Video SAR (Synthetic Aperture Radar) target tracking method and system based on spatio-temporal extension-accumulation strategy

Abstract

The invention discloses a video SAR (Synthetic Aperture Radar) target tracking method and system based on a spatio-temporal extension-accumulation strategy. The method comprises the following steps: dividing a video SAR image data set into N batch processing packets; extracting features of the initial position through a background perception correlation filter, obtaining initialization parameters and an initial filter template of the background perception correlation filter, and generating an initial candidate box; delimiting a target search area of a second frame based on the candidate frame of the first frame to obtain a plurality of test samples, calculating similarity and obtaining a response matrix; selecting S target prediction positions corresponding to the maximum response values from the response matrix; extracting features of each prediction position and updating a corresponding template to generate S candidate frames; performing repeated iteration to obtain # imgabs0 # candidate frames, and forming a multi-frame tree structure; the estimated position of the target in each frame of the tree structure is searched reversely, and candidate paths of # imgabs1 # target tracks are obtained; and selecting the candidate path with the maximum value function as the path of the target in one batch processing packet, and outputting the estimation state of the target in each frame.

Description

Background Art

[0002] According to the SAR imaging principle, in high frame rate video SAR images, the energy of moving targets will be greatly offset and defocused relative to the real position of the target, leaving a shadow at its real position. The state information of the moving target shadow is consistent with the real state information of the target. In video SAR images, because there is no phase information from the target shadow area, the shadow imaging of the target is not affected by any distortion effect; in addition, the contrast between the target shadow and clutter is independent of the target radar scattering cross section, which means that the shadow cannot be eliminated or reduced by stealth technology. Due to the above advantages, analyzing the characteristics of moving target shadows and developing a series of detection and tracking methods have become one of the research hotspots in the field of video SAR target tracking.

[0003] Algorithms based on correlation filtering have become popular in video SAR target tracking in recent years. Zhong et al. proposed a video SAR moving target tracking framework based on joint kernel correlation filter (JKCF), which combines the shadows in video SAR images with the energy in the corresponding range-Doppler spectrum to track small targets. Zhao et al. equipped the tracker with a detection mechanism based on spatiotemporal information and saliency for interference and background clutter. Tian et al. proposed a new BACF algorithm based on appearance-range information assisted probabilistic data association (AD-PDA), which uses video SAR images to estimate the target state.

[0004] TBD technology is a detection and tracking technology for weak targets. The basic steps of tracking targets are: no threshold is set for a single frame or the threshold is set very low, then multiple frames of data are accumulated according to the correlation of the target's motion between frames, and finally compared with the set threshold to judge the target and obtain the target's estimated track. The dynamic programming-based detection-before-tracking (DP-TBD) and particle filter-based detection-before-tracking (PF-TBD) algorithms are two typical algorithms. Because of their excellent comprehensive performance, they have become hot algorithms for scholars to track weak targets. Using the DP-TBD algorithm to process video SAR image sequences, on the one hand, a value function based on target amplitude can be designed, and the target's track correlation can be used to accumulate the target amplitude and improve the signal-to-noise ratio of weak targets; on the other hand, the value function accumulation error caused by the model assumption distribution is avoided when the video SAR measurement model is unknown. Tian Xiaoqing et al. proposed an improved ES-TBD algorithm to detect and track a fixed number of slow-moving targets. Based on the DP-TBD algorithm, Qin et al. proposed a JP-DP-TBD algorithm based on a joint processing strategy. By analyzing the position and radial velocity information of the target, this method can screen the state candidate areas and retain more states that conform to the target motion laws, thereby tracking multiple maneuvering targets.

[0005] Although the above trackers have achieved excellent balance between tracking accuracy and speed and achieved excellent performance, the following factors still limit their application in video SAR target tracking. First, the size of the shadow target is very small, and the complex texture features are masked by speckle noise. Second, the surrounding background is complex, and the tracker is likely to be disturbed by man-made structures or similar objects in the background. The above situations will lead to multiple peaks in the correlation filter response graph, and the real target is not at the highest peak, which is also the difficulty of video SAR weak shadow target tracking. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a video SAR target tracking method and system based on a spatiotemporal expansion-accumulation strategy in view of the deficiencies in the above-mentioned prior art, so as to solve the technical problem of target tracking frame drift in video SAR target tracking and achieve high-precision tracking of video SAR weak shadow moving targets.

[0007] The present invention adopts the following technical solutions: The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy includes the following steps: S1. M The video SAR image dataset of frames is divided into N Batch packages, each containing K Frame video SAR image; S2: Manually select the tracking target in the initial frame of the first batch package obtained in step S1, and extract the features of the initial position using the background-aware correlation filter F 1 , get the initialization parameters and initial filter template of the background-aware correlation filter , generate an initial target candidate box ; S3, read the second frame image in the first batch package obtained in step S1, based on the candidate frame of the first frame obtained in step S2 Define the target search area for the second frame , The background-aware correlation filter is divided into multiple test samples, each test sample and Calculate similarity and obtain a response matrix; S4. Select the response matrix with the maximum response value from the response matrix obtained in step S3. , and from Select S The maximum response value, each response value corresponds to a predicted position of the target; extract features for each predicted position and update the corresponding template, and finally generate S target candidate boxes; S5, for the subsequent k Frame No. j Candidate boxes , repeat steps S3 to S4 to obtain target candidate boxes; K Frame, repeat step S3 to step S4 to obtain target candidate boxes, at this time all target candidate boxes are generated to form a multi-frame tree structure; S6, using the trajectory backtracking algorithm to reversely search for the estimated position of each frame of the target in the tree structure obtained in step S5, K Frame No. Candidate frames are started to backtrack, and a total of Candidate paths of target trajectories; S7. Calculate the value function corresponding to each candidate path obtained in step S6, select the candidate path with the largest value function as the estimated path of the target in a batch package, and output the estimated state of the target in each frame.

