Athlete training assistance system based on multi-target tracking
By constructing a multi-target tracking athlete training assistance system, the problem of automating motion acquisition and feedback in competitive sports training has been solved, enabling real-time motion assessment and personalized guidance, thereby improving training effectiveness and data support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- MALANSHAN AUDIO & VIDEO LABORATORY
- Filing Date
- 2026-04-28
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies lack automated motion capture, quantitative comparison, and real-time feedback mechanisms in competitive sports training, resulting in strong coach subjectivity, difficulty in addressing multiple objectives, inconsistent training effects among individuals, and a lack of real-time feedback and visual recording.
A multi-target tracking-based athlete training assistance system was constructed, including a standard movement library module, a real-time acquisition module, a model building module, a movement alignment and comparison module, a real-time feedback module, and a report generation module. Through deep learning, a multi-target tracking model and a movement alignment and comparison model were constructed to achieve real-time acquisition of the athlete's 3D skeleton sequence and real-time feedback of deviation data.
It enables real-time, high-precision quantitative assessment and personalized guidance of athletes' movements, improving training efficiency and movement standardization, and providing data support for scientific training.
Smart Images
Figure CN122347833A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of motion analysis technology, specifically to an athlete training assistance system based on multi-target tracking. Background Technology
[0002] In the field of competitive sports training, the standardization and normalization of athletes' technical movements directly determine their competitive performance and have a key impact on sports injury prevention. Scientific and precise movement analysis has become an indispensable part of the modern sports training system. However, the current mainstream training guidance model still relies heavily on coaches' visual observation and experience judgment. This traditional approach has significant limitations: First, it is highly subjective. Different coaches have different evaluation standards for the same movement, and it is difficult to quantify the "degree of deviation," resulting in varying training effects from person to person. Second, it lacks real-time feedback. Athletes are usually evaluated only after completing the movement, making it impossible for them to self-correct during the movement, which can easily lead to incorrect muscle memory. Third, it is difficult to balance multiple goals. When coaches guide multiple athletes at the same time, they cannot capture the details and flaws of each individual. The training quality of some athletes cannot be guaranteed, and there is a lack of visual records of movement deviations during training. Post-training reviews lack data support, and the progress trajectory is difficult to quantify.
[0003] Therefore, there is an urgent need for a training assistance system that can automatically collect athletes' movements, quantitatively compare them with standard movements, and provide real-time feedback and correction prompts. To this end, we now provide an athlete training assistance system based on multi-target tracking. Summary of the Invention
[0004] The purpose of this invention is to provide an athlete training assistance system based on multi-target tracking.
[0005] The objective of this invention can be achieved through the following technical solution: an athlete training assistance system based on multi-target tracking, comprising the following:
[0006] The standard action library module is used to build standard action libraries;
[0007] The real-time acquisition module is used to collect raw multi-source data;
[0008] The model building module is used to build a multi-target tracking model based on raw multi-source data and a motion alignment and comparison model based on a standard motion library.
[0009] The motion alignment and comparison module is used to obtain the real-time 3D skeleton sequence of the corresponding athlete based on the multi-target tracking model, and to call the motion alignment and comparison model to obtain the corresponding deviation data.
[0010] The real-time feedback module is used to generate and output correction prompts based on the real-time 3D skeleton sequence and corresponding deviation data.
[0011] The report generation module is used to generate personalized training reports.
[0012] Furthermore, the process of building the standard action library module includes:
[0013] Acquire video data of standard technical movements demonstrated by professional athletes, synchronously collected standard three-dimensional skeleton sequences, kinematic parameters of human joints, multi-view two-dimensional images, and biomechanical constraints.
[0014] Extract motion structure parameters from standard technical motion video data and standard 3D skeleton sequences;
[0015] The motion structure parameters, human joint kinematic parameters, multi-view two-dimensional images, and biomechanical constraints are encapsulated into a standard motion data package, and a unique motion index is assigned to each standard motion data package.
[0016] A separate data storage area is created for each standard action data packet in the database, thereby forming a standard action library.
[0017] Furthermore, the process by which the real-time acquisition module acquires raw multi-source data includes:
[0018] Multi-view acquisition devices are deployed at preset locations in the training field to collect image data and unique identification of athletes. Each frame of image data is accompanied by a timestamp.
