Long-time single-target tracking method and storage device

Abstract

The application belongs to the field of visual target tracking, and particularly relates to a long-time single-target tracking method and a storage device. The method comprises detecting a target frame in a current first frame image of a video, and displaying the detected target frame on an image interface; in response to a selection operation of a user on a certain target frame, taking a target in the target frame of the selection operation as a target to be tracked; checking target information of the target to be tracked; obtaining corresponding variables in the target information; according to the variables in the target information, selecting to execute a pure tracking mode or a detection tracking mode, adopting a target detection algorithm based on deep learning and a single-target tracking algorithm based on correlation filtering, and setting target position judgment, target scale judgment, target coordinate jump judgment and target disappearance judgment to determine a tracking mode and an algorithm of a current frame. According to the application, different algorithms are adopted according to target conditions, and the real-time tracking performance is greatly improved while the tracking robustness is ensured.

Description

Technical Field

[0001] This invention belongs to the field of visual target tracking, and particularly relates to a long-term single target tracking method and storage device. Background Technology

[0002] Single-target tracking technology belongs to the field of visual target tracking and has a wide range of applications in daily life, medicine, transportation, aerospace, and military. Long-term single-target tracking technology also needs to take into account situations such as re-capture after the target disappears, and significant changes in target size, lighting, and shape, making it more difficult to implement than short-term single-target tracking.

[0003] Existing long-term single-target tracking algorithms are typically based on deep learning or correlation filtering. The former has high robustness but poor real-time performance, while the latter has high real-time performance but poor robustness. To simultaneously satisfy real-time performance and robustness, tracking strategies combining correlation filtering tracking algorithms and target detection algorithms have emerged. However, these strategies are usually either too simple, failing to adequately consider the appropriate use of the algorithm in situations such as target disappearance or excessively small target scale, leading to frequent target loss and poor tracking robustness, or too complex, adding a large amount of extra computation and reducing real-time performance. Neither of these strategies can adequately balance tracking real-time performance and robustness, resulting in poor overall tracking performance over long periods. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] This invention addresses the above-mentioned problems by proposing a long-term single-target tracking method and storage device. Its purpose is to solve the problem that the tracking real-time performance and robustness cannot be balanced, resulting in poor overall tracking performance over long periods of time.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides a long-term single-target tracking method, comprising the following steps:

[0008] Detect the bounding box in the first frame of the video and display the detected bounding box in the image interface;

[0009] In response to a user's selection of a target box in the image interface, the target within the selected target box is taken as the target to be tracked.

[0010] Check the target information of the target to be tracked, wherein the target information includes at least: target disappearance information, target scale information, and target coordinate jump information;

[0011] Obtain the corresponding variables in the target information, wherein the corresponding variable for the target disappearance information is whether the target has disappeared, the corresponding variable for the target scale information is whether the target scale is too small, and the corresponding variable for the target coordinate jump information is whether the target has jumped.

[0012] Based on the variables in the target information, select to execute either pure tracking mode or detection tracking mode.

[0013] Furthermore, if the following conditions are met: the target has not disappeared, the target is not a small target, and the target has not changed direction, then the pure detection mode is used; otherwise, the detection and tracking mode is used.

[0014] Furthermore, the steps for performing pure tracking mode include:

[0015] A single-target tracking algorithm based on correlation filtering is executed. The target in the previous frame is used as a template. Template matching is performed in the current frame to search for the target to be tracked and obtain the target coordinates.

[0016] The target detection algorithm is executed once every N1 frames. The detection box that is closest to the target in the previous frame is selected as the target box and replaced with the target box generated by the single target tracking algorithm in this frame to obtain the target coordinates.

[0017] The target coordinates obtained in the above steps are subjected to jump judgment; if the coordinates do not jump, the coordinates obtained in the frame are put into an array, and the template is updated based on the tracking result of the frame, and then the pure tracking mode execution condition judgment is performed; if the coordinates jump, the coordinates of the frame are not recorded, and the pure tracking mode execution condition judgment is performed directly. If the pure tracking mode execution condition is met, the pure tracking mode is continued to be executed; otherwise, the pure tracking mode is exited and the detection tracking mode is executed.