[0008] Preferably, in step S1, the last frame of the previous batch package and the first frame of the next batch package are the same frame, the target initial state of the first frame of the next batch package is the target estimation state of the previous batch package after screening by the trajectory backtracking module, and the corresponding candidate box information including the position-size state vector, features, templates, and response values ​​are all from the last frame of the previous batch package.

[0009] Preferably, in step S2, the feature F 1 is a 32-dimensional feature, including 31-dimensional directional histogram features and grayscale features. The target tracking frame is , The superscript indicates the number of the current candidate box, and the subscript indicates the frame number of the image in the current batch package.

[0010] Preferably, in step S3, the target search area The location is based on Position generation, target search area The size is right The position-size state vector is enlarged in width and height Double generation, is an adjustable parameter with a range of [4,8]; the resulting response matrix is the matrix with the maximum value among all response matrices.

[0011] Preferably, in step S4, each response value corresponds to the predicted position of the target, which satisfies the following relationship:

[0012] in,( ) is the response value The corresponding s The coordinates of the predicted target locations.

[0013] Preferably, in step S5, the tree structure constituting the multiple frames is specifically: S501, initial target candidate box and the position-size state vector of the candidate box ; S502: Generate a search area in the second frame , The background-aware correlation filter is used to crop the sample into multiple test samples, each with the same size and dimension. Same, each test sample and Perform cross-correlation similarity calculation and obtain a response matrix, and select the response matrix with the maximum response value , It is a two-dimensional matrix. The value in each grid is between 0 and 1, and each grid corresponds to a position coordinate of the current frame. S503. Generate response matrix from search area Select the front S The maximum response value is obtained S target candidate boxes, the size of each candidate box is determined by the scale adjustment factor of the background-aware correlation filter; S504: For any frame in the first batch processing package k -1, from k -Select one from all candidate frames in 1 frame , After repeating step S502 and step S503, k Frame Generation candidate boxes, represented as , ; And so on, in the last frame of the first batch package ( k=K )Total The candidate box of a target is represented as , ; Store all candidate box information in the first batch package in the buffer area for subsequent calculations. n When >1, the candidate box information of the initial frame directly retrieves the target estimation result of the last frame of the previous batch package, and the target candidate box generation process of the remaining frames is the same as that of the first batch package.

[0014] Preferably, an adaptive template updating method is used to calculate the first batch of initial frames. n The grayscale mean difference between the first frame of the batch package and the initial frame of the first batch package is used to adjust the online learning rate variables.

[0015] Preferably, the online learning rate for:

[0016] in, is the grayscale difference index, is the threshold value, and is the value of the online learning rate.

[0017] Preferably, in step S7, Select the function with the largest value from the candidate paths The path is:

[0018] in, Represents the set of all candidate paths in a batch package, is the estimated position of the target for all frames estimated by the algorithm, Estimate the position of the target for the Kth frame, All predicted positions for all frames of the target.

[0019] In a second aspect, an embodiment of the present invention provides a video SAR target tracking system based on a spatiotemporal expansion-accumulation strategy, comprising: Divide the module into a M The dataset of frame video SAR images is divided into N Batch packages, each containing K Frame image; The selection module manually selects the tracking target in the initial frame of the first batch package obtained by the division module, and the background-aware correlation filter extracts the features of the initial position F 1 , get the initialization parameters and initial filter template of the background-aware correlation filter , generate an initial candidate box ; The reading module reads the second frame image in the first batch package obtained by the segmentation module, based on the candidate box of the first frame obtained by the selection module Define the target search area for the second frame , The background-aware correlation filter is divided into multiple test samples, each test sample and Calculate similarity and obtain a response matrix; Update module, select the response matrix with the maximum response value from the response matrix obtained by the reading module , and from Select S The maximum response value, each response value corresponds to a target prediction position; extract features for each prediction position and update the corresponding template, and finally generate S target candidate boxes; Repeat the module, for the subsequent k Frame No. j Candidate boxes , after repeatedly reading the module and updating the module, we get candidate boxes; K Frame, repeatedly read the module and update the module to get candidate boxes, at this time all candidate boxes are generated to form a multi-frame tree structure; The backtracking module uses the trajectory backtracking module to reversely search for the estimated position of the target in each frame in the tree structure obtained by the repeating module, starting from the K Frame No. The target candidate boxes start backtracking, and a total of Candidate paths of target trajectories; The tracking module calculates the value function corresponding to each candidate path obtained by the backtracking module, selects the candidate path with the largest value function as the estimated path of the target in a batch package, and outputs the estimated state of the target in each frame.

[0020] In a third aspect, a computer device includes 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 above-mentioned video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy when executing the computer program.

