Underwater target tracking method and system based on adaptive trajectory correction

Abstract

The invention discloses an underwater target tracking method and system based on adaptive trajectory correction, and the method comprises the steps: constructing an underwater visual target tracking data set, and carrying out the marking of a target position and scene attributes, and obtaining a marked target tracking data set, the underwater visual target tracking data set comprises a plurality of types of underwater target video sequences; an underwater image enhancement module and a self-adaptive trajectory correction module are introduced, and a self-adaptive trajectory correction underwater target tracking model is constructed; and performing target tracking processing on the labeled target tracking data set based on the adaptive trajectory correction underwater target tracking model to obtain a final target tracking result. According to the invention, through image quality optimization and trajectory data stabilization, the tracking precision and robustness of a complex underwater scene are improved. The underwater target tracking method and system based on adaptive trajectory correction can be widely applied to the technical field of underwater visual target tracking.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater visual target tracking, and in particular to an underwater target tracking method and system based on adaptive trajectory correction. Background Art

[0002] Underwater visual object tracking (UVOT), a cutting-edge field in computer vision, aims to continuously locate underwater objects using video sequences. It has important applications in areas such as marine life conservation and underwater operations. However, compared to traditional open-air scenes, underwater environments present significant challenges: due to water absorption and scattering, underwater images generally suffer from low contrast, color distortion, and low resolution, significantly increasing the difficulty of feature extraction.

[0003] Related underwater tracking methods include: domain adaptation methods (such as Ustark and UOSTrack) alleviate environmental differences through image enhancement or hybrid training, but do not solve the problem of dynamic drift during the tracking process; spatiotemporal fusion methods (such as ODTrack and AQATrack) use the Transformer architecture to model temporal information, but have high computational complexity and are difficult to achieve real-time performance on embedded devices; template update strategies (such as UpdateNet and ARTrack) rely on confidence scores to update templates, but are prone to introducing noise in low-quality underwater images, resulting in a decrease in template credibility. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide an underwater target tracking method and system based on adaptive trajectory correction, which can improve the tracking accuracy and robustness of complex underwater scenes through image quality optimization and trajectory data stabilization.

[0005] The first technical solution adopted by the present invention is: an underwater target tracking method based on adaptive trajectory correction, comprising the following steps:

[0006] Constructing an underwater visual target tracking dataset and performing target position and scene attribute annotation processing to obtain an annotated target tracking dataset, wherein the underwater visual target tracking dataset includes underwater target video sequences of several categories;

[0007] Introducing underwater image enhancement module and adaptive trajectory correction module to build an adaptive trajectory correction underwater target tracking model;

[0008] The target tracking data set after annotation is processed based on the adaptive trajectory correction underwater target tracking model to obtain the final target tracking result.

[0009] Furthermore, the adaptive trajectory correction underwater target tracking model specifically includes an underwater image enhancement module, a tracking module and an adaptive trajectory correction module. The underwater image enhancement module, the tracking module and the adaptive trajectory correction module are connected in sequence, and the underwater image enhancement module adopts a FA+Net two-stage enhancement structure.

[0010] Furthermore, the step of performing target tracking processing on the labeled target tracking data set based on the adaptive trajectory correction underwater target tracking model to obtain a final target tracking result specifically includes:

[0011] The labeled target tracking dataset is input into the adaptive trajectory correction underwater target tracking model;

[0012] An underwater image enhancement module based on an adaptive trajectory correction underwater target tracking model performs image enhancement processing on the annotated target tracking dataset to obtain an enhanced target tracking dataset;

[0013] The tracking module based on the adaptive trajectory correction underwater target tracking model performs specified tracking on the enhanced target tracking dataset and obtains the quadruple coordinates of the tracking results;

[0014] The adaptive trajectory correction module of the underwater target tracking model based on the adaptive trajectory correction performs adaptive trajectory correction on the quadruple coordinates of the tracking result to obtain the final target tracking result.