[0018] Furthermore, the steps for implementing the detection and tracking mode include:

[0019] Check for information indicating that the target has disappeared;

[0020] If the target has not disappeared, execute the target detection algorithm and distance determination algorithm to obtain the target coordinates;

[0021] The target coordinates are checked for jumps. If the target coordinates do not jump, the calculated target coordinates are stored in an array, and then the execution condition of the detection and tracking mode is checked. If the coordinates jump, the execution condition of the detection and tracking mode is checked directly.

[0022] If the target disappears, check if the target disappeared at the edge. If the target disappeared at the edge, the program will not perform any operation in the subsequent N2 frame; if the target did not disappear at the edge, the program will not perform any operation in the subsequent N3 frame.

[0023] Perform condition judgment for the execution of the detection and tracking mode;

[0024] If the conditions for executing the detection and tracing mode are met, the detection and tracing mode will continue to be executed; otherwise, the detection and tracing mode will be exited and the pure tracing mode will be executed.

[0025] Furthermore, the target disappearance process includes edge disappearance and in-field disappearance.

[0026] Furthermore, the step for determining when the target disappears includes:

[0027] The target did not disappear by default in the initial frame;

[0028] Determine whether the target is not detected or a coordinate jump occurs within N4 consecutive frames;

[0029] If the target is not detected for N4 consecutive frames or a coordinate jump occurs, the target is determined to have disappeared.

[0030] Determine whether the target coordinates in the previous frame are at the image boundary.

[0031] If the target is at the image boundary, it is determined that the target edge has disappeared; otherwise, it is determined that the target has disappeared within the field of view.

[0032] If at least one of the N4 consecutive frames shows that the target can be detected and no coordinate jump occurs, then the target is determined not to have disappeared.

[0033] End of target disappearance judgment.

[0034] Furthermore, the coordinate jump determination step includes:

[0035] Check the target disappearance information and the number of coordinate jumps;

[0036] If the target has not disappeared, compare the difference between the coordinates in this frame and the coordinates in the previous frame;

[0037] If the difference is large, then perform two more target detections consecutively; compare the differences in coordinates in three consecutive tests; if the difference is large, determine that the coordinates have changed; otherwise, determine that the coordinates have not changed.

[0038] If the difference is small, then the coordinates are determined not to have changed.

[0039] If the target disappears, determine whether the target edge has disappeared. If the target edge has disappeared, perform another target detection and compare the difference between the two consecutive coordinates. If the difference is large, determine that the coordinates have changed; otherwise, determine that the coordinates have not changed.

[0040] If the target edge does not disappear, then perform two more target detections and compare the differences in the coordinates of the three consecutive detections. If the difference is large, then the coordinates are determined to have changed; otherwise, the coordinates are determined not to have changed.

[0041] End the jump judgment.

[0042] Furthermore, the single-target tracking algorithm based on correlation filtering is the KCF algorithm.

[0043] Furthermore, the target detection algorithm used is the YOLOv5 algorithm.

[0044] To achieve the above objectives, the present invention provides a storage device characterized in that it stores program instructions that can be executed by a processor, the program instructions being used to implement the long-term single-target tracking method described above.

[0045] (III) Beneficial Effects

[0046] The technical solution of this invention has the following advantages: It divides the tracking mode into a detection-tracking mode and a pure tracking mode, employs a target detection algorithm based on deep learning and a single-target tracking algorithm based on correlation filtering, and sets target position judgment, target scale judgment, target coordinate jump judgment, and target disappearance judgment to determine the tracking mode and algorithm for the current frame. By using different algorithms according to the target situation, the real-time tracking performance can be significantly improved while ensuring tracking robustness. Attached Figure Description

[0047] Figure 1 This is a flowchart of a long-term single-target tracking method disclosed in this invention.

[0048] Figure 2 This is an exemplary overall flowchart of a long-term single-target tracking method according to an embodiment of the present invention.

[0049] Figure 3 This is the execution flowchart for pure tracing mode.

[0050] Figure 4 This is a flowchart of the execution process for the detection and tracking mode.

[0051] Figure 5 Flowchart for determining target disappearance.

[0052] Figure 6 The flowchart shows the coordinate jump judgment process. Detailed Implementation

[0053] The technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of this disclosure.