[0021] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy.

[0022] Compared with the prior art, the present invention has at least the following beneficial effects: The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy aims at the multi-peak problem caused by the target being occluded or the interference of similar targets in the correlation filtering. Inspired by the TBD idea, a new tracking mode is designed, that is, the original single-frame prediction-single-frame update method is modified to a multi-frame prediction-accumulation and screening-multi-frame update method, which successfully reduces the probability of target loss.

[0023] Furthermore, dividing the video SAR images into multiple batch packages has the following advantages: first, it reduces the computational complexity of the algorithm and improves the calculation speed of the algorithm; second, by accumulating multiple frames of information in each batch package according to the batch images, it can effectively avoid the problems of missed tracking and wrong tracking caused by the loss of single-frame target information, thereby improving the success rate of target tracking.

[0024] Furthermore, the use of 32-dimensional features can more accurately and comprehensively characterize the target, making it easier for the tracker to identify the target.

[0025] Furthermore, in a batch package, searching the target position of the current frame based on the target prediction position of the previous frame can more efficiently find the target, because the target moves a smaller distance in two consecutive frames. At the same time, the maximum response value represents the maximum correlation between the test sample and the target template, and the probability that this part of the test sample is the target is also the highest.

[0026] Furthermore, the method of selecting the position corresponding to the maximum response value from a response matrix as the target prediction position is abandoned, because the area where similar targets are located may also obtain the maximum response value. If the tracker mistakenly tracks similar targets, the template will be contaminated, and the target will eventually be lost. In order to reduce the probability of template contamination, S response values ​​are selected and their corresponding positions are obtained as candidate positions of the target, so that the candidate positions contain the true position of the target as much as possible.

[0027] Further, such as Figure 2 As shown in the figure, the target prediction positions of all frames in a batch package together constitute a multi-frame tree. For each target prediction position at the bottom of the multi-frame tree, a target prediction trajectory can be found through the backtracking module, and the real motion trajectory of the target is hidden in these prediction trajectories. This multi-frame tree structure covers the real trajectory of the target as comprehensively as possible, reducing the probability of target loss. Reasonable setting of the total number of frames in the tree structure and the number of next-level branches for each target prediction position can not only control the computational complexity of the algorithm, but also improve the success rate of target tracking.

[0028] Furthermore, in correlation filtering, the response value represents the similarity between the predicted target and the real target. The value function of the candidate path in the present invention refers to the sum of the response values ​​of the target prediction position under the path. The maximum value function indicates that the target prediction position corresponding to the candidate path is closest to the real target position. If there is interference from a similar target in a certain frame, resulting in the maximum response value at this position, the traditional correlation filtering algorithm will update the template according to the predicted position here, resulting in the template being contaminated, and the tracker will eventually lose the target. However, in the present invention, the possible positions of all targets in a batch package are first predicted, and then the corresponding response values ​​are summed, which can effectively avoid the occurrence of the above situation.

[0029] It can be understood that the beneficial effects of the second aspect mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0030] In summary, the algorithm proposed in the present invention uses the method of accumulating multi-frame information to track weak shadow targets in video SAR. After accumulating multi-frame image information, it obtains enough real target information and can successfully distinguish the target from clutter; a multi-frame tree structure is built to deal with the multi-peak problem in the correlation filter response graph. By generating multiple candidate frames to cover the real target as much as possible, even if the candidate frame of the real target in a certain frame does not have the maximum response value, the real trajectory of the target can still be successfully found after the tree structure is expanded in the time domain and space domain; the online learning rate of the correlation filter is flexibly adjusted based on the change of the target appearance, and the appropriate online learning rate is adaptively selected to update the template, which can improve the ability of the algorithm to adapt to different tracking scenarios; it has the ability to continuously and stably track the target, and has achieved excellent tracking accuracy and success rate, as well as very small positioning error, and has achieved excellent performance compared with other algorithms based on correlation filtering.

[0031] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of batch division; Figure 2 This is a schematic diagram of the tree structure. Figure 3 It is a framework diagram of the method of the present invention; Figure 4 A schematic diagram of a computer device provided by an embodiment of the present invention; Figure 5 The present invention is a block diagram of an electronic device provided according to an embodiment. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] In the description of the present invention, it should be understood that the terms “include” and “comprises” indicate the presence of described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0035] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0036] It should be further understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects are in an "or" relationship.

[0037] It should be understood that, although the terms first, second, third, etc. may be used to describe preset ranges, etc. in the embodiments of the present invention, these preset ranges should not be limited to these terms. These terms are only used to distinguish preset ranges from each other. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0038] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0039] Various structural schematic diagrams of the embodiments disclosed in the present invention are shown in the accompanying drawings. 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.

[0040] The present invention provides a video SAR target tracking method based on a spatiotemporal expansion-accumulation strategy, which includes training a target detector; using the target detector to detect whether the predicted position of the next frame is a target; and using the new detection result to update the training set and then update the target detector.

[0041] The BACF target tracking process includes two key stages: filter training and target detection.

