture the lost target;

[0012] The single-frame image recapture mechanism is used to recapture the lost target.

[0013] Furthermore, in the recapture of the lost target using the single-frame image recapture mechanism, the coordinate offset (x c ,y c ), and weighted calculation with the recapture position (x, y), to update the position information of the lost target. The calculation formula is:

[0014] (x new ,y new )=α·(x,y)+β·(x c ,y c ).

[0015] In one aspect, the present invention provides a real-time continuous tracking system under complex background, comprising:

[0016] The state determination module is used to determine whether the tracking target is lost. When the tracking target is lost, the sub-template in the sub-template cache pool is called to calculate and determine the target state;

[0017] The recapture module is used to execute the LET-NET network loss recapture mechanism according to the sub-template to recapture the lost target.

[0018] The real-time continuous tracking system under complex background of the present invention is further improved in that:

[0019] Further, in the process of executing the LET-NET network loss recapture mechanism to recapture the lost target;

[0020] A single-frame image recapture mechanism is used to recapture lost targets.

[0021] Furthermore, in the recapture of the lost target using the single-frame image recapture mechanism, the coordinate offset (x c ,y c ), and weighted calculation with the recapture position (x, y), to update the position information of the lost target. The calculation formula is:

[0022] (x new ,y new )=α·(x,y)+β·(x c ,y c ).

[0023] In one aspect, the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the real-time continuous tracking method in a complex background when executing the computer program.

[0024] In a fourth aspect of the present invention, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the real-time continuous tracking method under a complex background are implemented.

[0025] The present invention has the following beneficial effects:

[0026] The real-time continuous tracking method, system, device and medium under complex background described in the present invention, during specific operation, achieves fast and accurate recapture of lost targets by utilizing the ultra-lightweight LET-NET network, and synchronously adjusts the tracker in combination with the sub-template cache pool, distinguishes the target motion state, and finally achieves dynamic continuous tracking. Compared with the existing improved methods, the present invention achieves stable continuous tracking while taking real-time into account, and achieves a good balance between accuracy and real-time, providing a more efficient and reliable solution for the field of long-term target tracking, and has broad practical application potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings constituting 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 an improper limitation of the present invention. In the accompanying drawings:

[0028] Figure 1 is a flow chart of the method of the present invention;

[0029] Figure 2 Schematic diagram of the sub-template cache pool and loss handling mechanism;

[0030] Figure 3 This is a flowchart for processing single-frame LET-NET network loss;

[0031] Figure 4 Flowchart for adaptive scale change processing;

[0032] Figure 5 is a system structure diagram of the present invention;

[0033] Figure 6 This is a time statistics chart for the verification experiment. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only an embodiment of a part of the present invention, not all embodiments, and is not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concepts disclosed in the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0035] The accompanying drawings show schematic diagrams of structures according to embodiments disclosed in the present invention. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0036] Embodiment 1

[0037] refer to Figures 1 to 5 The real-time continuous tracking method under complex background of the present invention includes:

[0038] 1) When the tracking target is lost, the sub-template in the sub-template cache pool is called;

[0039] 2) Calculating the tracking state of the current target according to the sub-template;

[0040] 3) When the tracking state is lost, the LET-NET network loss recapture mechanism is executed according to the sub-template to recapture the lost target.

[0041] The specific mechanism of LET-NET network loss recapture is as follows:

[0042] In order to improve the accuracy of target recapture, make full use of the feature information of the target image, and take into account the real-time requirements, the LET-NET ultra-lightweight convolutional network with low computational complexity and low time consumption was selected. The LET-NET network model was integrated into the KCF algorithm, and the feature descriptors and visualization key points of the target image area were output through the network, and combined with the color information of the target image, the algorithm's target loss recapture efficiency was effectively improved.

[0043] In combination with actual needs, a single-frame image recapture mechanism was designed and implemented.