[0015] Furthermore, the underwater image enhancement module based on the adaptive trajectory correction underwater target tracking model performs image enhancement processing on the labeled target tracking dataset to obtain an enhanced target tracking dataset, which specifically includes:

[0016] Input the labeled target tracking dataset into the underwater image enhancement module;

[0017] Determine whether the underwater target video sequence in the labeled target tracking dataset is the first frame;

[0018] If the underwater target video sequence is the first frame, the labeled target tracking dataset is subjected to image enhancement processing, and the image similarity values ​​before and after enhancement are obtained. The enhancement signal of the underwater target video sequence corresponding to the image similarity values ​​before and after enhancement that meet the preset enhancement interval value range is set to 1;

[0019] If the underwater target video sequence is not the first frame, the value of the enhancement signal of the underwater target video sequence is determined. If the enhancement signal of the underwater target video sequence is 1, the underwater target video sequence is subjected to image enhancement processing. If the enhancement signal of the underwater target video sequence is not 1, the underwater target video sequence is discarded.

[0020] Until all underwater target video sequences in the labeled target tracking dataset are traversed, an enhanced target tracking dataset is obtained.

[0021] Furthermore, the step of obtaining the image similarity values ​​before and after enhancement specifically includes:

[0022] Perform brightness contrast calculation on the images before and after enhancement to obtain brightness contrast similarity between the images before and after enhancement;

[0023] Performing contrast comparison calculation on the images before and after enhancement to obtain the contrast comparison similarity of the images before and after enhancement;

[0024] Performing structural comparison calculation on the images before and after enhancement to obtain structural comparison similarity of the images before and after enhancement;

[0025] The image similarity value before and after enhancement is determined by combining the brightness contrast similarity of the image before and after enhancement, the contrast contrast similarity of the image before and after enhancement, and the structure contrast similarity of the image before and after enhancement.

[0026] Furthermore, the adaptive trajectory correction module based on the adaptive trajectory correction underwater target tracking model performs adaptive trajectory correction on the quadruple coordinates of the tracking result to obtain the final target tracking result, which specifically includes:

[0027] Input the quaternion coordinates of the tracking result into the adaptive trajectory correction module;

[0028] Determine whether the underwater target video sequence corresponding to the quadruple coordinates of the tracking result is the first frame. If it is the first frame, track the next frame.

[0029] If it is not the first frame, the Euclidean metric between the center coordinate position of the previous frame of the corresponding underwater target video sequence and the center coordinate position of the current frame is obtained as the displacement offset of the tracking target in the current frame;

[0030] Constructing a correction threshold, where the correction threshold represents the sum of the Euclidean distances of the first t frames of the corresponding underwater target video sequence;

[0031] If the displacement offset of the tracking target in the current frame is less than or equal to the correction threshold, there is no need to correct the quadruple coordinates of the tracking result, and the correction threshold is adaptively adjusted according to the displacement offset of the tracking target in the current frame;

[0032] If the displacement offset of the tracking target in the current frame is greater than the correction threshold, the four-tuple coordinates of the tracking result are corrected through Kalman filtering;

[0033] When the number of consecutive occurrences of the quadruple coordinates of the tracking results that do not require correction processing reaches a preset number threshold, and all frames of the corresponding underwater target video sequence are traversed, the final target tracking result is output.

[0034] Furthermore, the step of correcting the four-tuple coordinates of the tracking result by using Kalman filtering specifically includes:

[0035] The historical coordinate data within the t-frame range of the underwater target video sequence is predicted by Kalman filtering to obtain the predicted coordinate position;

[0036] The predicted coordinate position is replaced with the quadruple coordinates of the current tracking result, and the width and height of the bounding box of the current frame are retained to achieve correction processing of the quadruple coordinates of the tracking result.

[0037] The second technical solution adopted by the present invention is: an underwater target tracking system based on adaptive trajectory correction, comprising:

[0038] The first module is used to construct an underwater visual target tracking dataset and perform target position and scene attribute annotation processing to obtain an annotated target tracking dataset. The underwater visual target tracking dataset includes several categories of underwater target video sequences;

[0039] The second module is used to introduce the underwater image enhancement module and the adaptive trajectory correction module to build an adaptive trajectory correction underwater target tracking model;

[0040] The third module is used to perform target tracking processing on the labeled target tracking data set based on the adaptive trajectory correction underwater target tracking model to obtain the final target tracking result.