[0019] The collected image data, unique identifiers, and corresponding timestamps are integrated in the format of "timestamp-data type-data content-unique identifier" to form structured raw multi-source data.
[0020] Furthermore, the process of constructing a multi-target tracking model by the model building module includes:
[0021] A multi-target tracking model is constructed based on deep learning. Several raw multi-source data are obtained in advance from the real-time acquisition module. Based on the raw multi-source data, the three-dimensional skeleton coordinates, bounding box positions and motion trajectories of each athlete in different time frames are extracted. Athlete identification is performed on the raw multi-source data. The labeled raw multi-source data is used as training data. The multi-target tracking model is trained using the training data to obtain the trained multi-target tracking model.
[0022] Furthermore, the process of constructing the action alignment and comparison model by the model building module includes:
[0023] An action alignment and comparison model is constructed based on deep learning, and the parameters of the constructed action alignment and comparison model are initialized.
[0024] Acquire several sample data, including several historical standard movement data packages extracted from the standard movement library and real-time three-dimensional skeleton sequences of athletes corresponding to the same technical movements obtained from the historical output results of the multi-target tracking model. Each historical standard movement data package contains the movement structure parameters, human joint kinematic parameters and biomechanical constraints of a type of technical movement.
[0025] The sample data is randomly divided into a training set and a test set according to a preset ratio to complete the training of the action alignment and comparison model and obtain the action alignment and comparison model.
[0026] Furthermore, the process by which the motion alignment and comparison module obtains the corresponding real-time 3D skeleton sequence of the athlete based on the multi-target tracking model includes:
[0027] After the real-time acquisition module outputs the raw multi-source data, it calls the multi-target tracking model that has been trained.
[0028] The raw multi-source data of the current frame is input into the multi-target tracking model, which directly outputs the real-time 3D skeleton sequence of each athlete in the current frame and the corresponding technical movement type.
[0029] Furthermore, the process by which the action comparison and alignment module calls the action alignment and comparison model to obtain the corresponding deviation data includes:
[0030] The obtained real-time 3D skeleton sequence is used as the input sequence;
[0031] Based on the type of technical movement the athlete is currently performing, retrieve the standard movement data package corresponding to that technical movement type from the standard movement library, and extract the standard three-dimensional skeleton sequence from the standard movement data package;
[0032] The input sequence and the corresponding standard 3D skeleton sequence are input into the trained motion alignment and comparison model, which outputs the deviation data of each athlete in the current frame.
[0033] For each athlete, the performance evaluation scores of all movements performed in chronological order are quantified and grouped into a sequence to obtain a quantitative performance evaluation score sequence.
[0034] Furthermore, the process by which the real-time feedback module generates and outputs correction prompts in real time includes:
[0035] Based on the joint angle deviation value, position deviation distance, and deviation location marker of each athlete's joints in the current frame, the joint angle deviation value of each joint is compared with the preset angle deviation threshold, and the position deviation distance of each joint is compared with the preset position deviation threshold. Based on the comparison results, corresponding correction prompts are generated.
[0036] Furthermore, the process by which the report generation module generates a personalized training report includes:
[0037] After a training session ends, the recorded data for each athlete during that training session is retrieved.
[0038] For each technical movement of the athlete, based on the start and end timestamps of the movement, as well as the joint angle deviation value and position deviation distance, the data corresponding to all frames within that time period are extracted as the deviation sequence of the technical movement, and then the total deviation value of the technical movement is obtained.
[0039] Based on the preset full score and total deviation value of the technical action, the action consistency score of the technical action is obtained. The scores of all technical actions are arranged in order of completion time to obtain the action consistency score sequence.
[0040] Based on the correction prompts recorded during training, the total number of times each athlete triggered correction prompts throughout the entire training process was counted, and the number of triggers for each part was counted according to the deviation part, and the number of triggers for the first level and the second level of correction prompts was counted according to the correction prompt level.
[0041] For each joint, the total deviation frequency of that joint is obtained based on the number of frames in all frames of all technical actions where the absolute value of the joint angle deviation exceeds a first angle threshold and the number of frames where the position deviation distance exceeds a first position threshold.