[0044] The following is an explanation of the specific steps of the recapture method: 21) Single-frame image recapture mechanism;

[0045] Considering the limited computing power resources of embedded devices, in order to reduce the computing time, the key points are calculated based on the single-frame image (the current lost frame) input network, combined with color information, and the target location information is obtained by focusing on the clustered area of ​​key points. The focus is on finding the logic of key point dense areas and obtaining target location information. The specific steps are as follows:

[0046] 211) Target area enhancement;

[0047] By using local image contrast enhancement and image sharpening, the amount of calculation is reduced by adjusting the contrast and sharpening threshold parameters to ensure the normal extraction of key points in the target blurred area.

[0048] 212) Feature extraction;

[0049] In order to reduce time consumption, the input image is lightweight processed, including adjusting the input image format to 8-bit integer and expanding the target size by 25 pixels based on the original target area to ensure that it meets the input requirements of the LET-NET network. On this basis, the LET-NET network is used to efficiently extract the feature descriptor of the target from the current image.

[0050] 213) Key point detection and processing;

[0051] The AKAZE feature (Accelerated-KAZE) detects and calculates key points. After the embedded parameter adjustment, the possibility of abnormal key points increases, so the screening of key points and the deletion of outliers are improved, 40% of the key points are eliminated, and the lowest and highest 20% of the remaining key points are removed. Then the key points are converted to ensure the robustness of subsequent calculations, tracking and recapture frames.

[0052] 214) The key point obtains the target position;

[0053] Input the current lost frame image, calculate the key point dense area of ​​the target, determine the key point clustering area through density analysis, focus on the center position of the most dense key point area, and obtain the target position information logic as follows:

[0054] Considering the prominent color of the target, the target position information is obtained by combining the target color information; the accuracy of recapture can be improved by weighting the color and key point features. Specifically, the coordinate offset (x c ,y c ), and the recaptured position (x, y) of the key point are weighted and calculated to update the position information of the lost target. The calculation formula is:

[0055] (x new ,y new )=α·(x,y)+β·(x c ,y c )

[0056] Among them, (x new ,y new ) is the calculated weighted specific coordinate position, α and β are parameters for controlling the weights of different parts. After testing, the recapture effect is good when α is 0.8 and β is 0.2.

[0057] 215) target position update;

[0058] The target position is updated based on the key point aggregation area, and the boundary of the new tracking box is restricted to prevent the box from going out of bounds and causing tracking interruption.

[0059] It should be noted that the relevant parameter settings of the target loss recapture mechanism based on the LET-NET network are not fixed. The present invention is based on single-threaded CPU processing, the test platform is PC, the input images are divided into two categories, including public data sets and self-built data sets, the number of image channels is three channels, and the AKAZE feature calculation is used. After the OTB data set test, the single-frame target recapture effect is good. Specifically, the tracking situation needs to adjust the relevant configuration parameters to ensure the real-time tracking, for example, multi-threading or GPU participation in processing, the use of more accurate SURF, ORB and other feature calculations, the whole image input and other related parameter settings.

[0060] In addition, considering that in the process of continuous tracking, the tracker may be interrupted due to various complex scenes. The fundamental reason is that a single tracker cannot perfectly adapt to various target template states in the tracking process, resulting in the inability to achieve stable continuous tracking. Therefore, the present invention designs a sub-template cache pool mechanism, which uses the historical frame information of the target template to feedback the target state, and performs value feedback according to the set logic. According to the expected threshold, the dynamic management of the tracker is realized to achieve robust and efficient continuous tracking, which specifically includes: a cache pool structure (capacity, quantity, sub-template), a sub-template structure (position coordinate information, part of the tracker parameters, pixel data), and a sub-template update response strategy (blur, deformation, occlusion), which is specifically:

[0061] 31) The operation logic of the sub-template cache pool;

[0062] The sub-template cache pool includes four major processing logics, specifically:

[0063] Initialization: The initialization of the sub-template cache pool includes initializing the cache pool capacity, the number of sub-templates in the cache pool, and the cache pool sub-template structure; initialization directly specifies the capacity C of the sub-template cache pool, the default number of sub-templates N is 0, and memory is allocated for the sub-templates.