[0041] The beneficial effects of the method and system of the present invention are as follows: the present invention constructs an underwater visual target tracking dataset and performs target position and scene attribute annotation processing, further introduces an underwater image enhancement module and an adaptive trajectory correction module, constructs an adaptive trajectory correction underwater target tracking model, constructs a collaborative framework including the underwater image enhancement module and the adaptive trajectory correction module, improves tracking accuracy through image quality optimization and trajectory data stabilization, and finally performs target tracking processing on the annotated target tracking dataset based on the adaptive trajectory correction underwater target tracking model to obtain a final target tracking result, thereby improving the tracking accuracy and robustness of complex underwater scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flowchart of the steps of an underwater target tracking method based on adaptive trajectory correction of the present invention;

[0043] Figure 2It is a structural block diagram of an underwater target tracking system based on adaptive trajectory correction of the present invention;

[0044] Figure 3 1 is a schematic diagram of the steps of underwater target tracking provided by a specific embodiment of the present invention;

[0045] Figure 4 1 is a schematic structural diagram of an underwater target tracking model with adaptive trajectory correction provided by a specific embodiment of the present invention;

[0046] Figure 5 1 is a schematic diagram of the data processing flow of the underwater image enhancement module provided by a specific embodiment of the present invention;

[0047] Figure 6 1 is a schematic diagram of a data processing flow for adaptive trajectory correction provided by a specific embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are provided for ease of description only and do not limit the order of the steps. The order of execution of the steps in the embodiments can be adaptively adjusted based on the understanding of those skilled in the art.

[0049] First, it should be noted that the lack of collaborative optimization in related technologies has led to the independent design of image enhancement and tracking modules, failing to form a closed-loop optimization. For example, UOSTrack's enhancement method only serves as a preprocessing step and is not linked to the trajectory correction module. Poor dynamic adaptability makes the traditional fixed threshold correction method unable to adapt to changes in the target motion pattern. The low credibility of the template makes the online update strategy susceptible to occlusion or interference, resulting in template library contamination, further exacerbating the risk of tracking failure.

[0050] Based on this, the embodiment of the present invention improves tracking accuracy by constructing a collaborative framework including an underwater image enhancement module and an adaptive trajectory correction module through image quality optimization and trajectory data stabilization. The underwater image enhancement module dynamically screens frames that need to be enhanced by calculating structural similarity, and uses the FA+Net network to improve image quality in real time; the adaptive trajectory correction module dynamically adjusts the correction threshold using historical trajectory data, and combines Kalman filtering to predict coordinate offsets. Among them, the dynamic threshold correction mechanism is a dynamic threshold calculation method based on the displacement of the previous t frames. When the current frame displacement exceeds the threshold, the Kalman filter prediction is triggered to achieve adaptive drift correction; the trusted template update strategy introduces the concept of "stationary period" and updates the trusted template library through a template sequence of consecutive ST frames with no drift. At the end of each "stationary period", the intermediate frame template is selected for update to avoid noise interference.

[0051] Reference Figure 1The present invention provides an underwater target tracking method based on adaptive trajectory correction, the method comprising the following steps:

[0052] S100, constructing an underwater visual target tracking dataset and performing target position and scene attribute annotation processing to obtain an annotated target tracking dataset, wherein the underwater visual target tracking dataset includes underwater target video sequences of several categories;

[0053] In this embodiment, an underwater visual target tracking dataset is constructed, which contains video sequences of multiple types of underwater targets (such as marine life and underwater equipment). The target positions and scene attributes (such as occlusion and deformation) are annotated and divided into training and test sets.

[0054] S200, introducing an underwater image enhancement module and an adaptive trajectory correction module to construct an adaptive trajectory correction underwater target tracking model;

[0055] In this embodiment, if Figure 4 As shown, the adaptive trajectory correction underwater target tracking model specifically includes an underwater image enhancement module, a tracking module and an adaptive trajectory correction module. The underwater image enhancement module, the tracking module and the adaptive trajectory correction module are connected in sequence, and the underwater image enhancement module adopts a FA+Net two-stage enhancement structure.

[0056] S300 , performing target tracking processing on the labeled target tracking data set based on the adaptive trajectory correction underwater target tracking model to obtain a final target tracking result.