[0054] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0055] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0056] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously, as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0057] Figure 1 , Figure 2 An exemplary flowchart of a long-duration single-target tracking method 100 according to an embodiment of the present disclosure is shown.

[0058] First, in step S101, the video to be detected is input, the target detection algorithm is executed on the first frame of the video to be detected, and the detected target boxes are displayed on the image interface.

[0059] The video images may be images captured in real time by a camera or recording device, or they may be video images obtained in advance by other means. This disclosure is not limited to the source or acquisition method of the video images. For example, they may be images directly captured by a camera, surveillance camera, drone, etc., or video images obtained after pre-processing by a computer, or video images provided by other individuals / organizations.

[0060] A frame in a video is a single image frame, the smallest unit, and the first frame is the first frame image played. In this embodiment, the target detection algorithm samples a large number of regions in the input video image, then determines whether these regions contain the target of interest, and adjusts the region boundaries to more accurately predict the true bounding box of the target. These region bounding boxes are called target boxes, and the target box obtained from the first frame image is displayed in the image interface.

[0061] In step S102, in response to the user's selection operation of a target box in the image interface, the target within the selected target box is taken as the target to be tracked.

[0062] Understandably, users or operators can select a target bounding box based on the graphical interface, and use the target within that box as the target to be tracked. If no target is detected or the detected target is not of interest to the operator, the program terminates.

[0063] In this embodiment, the target tracking modes are divided into pure tracking mode and detection tracking mode.

[0064] The pure tracking mode uses a single-target tracking algorithm for target tracking. The single-target tracking algorithm based on correlation filtering is executed in each frame, and the target position is corrected using a target detection algorithm. The target detection algorithm is executed only once every specified number of frames.

[0065] The aforementioned detection and tracking mode uses a target detection algorithm combined with target position determination to track targets. The target detection algorithm is executed in every frame, but the single target tracking algorithm is not executed.

[0066] Given that single-target tracking algorithms have good real-time performance but poor robustness, while target detection algorithms have poor real-time performance but good robustness, the real-time performance and robustness of the tracking strategy can be balanced by controlling the number of frames between the tracking frames.

[0067] The following steps will explain the execution conditions for pure tracking mode and detection tracking mode.

[0068] Step S103: Check the target information of the target to be tracked, wherein the target information includes at least: target disappearance information, target scale information, and target coordinate jump information.

[0069] Step S104: Obtain the corresponding variables in the target information, wherein the corresponding variable for the target disappearance information is whether the target has disappeared, the corresponding variable for the target scale information is whether the target scale is too small, and the corresponding variable for the target coordinate jump information is whether the target has jumped.

[0070] Step S105: Based on the variables in the target information, select to execute either pure tracking mode or detection tracking mode.

[0071] Specifically, in steps S103-S105, the target information is checked, including target disappearance information, target scale information, and target coordinate jump information. Each of these three pieces of target information corresponds to a specific variable: miss_flag, little_flag, and jump_flag. By default, miss_flag = False, indicating the target has not been lost; little_flag = False, indicating the target is not small; and jump_flag = False, indicating no coordinate jump. When all three variables are False, the frame executes pure tracking mode 200; otherwise, it executes detection tracking mode 300. If a next frame exists, steps S101-S105 are repeated; otherwise, tracking ends.

[0072] Understandably, this embodiment, by formulating multiple judgment strategies, makes the timing of calling the target detection algorithm and the correlation filtering-based target tracking algorithm more reasonable, which can not only ensure the robustness of target tracking and accurate target tracking, but also improve the real-time performance of target tracking. In addition, it can nest different types of target detectors and single-target trackers, and has strong scalability.

[0073] According to an embodiment of the present invention, an embodiment of executing a pure tracking mode 200 and executing a detection tracking mode 300 is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0074] The execution flow of the pure tracing mode 200 is as follows: Figure 3 As shown, the specific steps are as follows:

[0075] S201. Execute a single-target tracking algorithm based on correlation filtering. Using the target in the previous frame as a template, perform template matching in the current frame to search for the target to be tracked and obtain the target coordinates.