[0064] Add: This function adds a new sub-template based on an empty cache pool and counts the number of sub-templates. It is generally used to extract sub-template data for the first time and put it into the cache pool, or to open up new space in the cache pool for use.

[0065] Replace: Replace the sub-template in the cache pool. The data transmission comes from the target data of the current tracker, and the replacement content is the target historical data. It is mainly used for the sub-template strategy update mechanism in the tracking process, setting a fixed frequency update to replace the current sub-template.

[0066] Destruction: mainly releases sub-template members, including temporary pixel information and tracker data in the sub-template.

[0067] 32) Sub-template replacement and update strategy;

[0068] A sub-template update strategy mechanism is established. Four sub-templates are permanently stored in the cache pool. The template historical frame information is saved to determine the target's own state, including the determination of rapid movement, deformation, and occlusion. Finally, dynamic management of the tracker is achieved. Pay attention to the update frequency set when replacing the sub-template. The quality of the tracked target needs to be limited to ensure that the replaced current frame sub-template is a target in a normal tracking state. The specific process of state determination is as follows:

[0069] a. Rapid motion state determination;

[0070] According to the set replacement frequency, the sub-template data of the historical frame is intercepted and compared with the current sub-template information, the target offset in different periods is calculated, and the Euclidean distance is calculated using key points. The key points are the upper left, upper right, center, lower left, and lower right corners of the target frame. The single-frame threshold is set to 20. If it exceeds 20, it is determined that the target is currently in a fast-moving state. At the same time, the loss response threshold of the tracker is dynamically changed to ensure that the target can be continuously tracked. (x 1 ,y 1 ) and (x 2 ,y 2 )The Euclidean formula for two points is

[0071]

[0072] The five key points are calculated and averaged as the final value.

[0073] b. Determination of target deformation;

[0074] The process of determining whether the target has deformed is similar to the process of determining the fast moving state, except for the deformation calculation method, threshold setting, and dynamic management of the tracker's multi-scale parameters. The calculation method for determining whether the target has deformed is to use the rotation matrix norm of the affine transformation to determine whether the target has deformed. After testing, the deformation norm threshold is set to 0.2, and the rotation matrix norm ∥R∥ is obtained by calculation. F , it can be judged whether the target has deformed. When the norm value exceeds the set threshold, it is considered that the target has deformed. Norm ∥ R ∥ F (square root of the sum of the squares of the matrix elements) is calculated as:

[0075]

[0076] Among them, ∥R∥ F The Frobenius norm of the matrix R is the square root of the sum of the squares of all elements of the matrix. ij Represents the element in the i-th row and j-th column of the matrix R.

[0077] c. Target occlusion determination;

[0078] By calculating the similarity between the current lost frame and the simulated occlusion sub-template, we can further determine whether the target is occluded. If the target is occluded, it is directly judged as lost.

[0079] c1. Occlusion simulation process;

[0080] The target image is covered by a mask generated by code. The mask color and occlusion range can be specified, and the occlusion position is random. Currently, two simulated occlusion states are set in the cache pool (partial occlusion and full occlusion).

[0081] c2. Calculation of occluded sub-template matching similarity;

[0082] The perceptual hashing method is used to calculate the correlation, which can better reflect the differences between different sub-templates when calculating similarity. The specific calculation process is as follows:

[0083] After grayscale conversion of image I, discrete cosine transform (DCT) is performed:

[0084]

[0085] Where I(x,y) is the grayscale value of the input image, M,N are the width and height of the image, and u,v are the coordinate indexes in the frequency domain.

[0086] Take the 8×8 low-frequency part in the upper left corner as the feature area: DCT 8×8 (I)

[0087] Calculate the mean μ and generate the binary hash value pHash(I):

[0088]

[0089] The similarity is calculated as the Hamming distance d of the hash values. H :

[0090]

[0091] Among them, δ(a,b) is a Boolean function, which takes 1 when a and b are not equal, otherwise it takes 0; at the same time, the smaller the Hamming distance is, the better the image I 1 and I 2 The more similar.