[0057] S310, inputting the labeled target tracking data set into an adaptive trajectory correction underwater target tracking model;

[0058] S320, performing image enhancement processing on the labeled target tracking dataset by an underwater image enhancement module based on the adaptive trajectory correction underwater target tracking model to obtain an enhanced target tracking dataset;

[0059] First, it should be noted that if it is the first frame of each video sequence to be tracked in the test set, underwater image enhancement is performed, and then the image similarity before and after enhancement is calculated based on structural similarity, and the video sequence to be enhanced is dynamically selected, that is, the enhanced signal F is determined. enhance When the calculated result is between 0.1 and 0.9, the enhancement is triggered and the extremely similar / different frames are discarded; if it is not the first frame, the enhanced signal F calculated based on the first frame is used. enhance To determine whether the subsequent frames need to be enhanced, if so, the underwater image enhancement is performed; otherwise, the ARTrack visual target tracking algorithm is used to track the specified target in the current frame.

[0060] Specifically, the labeled target tracking dataset is input into the underwater image enhancement module; it is determined whether the underwater target video sequence in the labeled target tracking dataset is the first frame; if the underwater target video sequence is the first frame, the labeled target tracking dataset is image enhanced, and the image similarity value before and after the enhancement is obtained, and the underwater target video sequence corresponding to the image similarity value before and after the enhancement that meets the preset enhancement interval numerical range is set to have an enhancement signal of 1; if the underwater target video sequence is not the first frame, the value of the enhancement signal of the underwater target video sequence is determined, if the enhancement signal of the underwater target video sequence is 1, the underwater target video sequence is image enhanced, if the enhancement signal of the underwater target video sequence is not 1, the underwater target video sequence is discarded; until all underwater target video sequences in the labeled target tracking dataset are traversed to obtain the enhanced target tracking dataset.

[0061] In this embodiment, if Figure 5 As shown, the underwater image enhancement module is executed, and the specific process is as follows:

[0062] 1) Determine whether it is the first frame of the kth (k>0) video sequence to be tracked in the test set. If so, go to step 2) first, then go to step 3). If not, go to step 4).

[0063] 2) Enhancement using a specific underwater image enhancement method. We introduce an efficient, lightweight, real-time underwater image enhancement network, FA+Net, with only 9k parameters, capable of enhancing 1080p images in real time. FA+Net employs a two-stage enhancement architecture. The strong prior stage aims to decompose the challenging underwater degradation into subproblems, while the fine-grained stage contains a multi-branch color enhancement module and a pixel attention module to enhance the network's perception of details.

[0064] 3) Calculate the image similarity before and after enhancement. If the calculation result Within the preset threshold range [a, b], the signal F will be enhanced enhance Set to 1, otherwise set to 0.

[0065] 4) Determine the enhanced signal F enhance Is it 1? If so, go to step 2) first, and then go to the step of using the ARTrack visual target tracking algorithm to track the specified target in the current frame.

[0066] It should be noted that obtaining the image similarity value before and after enhancement includes performing a brightness contrast calculation on the images before and after enhancement to obtain the brightness contrast similarity of the images before and after enhancement; performing a contrast contrast calculation on the images before and after enhancement to obtain the contrast contrast similarity of the images before and after enhancement; performing a structure contrast calculation on the images before and after enhancement to obtain the structure contrast similarity of the images before and after enhancement; and determining the image similarity value before and after enhancement by combining the brightness contrast similarity of the images before and after enhancement, the contrast contrast similarity of the images before and after enhancement, and the structure contrast similarity of the images before and after enhancement.

[0067] In some specific embodiments, a structural similarity index is used to measure the similarity between two images. It simulates the human visual system to evaluate image quality and is mainly used to compare the differences between the original image and the processed image. The calculation of the structural similarity index is mainly based on the following three aspects:

[0068] 1) Brightness contrast: Calculate the average brightness of two image blocks and compare their similarity. The brightness contrast formula is as follows:

[0069]

[0070] Among them, μ x and μ y are the average brightness of the two image blocks, and C1 is a constant used to avoid the denominator being zero.

[0071] 2) Contrast comparison: Calculate the contrast of two image blocks and compare their similarity. The contrast comparison formula is as follows:

[0072]

[0073] Among them, σ x and σ y are the average brightness of the two image blocks, and C2 is another constant used to avoid the denominator being zero.