[0042] Filter training BACF applies a circular shift operator and a cropping operator to the image, and uses the circular shifted and then twice cropped samples as the real training samples. Then BACF obtains an optimal filter by minimizing the following objectives: h :

[0043] in, is an ideal two-dimensional Gaussian response centered on the target, and denote the samples and the correlation filter, respectively. channels, is the number of feature channels, the superscript represents the transpose operation, Indicates the sample Conduct a The circular shift operation of step, Is the size of The binary matrix is ​​used to extract the middle of the candidate area. elements to select appropriate training samples to alleviate boundary effects. In general, ,because is equal to the size of the target object, T is the size of the training sample (including both the target object and the fill area), is the regularization term coefficient.

[0044] Object Detection In the detection phase, the calculation starts from Samples cropped from the frame search area With Correlation filter obtained by frame training Correlation operation response ,Right now , according to the response The target position is updated by the position of the maximum value, and the samples of a new frame are obtained at the target position to retrain the filter and prepare for tracking in the next frame.

[0045] See also Figure 3 The present invention provides a video SAR target tracking method based on a spatiotemporal expansion-accumulation strategy, comprising the following steps: S1. M The dataset of frame video SAR images is divided into N batch packages, each batch package contains K Frame video SAR image; The last frame of the previous batch and the first frame of the next batch are the same frame. The target estimation state of the first frame of the next batch is the state of the previous batch after being filtered by the trajectory backtracking module. The corresponding candidate box information including position-size state vector, features, templates, response values, etc. are all from the last frame of the previous batch.

[0046] S2, in the initial frame of the first batch ( k =1) Manually select the tracking target, and its state vector is , the background-aware correlation filter (BACF) extracts features at the initial position F 1 , get the initialization parameters and initial filter template of BACF , generate an initial candidate box , is the response value corresponding to the candidate box, and its value is 1; The BACF here may also be replaced by other correlation filters, such as kernel correlation filter (KCF), etc. The present invention selects BACF according to the effect of the algorithm.

[0047] feature F 1 is a 32-dimensional feature, including a 31-dimensional directional histogram feature and a grayscale feature. The target tracking frame can be further expressed as ,in, The superscript indicates the number of the current candidate box, and the subscript indicates the frame number of the image in the current batch.

[0048] S3, read the second frame image in the first batch, based on the candidate box of the first frame Define the target search area for the second frame , It is an adjustable parameter; Divided into multiple test samples by BACF, each test sample and Calculate similarity and obtain a response matrix ; Search area , whose location is based on The position is generated, and its size is The width and height of Double generation, is an adjustable parameter, and its range is [4,8]. The final response matrix is the matrix with the maximum value among all response matrices.

[0049] S4. From the matrix Select S The maximum response value, each response value corresponds to the predicted position of the target, that is, it satisfies the relationship ; Extract features for each predicted position and update the corresponding template, and finally generate S candidate boxes, i.e. ; Each value of the response matrix corresponds to a target prediction position, and the BACF algorithm applies a subgrid interpolation strategy to calculate the response results on the test sample. R , use the same method to calculate the response corresponding to each scale in the scale pool, and then use the scale with the largest response value as the size of the current frame target, and the maximum response value is the position of the target. S The maximum response value is calculated and its corresponding position is finally generated. S target candidate boxes.

[0050] S5. For the subsequent k Frame No. j Candidate boxes , repeat steps S3 to S4 to obtain candidate boxes; K Frame, repeat steps S3 to S4 to get At the end of this step, all candidate boxes are generated, and the generation process forms a multi-frame tree structure; Building a tree structure is as follows: In the BACF tracking process, the response matrix The position corresponding to the maximum value of represents the estimated position of the target in the current frame. However, due to the existence of multi-peak problems, the position with the maximum response value is not necessarily the target, but may also be clutter or similar structures. Therefore, it is necessary to select as many positions as possible as candidate positions of the target in the current frame. Take the first batch as an example to introduce the process of building the tree structure: (1) Filter initialization Manually input the target tracking frame information of the first frame (initial frame), and its position-size state vector is , Represents the center position of the initial target tracking box, and Represent the width and height of the tracking box respectively. Extract features at the tracking box Including 31-dimensional directional histogram features and grayscale features, using the extracted features Generate an initial template for the target And store it in the buffer, and finally generate a candidate box of the target in the initial frame In order to more simply introduce the algorithm idea of ​​the present invention, Has two meanings: 1) The name of the initial candidate box; 2) The state vector of the initial candidate box.

[0051] Correspondingly, its state vector can be expressed as In the initial frame of the first batch, the filter mainly performs autocorrelation calculation. The value of is 1.

[0052] (2) Search area generation See also Figure 2 , based on the initial candidate box The position size state vector , generate the search area in the second frame ,and The definition of is similar to that of , and the state vector of the search area is expressed as , is an adjustable parameter. In the present invention, The adjustment range is [4,8]. It is cut into multiple test samples by BACF, and the size and dimension of each test sample are the same as Same, each test sample and Perform cross-correlation similarity calculation and obtain a response matrix, and select the response matrix with the maximum response value ,in, It is a two-dimensional matrix. The value in each grid is between 0 and 1, and each grid corresponds to a position coordinate of the current frame.

[0053] In the traditional BACF algorithm, the position corresponding to the maximum value of the response matrix represents the estimated position of the target. However, in video SAR shadow target tracking, due to the small size of the target and the lack of effective texture features, the target is easily disturbed by clutter and similar structures. There may be multiple higher peaks in the response matrix, but the real target is not at the highest peak position, which causes the tracker to lose the target. Therefore, the present invention does not directly select the position corresponding to the maximum value of the response matrix as the estimated position of the target in the current frame.