[0092] c3. Target occlusion status judgment;

[0093] The calculated matching similarity is compared with the set occlusion threshold of 0.8. When it is higher than 0.8, it is determined that the target is lost due to occlusion. After the tracker determines that the target is lost, it will skip tracking in the current frame until the target reappears and is recaptured.

[0094] In summary, the offset is calculated by the sub-template to dynamically manage the loss threshold parameter to ensure the tracking robustness during fast movement; the deformation value is the deformation norm to determine the degree of target deformation. The correlation parameter describes the similarity between the simulated occluded target and the current lost target, and then determines the target occlusion state.

[0095] Confirmatory experiments

[0096] The experimental data set uses two data sets, OTB50 / 100, and adopts the OPE evaluation method, where the CLE (center position error) threshold is set to 20, and the IOU (overlap) threshold is set to 0.5. The tracking success rate and accuracy of the data set are calculated to evaluate the tracking quality, and the tracking performance of the improved algorithm is analyzed in combination with the experimental data.

[0097] Experimental environment: Linux environment, CPU processing, c / c++ code implementation, opencv4.5.3 and other dependent library files.

[0098] Taking into account the hardware deployment requirements of the algorithm, five representative single-target tracking algorithms (correlation filter type) were selected for comparison with the RC-KCF (Regional Convolutional Kernelized Correlation Filter) proposed in this invention. In order to verify the continuity of tracking, the entire data set was tested at one time to simulate long-term tracking tasks. Finally, the RC-KCF improved algorithm achieved a continuous tracking accuracy of 78.2% and a success rate of 74.4% for a typical data set in OTB. The overall test frame rate of the algorithm was 41 frames per second, which can meet the daily real-time requirements. For its algorithm, its comprehensive tracking efficiency is higher than that of the listed algorithms. It reflects that the RC-KCF algorithm better balances the real-time and robustness of continuous tracking.

[0099] The time taken to recapture a single frame is also counted, and the average time taken is 28ms (the average time of the total time of 500 frames):

[0100] In order to highlight the continuous tracking performance of improvement measures such as the LET-NET network loss-recapture mechanism and sub-template cache pool in complex scenarios (fast movement, deformation, scale change, occlusion), separate tests and compilation were carried out for each scenario. Finally, the improved KCF algorithm performed well in four scenarios of fast movement, occlusion, target deformation and scale change. The overall tracking efficiency was better than the enumeration algorithm, especially in the fast-moving scenario, the accuracy and success rate were significantly improved.

[0101] In summary, the present invention improves the loss recapture accuracy and dynamic tracking capability of the KCF algorithm by introducing the sub-template cache pool and the LET-NET network loss recapture mechanism, and reflects the stability of the continuous tracking of the improved algorithm. Experimental results show that the present invention performs well in continuous tracking when dealing with complex scenes such as rapid movement, scale change, target deformation and target occlusion, especially in the secondary recapture process after rapid movement and target occlusion, the tracking efficiency is significantly better than the traditional correlation filtering algorithm. In addition, while maintaining high precision and robustness, the present invention still has good real-time performance, showing strong edge device application potential and practical engineering value. The present invention provides a more stable and efficient solution for the task of real-time continuous tracking of targets, and has broad application prospects.

[0102] Embodiment 2

[0103] The real-time continuous tracking system under complex background of the present invention comprises:

[0104] The judgment module is used to judge the tracking state of the tracking target. When the tracking target is lost, the sub-template in the sub-template cache pool is called;

[0105] The recapture module is used to execute the LET-NET network loss recapture mechanism according to the sub-template to recapture the lost target.

[0106] In this embodiment, the LET-NET network in the LET-NET network loss and recapture mechanism is a LET-NET ultra-lightweight convolutional network.