[0074] 3) Structural comparison: Calculate the structural similarity of two image blocks. The structural comparison formula is as follows:

[0075]

[0076] Among them, μ xy is the covariance between two image patches, and C3 is a constant.

[0077] Finally, the calculation formula of the structural similarity index combines the above three aspects, and its expression is:

[0078] SSIM(x,y)=[L(x,y)] α [C(x,y)]β [S(x,y)] γ

[0079] Here, α, β, and γ are weight parameters, typically set to 1. In practical applications, the structural similarity metric is often used to evaluate the impact of image compression, image denoising, and image enhancement algorithms on image quality. It better reflects the human visual system's perception of image quality than traditional mean square error (MSE) and peak signal-to-noise ratio (PSNR).

[0080] The threshold for the final image enhancement method is determined by the structural similarity index mentioned above. The SSIM value range is [0, 1]. A higher SSIM value indicates a higher degree of similarity between the enhanced images; a lower SSIM value indicates a greater difference between the two images.

[0081] For the enhanced images required for tracking tasks, if the degree of similarity after enhancement is too high, it can be considered that the difference between the original and original images is not significant. Furthermore, the computational overhead of enhancement is considerable, so they should be discarded. If the degree of similarity after enhancement is too low, the difference is too large. This is likely to cause target distortion, and therefore should also be discarded. In summary, the threshold for image enhancement should be a range. We discarded the two extreme cases: the maximum 10% similarity and the minimum 10% similarity. The final enhancement range was [0.1, 0.9].

[0082] S330, a tracking module based on an adaptive trajectory correction underwater target tracking model performs specified tracking on the enhanced target tracking data set to obtain a quadruple coordinate of the tracking result;

[0083] In this embodiment, the ARTrack visual target tracking algorithm is used to track the specified target in the current frame to obtain the four-tuple coordinates of the target.

[0084] S340 , an adaptive trajectory correction module based on the adaptive trajectory correction underwater target tracking model performs adaptive trajectory correction on the quadruple coordinates of the tracking result to obtain a final target tracking result.

[0085] First, it's important to note that if the frame is the first frame of each video sequence to be tracked in the test set, the underwater image enhancement step is directly executed to track the next frame. Otherwise, the coordinates acquired during the current frame's tracking of the specified target using the ARTrack visual object tracking algorithm are corrected. This method doesn't filter candidate frames based on a fixed threshold, but instead uses historical trajectory data within a specific frame range for adaptive online correction. In each correction phase, the trajectory data within that range is used to adjust the correction threshold for the current frame, while also predicting the trajectory coordinates for the current frame.

[0086] We then designed a trajectory-based online module update strategy to dynamically update the initial template image. Specifically, unlike traditional template updates based on confidence scores, we use a trajectory-defined "stationary period" center template for updates, avoiding the emergence of untrustworthy templates. When the "stationary period" reaches a certain threshold, the update is performed at the end of the "stationary period."

[0087] Finally, determine whether there is a next frame in the video sequence. If so, output the current frame tracking result and then switch to the underwater image enhancement step to perform the next frame tracking; otherwise, output the current frame result and end.

[0088] Specifically, the four-tuple coordinates of the tracking result are input into the adaptive trajectory correction module; it is determined whether the underwater target video sequence corresponding to the four-tuple coordinates of the tracking result is the first frame, and if it is the first frame, the next frame is tracked; if it is not the first frame, the Euclidean metric of the center coordinate position of the previous frame of the corresponding underwater target video sequence and the center coordinate position of the current frame is obtained as the displacement offset of the tracking target in the current frame; a correction threshold is constructed, and the correction threshold represents the sum of the Euclidean distances of the previous t frames of the corresponding underwater target video sequence; if the displacement offset of the tracking target in the current frame is less than or equal to the correction threshold, there is no need to correct the four-tuple coordinates of the tracking result, and the correction threshold is adaptively adjusted according to the displacement offset of the tracking target in the current frame; if the displacement offset of the tracking target in the current frame is greater than the correction threshold, the four-tuple coordinates of the tracking result are corrected by Kalman filtering; when the number of consecutive times that the four-tuple coordinates of the tracking result do not need to be corrected reaches a preset number threshold, and all frames of the corresponding underwater target video sequence are traversed, the final target tracking result is output.