[0042] Sort the total deviation frequencies from high to low, take the top three as the main deviation locations, and label the deviation types. Draw a trend curve of the score change based on the action consistency score sequence.
[0043] The results are quantified into a scoring sequence, a trend curve of scoring changes, the total deviation frequency of each joint, the main deviation locations and deviation types, and are integrated into a personalized training report for each athlete and each corresponding technical movement.
[0044] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a standard movement library through a standard movement library module and collects raw multi-source data from athletes' training processes through a real-time acquisition module, establishing standardized movement benchmarks and conducting multi-dimensional data collection, providing a reliable data foundation for subsequent precise comparisons; the model building module constructs a multi-target tracking model and a movement alignment and comparison model; the movement alignment and comparison module obtains the athlete's real-time three-dimensional skeleton sequence based on the multi-target tracking model and calls the movement alignment and comparison model to obtain corresponding deviation data, achieving real-time, high-precision quantitative evaluation of technical movements and eliminating the differences in coaches' subjective judgments; the real-time feedback module generates graded correction prompts based on deviation data and provides real-time feedback to athletes and coaches through visualization, providing immediate and intuitive movement correction guidance, helping athletes quickly form correct muscle memory and improve training efficiency; the report generation module comprehensively analyzes training data and generates personalized training reports, achieving traceability and quantitative analysis of training data, providing data support for long-term progress tracking and personalized training plan development; this system realizes real-time monitoring, precise evaluation, and personalized guidance of multiple athletes' technical movements, effectively improving training efficiency and movement standardization, and providing data support for scientific training. Attached Figure Description
[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0046] Figure 1 This is a schematic diagram of the present invention. Detailed Implementation
[0047] like Figure 1 As shown, the athlete training assistance system based on multi-target tracking includes a standard movement library module, a real-time acquisition module, a model building module, a movement alignment and comparison module, a real-time feedback module, and a report generation module.
[0048] The standard action library module is used to build the standard action library;
[0049] The real-time acquisition module is used to acquire raw multi-source data;
[0050] The model building module is used to build a multi-target tracking model based on the original multi-source data and to build an action alignment and comparison model based on the standard action library;
[0051] The motion alignment and comparison module is used to obtain the corresponding real-time three-dimensional skeleton sequence of the athlete based on the multi-target tracking model, and to call the motion alignment and comparison model to obtain the corresponding deviation data;
[0052] The real-time feedback module is used to generate and output correction prompts based on the real-time 3D skeleton sequence and corresponding deviation data;
[0053] The report generation module is used to generate personalized training reports.
[0054] It should be further explained that, in the specific implementation process, the standard action library module's process of building the standard action library includes:
[0055] Acquire video data of standard technical movements demonstrated by professional athletes, synchronously collected standard three-dimensional skeleton sequences, kinematic parameters of human joints, multi-view two-dimensional images, and biomechanical constraints.
[0056] Action structure parameters are extracted from the standard technical motion video data and the standard three-dimensional skeleton sequence. The standard three-dimensional skeleton sequence consists of the coordinates of human joints in several consecutive time frames, with each time frame corresponding to a posture.
[0057] The motion structure parameters include the three-dimensional coordinates of each joint in each frame, the joint angle change curve, the timestamp of key posture frames, and the motion speed baseline.
[0058] The kinematic parameters of human joints include the range of joint motion, limb length ratio, center of gravity trajectory data, and standard posture classification specifications based on human anatomy.
[0059] The biomechanical constraints include safety limits for joint angles and inviolable linkages in the kinetic chain;
[0060] The motion structure parameters, human joint kinematic parameters, multi-view two-dimensional images, and biomechanical constraints are encapsulated into a standard motion data package, and a unique motion index is assigned to each standard motion data package.
[0061] A separate data storage area is created for each standard action data packet in the database, thereby forming a standard action library.
[0062] It should be further explained that, in the specific implementation process, the process of the real-time acquisition module acquiring raw multi-source data includes:
[0063] Multi-view acquisition devices are deployed at predetermined locations on the training ground to collect image data and unique identification identifiers of athletes. Each frame of image data is accompanied by a timestamp.
[0064] The collected image data, unique identifiers, and corresponding timestamps are integrated in the format of "timestamp-data type-data content-unique identifier" to form structured raw multi-source data.