[0107] In this embodiment, the LET-NET network loss recapture mechanism is executed to recapture the lost target;

[0108] A single-frame image recapture mechanism is used to recapture lost targets.

[0109] In this embodiment, in the recapture of the lost target using the single-frame image recapture mechanism, the coordinate offset (x c ,y c ), and the weighted calculation of the key point recapture position (x, y), to update the position information of the lost target. The calculation formula is:

[0110] (x new ,y new )=α·(x,y)+β·(x c ,y c ).

[0111] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present application may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0112] Embodiment 3

[0113] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the real-time continuous tracking method under a complex background are implemented, for example, including: determining whether the tracking target is lost, and when the tracking target is lost, calling the sub-template in the sub-template cache pool; executing an external dynamic search mechanism according to the sub-template to recapture the lost target; when the lost target cannot be recaptured through the external dynamic search mechanism, executing the LET-NET network loss recapture mechanism according to the sub-template to recapture the lost target. Wherein, the memory may include a memory, such as a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk memory, etc.; the processor, the network interface, and the memory are interconnected through an internal bus, and the internal bus may be an industrial standard architecture bus, a peripheral component interconnection standard bus, an extended industrial standard structure bus, etc., and the bus may be divided into an address bus, a data bus, a control bus, etc. The memory is used to store programs. Specifically, the program may include a program code, and the program code includes computer operation instructions. The memory may include a memory and a non-volatile memory, and provide instructions and data to the processor.

[0114] Embodiment 4

[0115] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the real-time continuous tracking method under a complex background are implemented, for example, including: determining whether the tracking target is lost, and when the tracking target is lost, calling a sub-template in a sub-template cache pool; executing an external dynamic search mechanism according to the sub-template to recapture the lost target; when the lost target cannot be recaptured by the external dynamic search mechanism, executing a LET-NET network loss recapture mechanism according to the sub-template to recapture the lost target. Specifically, the computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include a read-only memory (ROM), a hard disk, a flash memory, an optical disk, a magnetic disk, etc.

[0116] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0117] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A real-time continuous tracking method under complex background, characterized in that: include: Determine the tracking status of the tracking target. When the tracking target is lost, call the sub-template in the sub-template cache pool to calculate and determine the target status; The LET-NET network loss recapture mechanism is executed according to the sub-template to recapture the lost target.

2. The real-time continuous tracking method under complex background according to claim 1, characterized in that: The LET-NET network loss recapture mechanism is executed to recapture the lost target; The lost target is recaptured according to the single-frame image recapture mechanism.

3. The real-time continuous tracking method under complex background according to claim 2, characterized in that: In the recapture of the lost target using the single-frame image recapture mechanism, the coordinate offset (x c ,y c ), and weighted calculation with the key point recapture position (x, y), to update the position information of the lost target. The calculation formula is: (x new ,and new )=α·(x,y)+β·(x c ,and c )。 4. A real-time continuous tracking system under complex background, characterized in that: include: The judgment module is used to judge the tracking state of the tracking target. When the tracking target is lost, the sub-template in the sub-template cache pool is called to calculate and determine the target state; The recapture module is used to execute the LET-NET network loss recapture mechanism according to the sub-template to recapture the lost target.

5. The real-time continuous tracking system under complex background according to claim 4, characterized in that: The LET-NET network loss recapture mechanism is executed to recapture the lost target; The lost target is recaptured according to the single-frame image recapture mechanism.

6. The real-time continuous tracking system under complex background according to claim 5, characterized in that: In the recapture of the lost target using the single-frame image recapture mechanism, the coordinate offset (x c ,y c ), and weighted calculation with the key point recapture position (x, y), to update the position information of the lost target. The calculation formula is: (x new ,and new )=α·(x,y)+β·(x c ,and c )。 7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the real-time continuous tracking method under a complex background as described in any one of claims 1-3 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the real-time continuous tracking method under a complex background as claimed in any one of claims 1 to 3 are implemented.

Images