[0054] (3) Selection of multiple candidate boxes Based on the problem of multiple peaks described in the search area generation part and inspired by the idea of ​​DP-TBD algorithm, the present invention obtains the maximum response matrix Select the front S The maximum response value is S The response values ​​satisfy the following relationship:

[0055] in, Represent this S The position corresponding to the response value is obtained. S The size of each candidate box is determined by the scale adjustment factor of BACF. The state vector of the target candidate box is defined as ,in , the candidate box includes the position-size state vector ,feature ,template and the response value .

[0056] (4)Repetition and iteration For any frame in the first batch k -1, from k -Select one from all candidate frames in 1 frame , , repeating process (2) and (3) in the k Frames can be generated candidate boxes, represented as , . And so on, in the last frame of the first batch ( k=K )Total The candidate box of a target is represented as , All candidate box information in the first batch is stored in the buffer area for calculation of the trajectory backtracking module.

[0057] The process of building the multi-frame tree structure in all subsequent batches is the same as that of the first batch, but one thing needs to be explained: when the batch number n When >1, the candidate box information of the initial frame directly retrieves the target estimation result of the last frame of the previous batch instead of manual delineation.

[0058] In order to balance computational complexity, speed and accuracy, the present invention K The value of is 4 but not limited to 4. The specific situation is analyzed specifically. The schematic diagram of building a multi-frame tree structure is as follows Figure 2 shown.

[0059] The template of the extension module needs to be updated in the template update module, and the updated template is transmitted back to the extension module. When the template update module updates the template, the features at the candidate box of the extension module need to be updated, and the two cannot be separated.

[0060] The update method is the same as the original BACF template update, that is,

[0061] in, Represents a multidimensional feature read from an extension module. Represents the template of the previous frame, represents the online learning rate, which represents the learning ability of the filter for changes in target appearance.

[0062] During the long-term tracking process of video SAR, factors such as illumination changes, target deformation, and noise interference will affect the appearance characteristics of shadow targets. At this time, a fixed learning rate is likely to cause target loss.

[0063] In order to improve the algorithm's adaptability to changes in target appearance features, the present invention proposes an adaptive template updating method, which takes the initial frame of the first batch as the benchmark and calculates the first batch of n The grayscale mean difference between the first frame of a batch and the initial frame of the first batch is used to adjust the online learning rate. Variables; First calculate the grayscale difference index :

[0064] Among them, the grayscale mean of the target area of ​​the initial frame of the first batch is , No. n The grayscale mean of the target area in the initial frame of the batch is .

[0065] Secondly, the algorithm follows Choose an appropriate online learning rate :

[0066] in, , is the threshold value.

[0067] In the data sample used in the present invention, , , =17.

[0068] S6, using the trajectory backtracking algorithm to reversely find the estimated position of the target in each frame, for the tree structure in step S5, from K Frame No. Candidate frames are started to backtrack, and a total of The candidate paths of the target trajectory are , ; S7. Calculate the value function corresponding to each candidate path, that is, , select the candidate path with the largest value function as the estimated path of the target in a batch, and output the estimated state of the target in each frame.

[0069] For each subsequent batch, the target state of its initial frame is extracted from the last frame of the previous batch, and steps S3 to S7 are repeated to finally complete the target tracking of the entire dataset.

[0070] The multi-frame tree structure of each batch is built, and the information of all candidate boxes is stored in the data buffer. The backtracking process is performed to filter out the estimated state of the target from a large number of target candidate boxes.

[0071] Inspired by the DP-TBD idea, this paper uses the response value of the candidate box as the accumulated value of the value function. The detailed ideas are as follows: Taking the first batch as an example, the last frame has target candidate boxes, and the corresponding target candidate paths have Any candidate path consisting of multiple target candidate boxes from different frames It is expressed as:

[0072] in, ; Candidate Path The accumulated value function is expressed as:

[0073] from Select the function with the largest value from the candidate paths The path, that is

[0074] in, Represents the set of all candidate paths in a batch.

[0075] Finally, the state vector of the filtered path is output, which is the estimated state of the target in each frame within a batch.

[0076] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as systems, methods or program products. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits", "modules" or "platforms".

[0077] In one embodiment of the present invention, a video SAR target tracking system based on a spatiotemporal expansion-accumulation strategy is provided. The system can be used to implement the above-mentioned video SAR target tracking method based on a spatiotemporal expansion-accumulation strategy. Specifically, the video SAR target tracking system based on a spatiotemporal expansion-accumulation strategy includes a division module, a selection module, a reading module, an update module, a repetition module, a backtracking module and a tracking module.