[0065] It should be further explained that, in the specific implementation process, the process of the model building module constructing the multi-target tracking model includes:
[0066] A multi-target tracking model is constructed based on deep learning. Several raw multi-source data are obtained in advance from the real-time acquisition module. Based on the raw multi-source data, the three-dimensional skeleton coordinates, bounding box positions and motion trajectories of each athlete in different time frames are extracted. Athlete identification is performed on the raw multi-source data. The labeled raw multi-source data is used as training data. The multi-target tracking model is trained using the training data to obtain the trained multi-target tracking model.
[0067] It should be further explained that, in the specific implementation process, the model building module's process of aligning and comparing models includes:
[0068] An action alignment and comparison model is constructed based on deep learning, and the parameters of the constructed action alignment and comparison model are initialized.
[0069] Acquire several sample data, including several historical standard movement data packages extracted from the standard movement library and real-time three-dimensional skeleton sequences of athletes corresponding to the same technical movements obtained from the historical output results of the multi-target tracking model. Each historical standard movement data package contains the movement structure parameters, human joint kinematic parameters and biomechanical constraints of a technical movement.
[0070] The sample data is randomly divided into a training set and a test set according to a preset ratio;
[0071] The training set is used to supervise the training of the constructed action alignment and comparison model. The parameters of the action alignment and comparison model are adjusted by optimizing the loss function, so that the model learns to perform temporal and spatial alignment based on the input real-time 3D skeleton sequence of the athlete and the corresponding original multi-source data, and outputs the angle deviation value, position deviation distance and quantitative score of the completion effect of each complete technical movement for each frame.
[0072] The prediction accuracy of the trained action alignment and contrast model is verified using a test set. If the prediction accuracy of the training results meets the preset threshold, the training of the action alignment and contrast model is completed. If the training results do not meet the preset threshold, the model hyperparameters or training strategy are adjusted and the training is repeated until the training results meet the preset threshold or the number of iterations reaches the preset upper limit, thereby completing the training of the action alignment and contrast model and obtaining the action alignment and contrast model.
[0073] It should be further explained that, in the specific implementation process, the process by which the motion alignment and comparison module obtains the corresponding real-time 3D skeleton sequence of the athlete based on the multi-target tracking model includes:
[0074] After the real-time acquisition module outputs the raw multi-source data, it calls the multi-target tracking model that has been trained.
[0075] The raw multi-source data is input into the multi-target tracking model, which directly outputs the real-time 3D skeleton sequence of each athlete in the current frame and the corresponding technical movement type.
[0076] It should be further explained that, in the specific implementation process, the process by which the action comparison and alignment module calls the action alignment and comparison model to obtain the corresponding deviation data includes:
[0077] The obtained real-time 3D skeleton sequence is used as the input sequence;
[0078] Based on the type of technical movement the athlete is currently performing, retrieve the standard movement data package corresponding to that technical movement type from the standard movement library, and extract the standard three-dimensional skeleton sequence from the standard movement data package;
[0079] The input sequence and the corresponding standard 3D skeleton sequence are input into the trained motion alignment and comparison model. The motion alignment and comparison model outputs the deviation data of each athlete in the current frame. The deviation data includes joint angle deviation value, position deviation distance, deviation part marking and performance quantitative score.
[0080] For each athlete, the performance evaluation scores of all movements performed in chronological order are quantified and grouped into a sequence to obtain a quantitative performance evaluation score sequence.
[0081] It should be further explained that, in the specific implementation process, the process by which the real-time feedback module generates and outputs correction prompts in real time based on real-time 3D skeleton data and corresponding deviation data includes:
[0082] Based on the joint angle deviation value, position deviation distance, and deviation location marker of each athlete's joints in the current frame, the joint angle deviation value of each joint is compared with the preset angle deviation threshold, and the position deviation distance of each joint is compared with the preset position deviation threshold.
[0083] The angle deviation threshold includes a first angle threshold and a second angle threshold, wherein the first angle threshold is less than the second angle threshold.
[0084] The position deviation threshold includes a first position threshold and a second position threshold, wherein the first position threshold is less than the second position threshold;
[0085] If the absolute value of the joint angle deviation is greater than or equal to the first angle threshold and less than the second angle threshold, or the position deviation distance is greater than or equal to the first position threshold and less than the second position threshold, then a first-level correction prompt is generated.