[0089] In some specific embodiments, Figure 6 As shown, the adaptive trajectory correction module is executed, and the specific process is as follows:

[0090] 1) Determine whether it is the first frame of each video sequence to be tracked in the test set. If so, directly execute the underwater image enhancement module step to perform tracking of the next frame; otherwise, go to step 2);

[0091] 2) Perform trajectory coordinate correction on the acquired coordinates. This method does not filter candidate frames based on a fixed threshold. Instead, it uses historical trajectory data within a specific frame range for adaptive online correction. During each correction phase, the trajectory data within that range is used to adjust the correction threshold for the current frame, while simultaneously predicting the trajectory coordinates for the current frame. Then, proceed to step 3) to perform trajectory template updates.

[0092] Correction threshold acquisition. Before adopting the trajectory correction method, the first problem that needs to be solved is how to adaptively determine whether the current frame tracking result needs trajectory correction based on historical trajectory information. In order to avoid unnecessary overhead caused by invalid correction and adverse results caused by malicious correction, it is necessary to use historical information as much as possible to design a reasonable judgment basis for judgment. By calculating the Euclidean metric between the center coordinate position of the previous frame and the center coordinate position of the current frame, the displacement offset of the tracking target in the current frame is simulated, and then compared with the correction threshold. Finally, based on the comparison result, it is decided whether the current frame should adopt trajectory correction.

[0093] The Euclidean metric (also known as Euclidean distance) is a commonly used distance definition that refers to the actual distance between two points in m-dimensional space. The Euclidean distance in two-dimensional and three-dimensional space is the actual distance between two points. It is defined as follows in the current method:

[0094]

[0095] where x pre 、y pre Respectively represent the horizontal and vertical coordinates of the center position of the target bounding box in the previous frame, x cur 、y cur It represents the horizontal and vertical coordinates of the center position in the target bounding box of the current frame. cur It represents the Euclidean distance of the current frame.

[0096] The correction threshold is defined as the sum of the Euclidean distances of the previous t frames, and its expression is:

[0097]

[0098] Where T cur represents the correction threshold, ρ i The correction threshold T is defined as cur It can adaptively adjust its value based on the t-1 Euclidean distances within the range of t frames. Specifically, when the target moves quickly within a certain period of time, the correction threshold will be adaptively adjusted to a larger value; when the target moves slowly, the correction threshold will be adaptively adjusted to a smaller value.

[0099] Further trajectory coordinate correction is performed. If the Euclidean distance ρ of the current frame cur Less than or equal to the correction threshold T cur , then no correction is required, the original tracking result is output, and normal visual target tracking continues. At the same time, the ρ of the current frame is cur Participate in the correction threshold T curThe adjustment is used for the next trajectory coordinate correction judgment, thus achieving an adaptive effect. Conversely, if trajectory correction is determined to be necessary, the Kalman filter is used to predict the historical coordinate data within the range of t frames and replace the predicted coordinate position with the current tracking coordinate position. As for the width and height of the bounding box of the current frame, we believe that the deformation within a frame does not change much, so we choose to directly inherit the width and height of the bounding box of the previous frame. In addition, considering that the Kalman filter itself requires training some parameters and the trajectories of the tracked targets in different video sequences vary, at the beginning of tracking for each video sequence, the Kalman filter will use the historical information of the previous s frames to learn the relevant parameters required for each video sequence, so as to ensure the stability and accuracy of the subsequent prediction results.

[0100] 3) Implement a trajectory-based online module update strategy to dynamically update the initial template image. Specifically, unlike traditional template updates based on confidence scores, we use a trajectory-defined "stationary period" center template for updates, avoiding the appearance of untrustworthy templates. When the "stationary period" reaches a certain threshold, the update is performed at the end of the "stationary period."