[0076] S202. Execute the target detection algorithm once every N1 frames, select the detection box closest to the target in the previous frame as the target box, and replace the target box generated by the single target tracking algorithm in this frame to obtain the target coordinates;

[0077] S203. Perform a jump judgment on the target coordinates obtained in steps S201 and S202. If the coordinates do not jump, put the coordinates obtained in the frame into array A, which is the target's real position array, used to store the final target coordinates obtained in each frame, and update the template based on the tracking result of the frame, and then perform the judgment on the execution conditions of pure tracking mode in steps S103-S105. If the coordinates jump, do not record the coordinates of the frame, and jump_flag = True, directly perform the judgment on the execution conditions of pure tracking mode in steps S103-S105. If the execution conditions of pure tracking mode are met, continue to execute pure tracking mode; otherwise, exit pure tracking mode and execute detection tracking mode.

[0078] The execution process of the detection and tracking mode 300 is as follows: Figure 4 As shown, the specific steps are as follows:

[0079] S301. Check for target disappearance information;

[0080] S302. If the target has not disappeared, i.e., miss_flag = False, then execute the target detection algorithm and distance determination algorithm to obtain the target coordinates;

[0081] S303. Perform a jump judgment on the target coordinates. If the target coordinates do not jump, store the calculated target coordinates into array A, and then perform the judgment on the execution conditions of the detection and tracking mode in steps S103-S105. If the coordinates jump, then directly perform the judgment on the execution conditions of the detection and tracking mode in steps S103-S105.

[0082] S304. If the target disappears, i.e., miss_flag = True, then check whether the target disappears at the edge. If the target disappears at the edge, i.e., edgeMiss_flag = True, then the program in the subsequent N2 frame does not perform any operation; if the target does not disappear at the edge, i.e., the target disappears within the field of view, edgeMiss_flag = False, then the program in the subsequent N3 frame does not perform any operation.

[0083] S305. Perform a condition check for the execution of the detection and tracking mode;

[0084] S306. If the conditions for executing the detection and tracing mode are met, continue executing the detection and tracing mode; otherwise, exit the detection and tracing mode and execute the pure tracing mode.

[0085] In step S304, the target disappearance includes edge disappearance and in-view disappearance. Edge disappearance refers to the target disappearing from the image boundary, with its coordinates close to the image boundary line before disappearance. Edge disappearance corresponds to the variable edgeMiss_flag, which defaults to False, indicating that the target has not disappeared at the edge. If miss_flag = True and edgeMiss_flag = false, it means the target has disappeared within the view; if miss_flag = True and edgeMiss_flag = false, it means the target has disappeared at the edge; there is no case where miss_flag = False and edgeMiss_flag = True.

[0086] After a target disappears, it usually takes a relatively long time to reappear, especially when it disappears at the edge. Skipping N2 or N3 frames can reduce unnecessary calculations and improve the real-time performance of the algorithm.

[0087] It should be further explained that "disappearance within the field of view" refers to the target disappearing in the middle of the image, which is mostly due to occlusion, while "disappearance at the edge" refers to the target disappearing at the boundary of the image, which is mostly due to flying out of the field of view. For both "disappearance within the field of view" and "disappearance at the edge" scenarios, the target detection algorithm is no longer executed every frame, but rather once every specified number of frames, thereby improving real-time performance.

[0088] The target disappearance judgment process 400 will be explained in detail below, such as... Figure 5 As shown, the specific steps are as follows:

[0089] Step S401: In the initial frame, the target is assumed to have not disappeared, i.e., miss_flag = False.

[0090] Step S402: Determine whether the target cannot be detected or whether a coordinate jump occurs within N4 consecutive frames.

[0091] Step S403: If the target is not detected for N4 consecutive frames or a coordinate jump occurs, the target is determined to have disappeared, and miss_flag = True.

[0092] Step S404: Determine whether the target coordinates in the previous frame are at the image boundary.

[0093] Step S405: If the target is at the image boundary, then the target edge is determined to disappear, and edgeMiss_flag = True; otherwise, the target is determined to disappear within the field of view.

[0094] Step S406: If at least one of the N4 consecutive frames can detect the target and no coordinate jump occurs, then it is determined that the target has not disappeared.