[0086] If the absolute value of the joint angle deviation is greater than or equal to the second angle threshold, or the position deviation distance is greater than or equal to the second position threshold, then a second-level correction prompt is generated.
[0087] If the absolute value of the joint angle deviation is less than the first angle threshold and the position deviation distance is less than the first position threshold, no correction prompt will be generated.
[0088] Specifically:
[0089] The real-time 3D skeleton model of each athlete is displayed on the sideline display screen, and the display color of the corresponding joint point is set to green, yellow or red according to the angle threshold range where the joint angle deviation value is located and the position threshold range where the position deviation distance is located.
[0090] Green indicates that the angle deviation value is less than the first angle threshold and the position deviation distance is less than the first position threshold; yellow indicates that the angle deviation value is greater than or equal to the first angle threshold and less than the second angle threshold, or the position deviation distance is greater than or equal to the first position threshold and less than the second position threshold; red indicates that the angle deviation value is greater than or equal to the second angle threshold, or the position deviation distance is greater than or equal to the second position threshold.
[0091] It should be further explained that, in the specific implementation process, the process of the report generation module generating a personalized training report includes:
[0092] After a training session ends, the recorded data for each athlete during the training session is read. The recorded data includes the timestamp of each technical movement, the joint angle deviation value of each joint point in each frame, the position deviation distance, and the corresponding correction prompts.
[0093] For each technical movement of the athlete, based on the start and end timestamps of the movement, as well as the joint angle deviation value and position deviation distance, the data corresponding to all frames within that time period are extracted as the deviation sequence of the technical movement.
[0094] Assume that this technical action has a total of Frames, each frame has There are 1 key points, among which ;
[0095] Calculate the total deviation value of this technical movement, denoted as . ,Right now:
[0096] ;
[0097] in, This represents the absolute value of the joint angle deviation. Indicates the positional deviation distance. These are preset angle weighting coefficients. This represents the preset position weight coefficient;
[0098] Based on the preset full score for the technical action, obtain the action consistency score for this technical action, and record it as follows. ,Right now:
[0099] ;
[0100] in, The preset full score for the technical action;
[0101] Arrange the scores of all N technical movements in order of completion time to obtain the movement consistency score sequence;
[0102] Based on the correction prompts recorded during training, the total number of times each athlete triggered correction prompts throughout the entire training process was counted, and the number of triggers for each part was counted according to the deviation part, and the number of triggers for the first level and the second level of correction prompts was counted according to the correction prompt level.
[0103] For each joint, the total deviation frequency of that joint is obtained by considering the number of frames in all frames of all technical actions where the absolute value of the joint angle deviation exceeds a first angle threshold and the number of frames where the position deviation distance exceeds a first position threshold. This deviation frequency is denoted as... ,Right now:
[0104] ;
[0105] in, This indicates the number of frames in which the absolute value of the joint angle deviation exceeds the first angle threshold. This indicates the number of frames where the position deviation exceeds the first position threshold. Indicates for the first The total number of frames where the absolute value of the joint angle deviation at a given joint point exceeds the first angle threshold. Indicates for the first The sum of the number of frames in which the positional deviation of a key point exceeds the first position threshold. and These are the normalized weights for angular deviation and positional deviation, respectively. and Relationship satisfaction, ;
[0106] Sort the total deviation frequencies from high to low, take the top three as the main deviation locations, and label the deviation types;
[0107] like If the deviation type is angular, then the deviation type is mainly angular; otherwise, the deviation type is mainly positional.
[0108] A trend curve of score change is plotted based on the action consistency score sequence, wherein the horizontal axis of the trend curve represents the technical action and the vertical axis represents the action consistency score.
[0109] The results are quantified into a scoring sequence, a trend curve of scoring changes, the total deviation frequency of each joint, the main deviation locations and deviation types, and are integrated into a personalized training report for each athlete and each corresponding technical movement.