[0101] Trajectory template updates. This module contains a counter variable, count. It increments by 1 each time it passes through the Adaptive Trajectory Correction module. During this period, the template sequence automatically stores the template. However, after each trajectory coordinate correction, count is reset, and the template sequence begins storing templates anew. When ST frames of the Adaptive Trajectory Correction module are passed through without any trajectory coordinate correction, this ST-frame period is considered a "stationary period." At this point, the template sequence has automatically stored the templates during the "stationary period." Each trajectory template update occurs at the first trajectory coordinate correction after the "stationary period." This module reads the centralmost template stored during the "stationary period" to perform the template update.

[0102] 4) Determine whether there is a next frame in the video sequence. If so, output the current frame tracking result and then execute the underwater image enhancement module step to perform the next frame tracking. Otherwise, output the current frame result and end.

[0103] In summary, if Figure 3As shown, the embodiment of the present invention first designs an adaptive underwater image enhancement module (UIEM), dynamically selects image frames to be enhanced through structural similarity (SSIM), and uses the FA+Net network to improve image quality and stabilize trajectory data. It further constructs a dual-module adaptive trajectory correction framework (ATCM) including trajectory coordinate correction (TCC) and trajectory template update (TTU), dynamically adjusts the correction threshold using historical coordinate data, combines Kalman filtering to predict coordinate offsets, and updates the trusted template through a "stationary period update" strategy. It also designs an online trajectory data storage mechanism to maintain coordinate sequences and template sequences separately to support real-time correction. Through the coordinated optimization of underwater image enhancement and trajectory correction, the robustness and accuracy of underwater target tracking are significantly improved, achieving optimal performance on benchmark datasets such as UTB180, VMAT, and UVOT400, successfully solving the tracking problems in complex scenarios such as target deformation, occlusion, and low resolution.

[0104] Reference Figure 2 , an underwater target tracking system based on adaptive trajectory correction, comprising:

[0105] The first module 201 is used to construct an underwater visual target tracking dataset and perform target position and scene attribute annotation processing to obtain an annotated target tracking dataset, wherein the underwater visual target tracking dataset includes underwater target video sequences of several categories;

[0106] The second module 202 is used to introduce an underwater image enhancement module and an adaptive trajectory correction module to build an adaptive trajectory correction underwater target tracking model;

[0107] The third module 203 is used to perform target tracking processing on the labeled target tracking data set based on the adaptive trajectory correction underwater target tracking model to obtain a final target tracking result.

[0108] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0109] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. An underwater target tracking method based on adaptive trajectory correction, characterized in that: The following steps are involved: Constructing an underwater visual target tracking dataset and performing target position and scene attribute annotation processing to obtain an annotated target tracking dataset, wherein the underwater visual target tracking dataset includes underwater target video sequences of several categories; Introducing underwater image enhancement module and adaptive trajectory correction module to build an adaptive trajectory correction underwater target tracking model; The target tracking data set after annotation is processed based on the adaptive trajectory correction underwater target tracking model to obtain the final target tracking result.

2. The underwater target tracking method based on adaptive trajectory correction according to claim 1, characterized in that: The adaptive trajectory correction underwater target tracking model specifically includes an underwater image enhancement module, a tracking module and an adaptive trajectory correction module. The underwater image enhancement module, the tracking module and the adaptive trajectory correction module are connected in sequence, and the underwater image enhancement module adopts a FA+Net two-stage enhancement structure.

3. The underwater target tracking method based on adaptive trajectory correction according to claim 2, characterized in that: The step of performing target tracking processing on the labeled target tracking data set based on the adaptive trajectory correction underwater target tracking model to obtain a final target tracking result specifically includes: The labeled target tracking dataset is input into the adaptive trajectory correction underwater target tracking model; An underwater image enhancement module based on an adaptive trajectory correction underwater target tracking model performs image enhancement processing on the annotated target tracking dataset to obtain an enhanced target tracking dataset; The tracking module based on the adaptive trajectory correction underwater target tracking model performs specified tracking on the enhanced target tracking dataset and obtains the quadruple coordinates of the tracking results; The adaptive trajectory correction module based on the adaptive trajectory correction underwater target tracking model performs adaptive trajectory correction on the quadruple coordinates of the tracking result to obtain the final target tracking result.