[0095] S407, End target disappearance judgment.

[0096] In steps S401-S407, if the target detection algorithm fails to detect a target, and if it is in pure tracking mode, the target detection algorithm will be executed in the subsequent (N4-1) frames, and the single target tracking algorithm will not be executed; if it is in detection-tracking mode, the target detection will continue in the subsequent (N4-1) frames.

[0097] The coordinate jump judgment process 500 described below will be explained in detail, such as... Figure 6 As shown, the specific steps are as follows:

[0098] Step S501: Check the target disappearance information and the number of coordinate jumps.

[0099] Step S502: If the target has not disappeared, i.e., miss_flag = False, then compare the difference between the coordinates of this frame and the coordinates of the previous frame.

[0100] Step S503: If the difference is large, perform two more target detections consecutively.

[0101] Step S504: Compare the differences between the coordinates in three consecutive measurements.

[0102] Step S505: If the difference is large, determine that the coordinates have changed and jump_flag = True; otherwise, determine that the coordinates have not changed.

[0103] Step S506: If the difference is small, then it is determined that the coordinates have not changed.

[0104] Step S507: If the target disappears, i.e., miss_flag = True, then determine whether the target has disappeared at the edge.

[0105] Step S508: If the target edge disappears, i.e., edgeMiss_flag=True, then perform target detection one more time.

[0106] Step S509: Compare the difference between two consecutive coordinates.

[0107] Step S510: If the difference is large, determine that the coordinates have changed and jump_flag = True; otherwise, determine that the coordinates have not changed.

[0108] Step S511: If the target does not disappear at the edge, that is, the target disappears within the field of view edgeMiss_flag=False, then perform target detection twice more.

[0109] Step S512: Compare the differences between the coordinates in three consecutive measurements.

[0110] Step S513: If the difference is large, determine that the coordinates have changed and jump_flag = True; otherwise, determine that the coordinates have not changed.

[0111] Step S514: End transition judgment.

[0112] This embodiment uses a coordinate jump judgment strategy to determine whether the target coordinates obtained by the target detection algorithm and the single target tracking algorithm have jumped relative to the previous frame, thereby judging the reliability of the obtained coordinates.

[0113] Furthermore, N1, N2, N3, and N4 can all be adjusted according to actual usage requirements. A smaller N1 results in a higher frequency of target detection algorithm execution in pure tracking mode, leading to stronger tracking robustness but reduced real-time performance. Larger N2 and N3 result in more frames skipped after the target disappears, improving the algorithm's real-time performance but potentially losing more information. A larger N4 leads to more accurate judgment of target disappearance but may cause unnecessary computation, wasting computing resources and reducing real-time performance. As an example in this embodiment, N1 = 10, N2 = 3, N3 = 5, and N4 = 3.

[0114] Furthermore, the tracking strategy described in this invention is a complete tracking framework that can call different single-target tracking algorithms and target detection algorithms according to actual usage requirements, and has strong scalability; the single-target tracking algorithm based on correlation filtering used in this embodiment of the invention is KCF, and the target detection algorithm is YOLOv5.

[0115] Furthermore, those skilled in the art will understand that aspects of this application can be described and illustrated through several patentable types or situations, including any new and useful combination of processes, machines, products, or substances, or any new and useful improvements thereof. Accordingly, aspects of this application can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. All of the above hardware or software may be referred to as a “data block,” “module,” “engine,” “unit,” “component,” or “system.” Furthermore, aspects of this application may manifest as a computer product located on one or more computer-readable media, the product including computer-readable program code.