[0078] Among them, the module is divided into a M The dataset of frame video SAR images is divided into N Batch packages, each containing K Frame image; The selection module manually selects the tracking target in the initial frame of the first batch package obtained by the division module, and the background-aware correlation filter extracts the features of the initial position F 1 , get the initialization parameters and initial filter template of the background-aware correlation filter , generate an initial candidate box ; The reading module reads the second frame image in the first batch package obtained by the segmentation module, based on the candidate box of the first frame obtained by the selection module Define the target search area for the second frame , The background-aware correlation filter is divided into multiple test samples, each test sample and Calculate similarity and obtain a response matrix; Update module, select the response matrix with the maximum response value from the response matrix obtained by the reading module , and from Select S The maximum response value, each response value corresponds to a predicted position of the target, extract features for each predicted position and update the corresponding template, and finally generate S target candidate boxes; Repeat the module, for the subsequent k Frame No. j Candidate boxes , after repeatedly reading the module and updating the module, we get candidate boxes; K Frame, repeatedly read the module and update the module to get candidate boxes, at this time all the candidate boxes are generated to form a multi-frame tree structure; The backtracking module uses the trajectory backtracking module to reversely find the estimated position of the target in each frame in the tree structure obtained by the repeating module, starting from the K Frame No. Candidate frames are started to backtrack, and a total of Candidate paths of target trajectories; The tracking module calculates the value function corresponding to each candidate path obtained by the backtracking module, selects the candidate path with the largest value function as the estimated path of the target in a batch, and outputs the estimated state of the target in each frame.

[0079] In another embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy, including: Include a M The dataset of frame video SAR images is divided into NBatch packages, each containing K Frame image; the tracking target is manually selected in the initial frame of the first batch package, and the background-aware correlation filter extracts the features of the initial position F 1 , get the initialization parameters and initial filter template of the background-aware correlation filter , generate an initial target candidate box ; Read the second frame image in the first batch, based on the candidate box of the first frame Define the target search area for the second frame , Divided into multiple test samples by BACF, each test sample and Calculate the similarity and obtain a response matrix, select the response matrix with the maximum response value ; and from Select S The maximum response value, each response value corresponds to a predicted position of the target; extract features for each predicted position and update the corresponding template, and finally generate S target candidate box; for the subsequent k Frame No. j Candidate boxes , repeat the above steps to get candidate boxes; K Frame, repeat the above steps to get candidate frames, and at the end, all candidate frames are generated to form a multi-frame tree structure; the trajectory backtracking algorithm is used to reversely find the estimated position of the target in each frame in the tree structure, starting from the first K Frame No. Candidate frames are started to backtrack, and a total of The candidate paths of the target trajectory are calculated; the value function corresponding to each candidate path is calculated, and the candidate path with the largest value function is selected as the estimated path of the target in a batch package, and the estimated state of the target in each frame is output.

[0080] In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device, and can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that more specific examples (non-exhaustive list) of the computer-readable storage medium here include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0081] Computer readable storage media also include data signals propagated in baseband or as part of a carrier wave, which carry readable program codes. Such propagated data signals can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium can also be any readable medium other than a readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program code contained on the readable storage medium can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0082] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0083] The processor may load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy in the above embodiment; the processor may load and execute the following steps of one or more instructions in the computer-readable storage medium: Include a M The dataset of frame video SAR images is divided into N Batch packages, each containing K Frame image; the tracking target is manually selected in the initial frame of the first batch package, and the background-aware correlation filter extracts the features of the initial position F 1 , get the initialization parameters and initial filter template of the background-aware correlation filter , generate an initial target candidate box ; Read the second frame image in the first batch, based on the candidate box of the first frame Define the target search area for the second frame , Divided into multiple test samples by BACF, each test sample and Calculate the similarity and obtain a response matrix, select the response matrix with the maximum response value ; and from Select S The maximum response value, each response value corresponds to a predicted position of the target; extract features for each predicted position and update the corresponding template, and finally generate S target candidate box; for the subsequent k Frame No. j Candidate boxes , repeat the above steps to get candidate boxes; K Frame, repeat the above steps to get candidate frames, and at the end, all candidate frames are generated to form a multi-frame tree structure; the trajectory backtracking algorithm is used to reversely find the estimated position of the target in each frame in the tree structure, starting from the first K Frame No. Candidate frames are started to backtrack, and a total of The candidate paths of the target trajectory are calculated; the value function corresponding to each candidate path is calculated, and the candidate path with the largest value function is selected as the estimated path of the target in a batch package, and the estimated state of the target in each frame is output.

[0084] See also Figure 4, the terminal device is a computer device, and the computer device 60 of this embodiment includes: a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the computer program 63 is executed by the processor 61, the method for calculating the composition of fluid in the reservoir transformation wellbore in the embodiment is realized. To avoid repetition, it is not described one by one here. Alternatively, when the computer program 63 is executed by the processor 61, the functions of each model / unit in the calculation system for the composition of fluid in the reservoir transformation wellbore in the embodiment are realized. To avoid repetition, it is not described one by one here.

[0085] The computer device 60 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will appreciate that Figure 4 This is only an example of the computer device 60 and does not constitute a limitation of the computer device 60. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.

[0086] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, central processing units, graphics processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, data processing logic based on quantum computing, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0087] The memory 62 may be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 may also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 60.

[0088] Furthermore, the memory 62 may include both an internal storage unit of the computer device 60 and an external storage device. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 may also be used to temporarily store data that has been output or is to be output.