4. The underwater target tracking method based on adaptive trajectory correction according to claim 3, characterized in that: The underwater image enhancement module based on the adaptive trajectory correction underwater target tracking model performs image enhancement processing on the labeled target tracking dataset to obtain an enhanced target tracking dataset, which specifically includes: Input the labeled target tracking dataset into the underwater image enhancement module; Determine whether the underwater target video sequence in the labeled target tracking dataset is the first frame; If the underwater target video sequence is the first frame, the labeled target tracking dataset is subjected to image enhancement processing, and the image similarity values ​​before and after enhancement are obtained. The enhancement signal of the underwater target video sequence corresponding to the image similarity values ​​before and after enhancement that meet the preset enhancement interval value range is set to 1; If the underwater target video sequence is not the first frame, the value of the enhancement signal of the underwater target video sequence is determined. If the enhancement signal of the underwater target video sequence is 1, the underwater target video sequence is subjected to image enhancement processing. If the enhancement signal of the underwater target video sequence is not 1, the underwater target video sequence is discarded. Until all underwater target video sequences in the labeled target tracking dataset are traversed, an enhanced target tracking dataset is obtained.

5. The underwater target tracking method based on adaptive trajectory correction according to claim 4, characterized in that: The step of obtaining the image similarity values ​​before and after enhancement specifically includes: Perform brightness contrast calculation on the images before and after enhancement to obtain brightness contrast similarity between the images before and after enhancement; Performing contrast comparison calculation on the images before and after enhancement to obtain the contrast comparison similarity of the images before and after enhancement; Performing structural comparison calculation on the images before and after enhancement to obtain structural comparison similarity of the images before and after enhancement; The image similarity value before and after enhancement is determined by combining the brightness contrast similarity of the image before and after enhancement, the contrast contrast similarity of the image before and after enhancement, and the structure contrast similarity of the image before and after enhancement.

6. The underwater target tracking method based on adaptive trajectory correction according to claim 5, characterized in that: The adaptive trajectory correction module based on the adaptive trajectory correction underwater target tracking model performs adaptive trajectory correction on the quadruple coordinates of the tracking result to obtain the final target tracking result. This step specifically includes: Input the quaternion coordinates of the tracking result into the adaptive trajectory correction module; Determine whether the underwater target video sequence corresponding to the quadruple coordinates of the tracking result is the first frame. If it is the first frame, track the next frame. If it is not the first frame, the Euclidean metric between the center coordinate position of the previous frame of the corresponding underwater target video sequence and the center coordinate position of the current frame is obtained as the displacement offset of the tracking target in the current frame; Constructing a correction threshold, where the correction threshold represents the sum of the Euclidean distances of the first t frames of the corresponding underwater target video sequence; If the displacement offset of the tracking target in the current frame is less than or equal to the correction threshold, there is no need to correct the quadruple coordinates of the tracking result, and the correction threshold is adaptively adjusted according to the displacement offset of the tracking target in the current frame; If the displacement offset of the tracking target in the current frame is greater than the correction threshold, the four-tuple coordinates of the tracking result are corrected through Kalman filtering; When the number of consecutive occurrences of the quadruple coordinates of the tracking results that do not require correction processing reaches a preset number threshold, and all frames of the corresponding underwater target video sequence are traversed, the final target tracking result is output.

7. The underwater target tracking method based on adaptive trajectory correction according to claim 6, characterized in that: The step of correcting the four-tuple coordinates of the tracking result by using Kalman filtering specifically includes: The historical coordinate data within the t-frame range of the underwater target video sequence is predicted by Kalman filtering to obtain the predicted coordinate position; The predicted coordinate position is replaced with the quadruple coordinates of the current tracking result, and the width and height of the bounding box of the current frame are retained to achieve correction processing of the quadruple coordinates of the tracking result.

8. An underwater target tracking system based on adaptive trajectory correction, characterized in that: Includes the following modules: The first module is used to construct an underwater visual target tracking dataset and perform target position and scene attribute annotation processing to obtain an annotated target tracking dataset. The underwater visual target tracking dataset includes several categories of underwater target video sequences; The second module is used to introduce the underwater image enhancement module and the adaptive trajectory correction module to build an adaptive trajectory correction underwater target tracking model; The third module is used to perform target tracking processing on the labeled target tracking data set based on the adaptive trajectory correction underwater target tracking model to obtain the final target tracking result.

Images