[0116] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A long-time single-target tracking method, characterized by, Includes the following steps: Detect the bounding box in the first frame of the video and display the detected bounding box in the image interface; In response to a user's selection of a target box in the image interface, the target within the selected target box is taken as the target to be tracked. Check the target information of the target to be tracked, wherein the target information includes at least: target disappearance information, target scale information, and target coordinate jump information; Obtain the corresponding variables in the target information, wherein the corresponding variable for the target disappearance information is whether the target has disappeared, the corresponding variable for the target scale information is whether the target scale is too small, and the corresponding variable for the target coordinate jump information is whether the target has jumped. Based on the variables in the target information, select to execute either pure tracking mode or detection tracking mode; The steps to perform pure trace mode include: A single-target tracking algorithm based on correlation filtering is executed. The target in the previous frame is used as a template. Template matching is performed in the current frame to search for the target to be tracked and obtain the target coordinates. The target detection algorithm is executed once every N1 frames. The detection box that is closest to the target in the previous frame is selected as the target box and replaced with the target box generated by the single target tracking algorithm in this frame to obtain the target coordinates. The target coordinates obtained in the above steps are subjected to jump judgment; if the coordinates do not jump, the coordinates obtained in the frame are put into the array, and the template is updated based on the tracking result of the frame, and then the pure tracking mode execution condition judgment is performed; if the coordinates jump, the coordinates of the frame are not recorded, and the pure tracking mode execution condition judgment is performed directly. If the pure tracking mode execution condition is met, the pure tracking mode is continued to be executed; otherwise, the pure tracking mode is exited and the detection tracking mode is executed. The single-target tracking algorithm based on correlation filtering is the KCF algorithm.

2. The long-time single object tracking method of claim 1, wherein, If the following conditions are met: the target has not disappeared, the target is not a small target, and the target has not changed direction, use the pure detection mode; otherwise, use the detection and tracking mode.

3. The long-time single object tracking method of claim 1, wherein, The steps for executing the detection and tracking mode include: Check for information indicating that the target has disappeared; If the target has not disappeared, execute the target detection algorithm and distance determination algorithm to obtain the target coordinates; The target coordinates are checked for jumps. If the target coordinates do not jump, the calculated target coordinates are stored in an array, and then the execution condition of the detection and tracking mode is checked. If the coordinates jump, the execution condition of the detection and tracking mode is checked directly. If the target disappears, check if the target disappeared at the edge. If the target disappeared at the edge, the program will not perform any operation in the subsequent N2 frame. If the target did not disappear at the edge, the program will not perform any operation in the subsequent N3 frame. Perform condition judgment for the execution of the detection and tracking mode; If the conditions for executing the detection and tracing mode are met, the detection and tracing mode will continue to be executed; otherwise, the detection and tracing mode will be exited and the pure tracing mode will be executed.

4. The long-time single object tracking method of claim 1, wherein, The target disappearance process includes edge disappearance and disappearance within the field of view.

5. The long-time single object tracking method of claim 4, wherein, The steps for determining when a target disappears include: The target did not disappear by default in the initial frame; Determine whether the target is not detected or a coordinate jump occurs within N4 consecutive frames; If the target is not detected for N4 consecutive frames or a coordinate jump occurs, the target is determined to have disappeared. Determine if the target coordinates in the previous frame are at the image boundary; If the target is at the image boundary, it is determined that the target edge has disappeared; otherwise, it is determined that the target has disappeared within the field of view. If at least one of the N4 consecutive frames shows that the target can be detected and no coordinate jump occurs, then the target is determined not to have disappeared. End of target disappearance judgment.

6. The long-time single object tracking method of claim 1, wherein, The steps for determining coordinate jumps include: Check the target disappearance information and the number of coordinate jumps; If the target has not disappeared, compare the difference between the coordinates in this frame and the coordinates in the previous frame; If the difference is large, then perform two more target detections consecutively; compare the differences in coordinates in three consecutive tests; if the difference is large, determine that the coordinates have changed; otherwise, determine that the coordinates have not changed. If the difference is small, then the coordinates are determined not to have changed. If the target disappears, determine whether the target edge has disappeared. If the target edge has disappeared, perform another target detection and compare the difference between the two consecutive coordinates. If the difference is large, determine that the coordinates have changed; otherwise, determine that the coordinates have not changed. If the target edge does not disappear, then perform two more target detections and compare the differences in the coordinates of the three consecutive detections. If the difference is large, then the coordinates are determined to have changed; otherwise, the coordinates are determined not to have changed. End the jump judgment.

7. The long-time single object tracking method of claim 3, wherein, The target detection algorithm used is YOLOv5.

8. A memory device, comprising: The system stores program instructions that can be executed by a processor, the program instructions being used to implement the long-term single-target tracking method according to any one of claims 1 to 7.

Images