[0089] Any reference to a memory, database or other medium used in the embodiments provided in this application may include at least one of a non-volatile and a volatile memory. Non-volatile memory may include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical memory, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetic random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. Volatile memory may include a random access memory (RAM) or an external cache memory, etc. As an illustration and not limitation, RAM may be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0090] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0091] See also Figure 5 The terminal device 600 is an electronic device, and the electronic device is in the form of a general computing device. The components of the electronic device may include but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0092] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 performs the steps according to various exemplary embodiments of the present invention described in the above method section of this specification. For example, the processing unit 610 can perform the following steps: Figure 3 Follow the steps shown in .

[0093] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 6201 and / or a cache memory unit 6202 , and may further include a read-only memory unit (ROM) 6203 .

[0094] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0095] Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0096] The electronic device 600 may also communicate with one or more external devices 700 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 650. Furthermore, the electronic device 600 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 via a bus 630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0097] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings here can usually be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of 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 are within the scope of protection of the present invention. Specific embodiments Input Data The method of the present invention is tested on a video SAR dataset published by Sandia National Laboratories in the United States, including Victr and Highway datasets; the Victr dataset has a total of 398 pictures, and only one target is driving on the highway, and the image resolution is 324*267; the Highway dataset has a total of 300 frames of pictures, in which there are multiple targets turning on the highway, and the image resolution is 660*720. In the implementation example (Highway) of the present invention, a total of 5 targets are selected for tracking.

[0099] Parameter initialization The parameters of the entire algorithm are all taken from the optimal values ​​in the experiment, where the number of frames in each batch is , the number of maximum response values ​​selected , a multiplication factor of the search area , online learning rate , , the intensity difference threshold .

[0100] Evaluation indicators The evaluation indexes used in the present invention are: Success rate Indicates The target box tracked by the frame algorithm, Represents the real target box, overlap rate Defined as:

[0101] in, , represents the intersection and union of two target boxes, and |∙| represents the number of pixels in the region. If the overlap rate is greater than a given threshold (0.5), the tracking of the current frame is considered successful. The success rate is defined as the percentage of successfully tracked frames to the total number of frames.

[0102] CLE and precision Center Location Error (CLE) is defined as the center coordinate of the predicted target box The center coordinates of the real target frame The average Euclidean distance between them is expressed as:

[0103] The accuracy is defined as the ratio of the number of accurately tracked frames to the total number of frames in the dataset. When the center position error CLE is less than a certain threshold, the current frame can be considered to be accurately tracked.

[0104] FPS (Frames Per Second, FPS) The processing time of the tracking algorithm is one of the important indicators for evaluating the real-time performance of the algorithm. FPS refers to the number of images processed by the tracking algorithm in one second.

[0105] Tracking results display In the table, red represents the best, green represents the second, and blue represents the third.

[0106] Table 1 Results and comparison of algorithms on the Victr dataset

[0107] In summary, the video SAR target tracking method and system based on the spatiotemporal expansion-accumulation strategy of the present invention has the following characteristics: (1) The method of accumulating multiple frames of information is used to track weak shadow targets in video SAR. Since the video SAR target exists in the surveillance area for a long time and continuously, but the clutter is randomly distributed and does not exist for a long time, after accumulating multiple frames of image information, enough real target information can be obtained, and the target can be successfully distinguished from the clutter.

[0108] (2) Build a multi-frame tree structure to deal with the multi-peak problem in the correlation filter response graph. By generating multiple candidate boxes, the real target is covered as much as possible. Even if the candidate box of the real target in a certain frame does not have the maximum response value, the real trajectory of the target can still be successfully found after expanding the tree structure in the time domain and space domain.

[0109] (3) The online learning rate of the correlation filter is flexibly adjusted based on the change of the target appearance. By comparing the grayscale difference between the first frame of the subsequent batch and the initial frame of the first batch, the appropriate online learning rate is adaptively selected to update the template, thereby improving the algorithm's ability to adapt to different tracking scenarios. During long-term video SAR target tracking, the appearance characteristics of the target change significantly due to factors such as deformation and illumination changes. At this time, the template update method proposed in the present invention can better adapt to these changes and achieve long-term target tracking.

[0110] (4) It has the ability to continuously and stably track the target, and has achieved excellent tracking accuracy and success rate, as well as very small positioning error, achieving excellent performance compared with other correlation filtering-based algorithms.

[0111] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0112] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0113] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0114] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0115] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0116] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0117] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0118] 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.

[0119] 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 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0120] 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 A step that specifies a function in one or more boxes.

[0121] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A video SAR target tracking method based on a spatiotemporal expansion-accumulation strategy, characterized in that: The following steps are involved: S1. M The video SAR image dataset of frames is divided into N batch packages, each of which contains K Frame video SAR image; S2: Manually select the tracking target in the initial frame of the first batch package obtained in step S1, and extract the features of the initial position using the background-aware correlation filter. F 1. Obtain the initialization parameters and initial filter template of the background perception correlation filter , generate an initial candidate box ; S3, read the second frame image in the first batch package obtained in step S1, based on the candidate frame of the first frame obtained in step S2 Define the target search area for the second frame , The background-aware correlation filter is divided into multiple test samples, each test sample and Calculate similarity and obtain a response matrix; S4. Select the response matrix with the maximum response value from the response matrix obtained in step S3. , from the response matrix Select S The maximum response value, each response value corresponds to a predicted position of the target; extract features for each predicted position and update the corresponding template, and finally generate S Candidate boxes; S5, for the subsequent k Frame No. j Candidate boxes , repeat steps S3 to S4 to obtain candidate boxes; K Frame, repeat step S3 to step S4 to obtain candidate frames, and at the end all candidate frames are generated to form a multi-frame tree structure; S6, using the trajectory backtracking to reversely search for the estimated position of the target in each frame in the tree structure obtained in step S5, K Frame No. Candidate frames are started to backtrack, and a total of Candidate paths of target trajectories; S7. Calculate the value function corresponding to each candidate path obtained in step S6, select the candidate path with the largest value function as the predicted path of the target in a batch package, and output the estimated state of the target in each frame.

2. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 1 is characterized in that: In step S1, the last frame of the previous batch package and the first frame of the next batch package are the same frame. The target initial state of the first frame of the next batch package is the target estimation state of the previous batch package after screening by the trajectory backtracking module. The corresponding candidate box information including position-size state vector, features, templates, and response values ​​are all from the last frame of the previous batch package.

3. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 1 is characterized in that: In step S2, the feature F 1 is a 32-dimensional feature, including a 31-dimensional directional histogram feature and a grayscale feature; The target tracking box is , The superscript indicates the number of the current candidate box, and the subscript indicates the frame number of the image in the current batch package.

4. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 1 is characterized in that: In step S3, the target search area The location is based on Position generation, target search area The size is right The position-size state vector is enlarged in both width and height Double generation, is an adjustable parameter with a range of [4,8]; the resulting response matrix is the matrix with the maximum value among all response matrices.

5. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 1 is characterized in that: In step S4, each response value corresponds to the predicted position of the target and satisfies the following relationship: in,( ) are the response values The corresponding s The coordinates of the predicted target locations.

6. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 1 is characterized in that: In step S5, the tree structure constituting the multiple frames is specifically: S501, initial target candidate box and the position-size state vector of the candidate box ; S502: Generate a search area in the second frame , The background-aware correlation filter is used to crop the sample into multiple test samples, each with the same size and dimension. Same, each test sample and Perform cross-correlation similarity calculation and obtain a response matrix , select the response matrix with the maximum response value , It is a two-dimensional matrix. The value in each grid is between 0 and 1, and each grid corresponds to a position coordinate of the current frame. S503. Generate response matrix from search area Select the front S The maximum response value is obtained S target candidate boxes, the size of each candidate box is determined by the scale adjustment factor of the background-aware correlation filter; S504: For any frame in the first batch processing package k -1, from k -Select one from all candidate frames in 1 frame , After repeating step S502 and step S503, k Frame Generation candidate boxes, represented as , ; And so on, in the last frame of the first batch package ( k=K )Total The candidate box of a target is represented as , ; Store all candidate box information in the first batch package in the buffer area, where the batch package number n When >1, the candidate box information of the initial frame directly retrieves the target estimation result of the last frame of the previous batch package, and the target candidate box generation process of the remaining frames is the same as that of the first batch package.

7. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 6 is characterized in that: Adopting the adaptive template updating method, taking the initial frame of the first batch package as the benchmark, the grayscale mean difference between the first frame of the nth batch package and the initial frame of the first batch package is calculated, and the difference is used as the adjustment of the online learning rate. variables.

8. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 7 is characterized in that: Online learning rate for: in, is the grayscale difference index, is the threshold value, and is the value of the online learning rate.

9. The video SAR target tracking method based on the spatiotemporal expansion-accumulation strategy according to claim 1, characterized in that: In step S7, from Select the function with the largest value from the candidate paths The path is: in, Represents the set of all candidate paths in a batch package, is the estimated position of the target for all frames estimated by the algorithm, Estimate the position of the target for the Kth frame, All predicted positions for the targets in a batch.

10. A video SAR target tracking system based on spatiotemporal expansion-accumulation strategy, characterized in that: include: Divide the module into a M The dataset of frame video SAR images is divided into N Batch packages, each containing K Frame video SAR image; The selection module manually selects the tracking target in the initial frame of the first batch package obtained by the division module, and the background-aware correlation filter extracts the features of the initial position F 1. Obtain the initialization parameters and initial filter template of the background perception correlation filter , generate an initial candidate box ; The reading module reads the second frame image in the first batch package obtained by the segmentation module, based on the candidate box of the first frame obtained by the selection module Define the target search area for the second frame , The background-aware correlation filter is divided into multiple test samples, each test sample and Calculate similarity and obtain a response matrix; Update module, select the response matrix with the maximum response value from the response matrix obtained by the reading module , from the response matrix Select S The maximum response value, each response value corresponds to the predicted position of the target; extract features for each predicted position and update the corresponding template, and finally generate S Candidate boxes; Repeat the module, for the subsequent k Frame No. j Candidate boxes , after repeatedly reading the module and updating the module, we get candidate boxes; K Frame, repeatedly read the module and update the module to get candidate frames, and at the end all candidate frames are generated to form a multi-frame tree structure; The backtracking module uses the trajectory backtracking algorithm to reversely find the estimated position of the target in each frame in the tree structure obtained by the repeated module. K Frame No. Candidate frames are started to backtrack, and a total of Candidate paths of target trajectories; The tracking module calculates the value function corresponding to each candidate path obtained by the backtracking module, selects the candidate path with the largest value function as the estimated path of the target in a batch package, and outputs the estimated state of the target in each frame.

Images