[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications or equivalent substitutions made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. An athlete training assistance system based on multi-target tracking, characterized in that, include: The standard action library module is used to build standard action libraries; The real-time acquisition module is used to collect raw multi-source data; The model building module is used to build a multi-target tracking model based on raw multi-source data and a motion alignment and comparison model based on a standard motion library. The motion alignment and comparison module is used to obtain the real-time 3D skeleton sequence of the corresponding athlete based on the multi-target tracking model, and to call the motion alignment and comparison model to obtain the corresponding deviation data. The real-time feedback module is used to generate and output correction prompts based on the real-time 3D skeleton sequence and corresponding deviation data. The report generation module is used to generate personalized training reports.
2. The athlete training assistance system based on multi-target tracking according to claim 1, characterized in that, The process of building a standard action library module includes: Acquire video data of standard technical movements demonstrated by professional athletes, synchronously collected standard three-dimensional skeleton sequences, kinematic parameters of human joints, multi-view two-dimensional images, and biomechanical constraints. Extract motion structure parameters from standard technical motion video data and standard 3D skeleton sequences; The motion structure parameters, human joint kinematic parameters, multi-view two-dimensional images, and biomechanical constraints are encapsulated into a standard motion data package, and a unique motion index is assigned to each standard motion data package. A separate data storage area is created for each standard action data packet in the database, thereby forming a standard action library.
3. The athlete training assistance system based on multi-target tracking according to claim 2, characterized in that, The process by which the real-time acquisition module acquires raw multi-source data includes: Multi-view acquisition devices are deployed at preset locations in the training field to collect image data and unique identification of athletes. Each frame of image data is accompanied by a timestamp. The collected image data, unique identifiers, and corresponding timestamps are integrated in the format of "timestamp-data type-data content-unique identifier" to form structured raw multi-source data.
4. The athlete training assistance system based on multi-target tracking according to claim 3, characterized in that, The process of building a multi-target tracking model by the model building module includes: A multi-target tracking model is constructed based on deep learning. Several raw multi-source data are obtained in advance from the real-time acquisition module. Based on the raw multi-source data, the three-dimensional skeleton coordinates, bounding box positions and motion trajectories of each athlete in different time frames are extracted. Athlete identification is performed on the raw multi-source data. The labeled raw multi-source data is used as training data. The multi-target tracking model is trained using the training data to obtain the trained multi-target tracking model.
5. The athlete training assistance system based on multi-target tracking according to claim 4, characterized in that, The process of building the action alignment and comparison model by the model building module includes: An action alignment and comparison model is constructed based on deep learning, and the parameters of the constructed action alignment and comparison model are initialized. Acquire several sample data, including several historical standard movement data packages extracted from the standard movement library and real-time three-dimensional skeleton sequences of athletes corresponding to the same technical movements obtained from the historical output results of the multi-target tracking model. Each historical standard movement data package contains the movement structure parameters, human joint kinematic parameters and biomechanical constraints of a type of technical movement. The sample data is randomly divided into a training set and a test set according to a preset ratio to complete the training of the action alignment and comparison model and obtain the action alignment and comparison model.
6. The athlete training assistance system based on multi-target tracking according to claim 5, characterized in that, The process by which the motion alignment and comparison module obtains the corresponding real-time 3D skeleton sequence of the athlete based on the multi-target tracking model includes: After the real-time acquisition module outputs the raw multi-source data, it calls the multi-target tracking model that has been trained. The raw multi-source data of the current frame is input into the multi-target tracking model, which directly outputs the real-time 3D skeleton sequence of each athlete in the current frame and the corresponding technical movement type.
7. The athlete training assistance system based on multi-target tracking according to claim 6, characterized in that, The process by which the action comparison and alignment module calls the action alignment and comparison model to obtain the corresponding deviation data includes: The obtained real-time 3D skeleton sequence is used as the input sequence; Based on the type of technical movement the athlete is currently performing, retrieve the standard movement data package corresponding to that technical movement type from the standard movement library, and extract the standard three-dimensional skeleton sequence from the standard movement data package; The input sequence and the corresponding standard 3D skeleton sequence are input into the trained motion alignment and comparison model, which outputs the deviation data of each athlete in the current frame. For each athlete, the performance evaluation scores of all movements performed in chronological order are quantified and grouped into a sequence to obtain a quantitative performance evaluation score sequence.
8. The athlete training assistance system based on multi-target tracking according to claim 7, characterized in that, The process by which the real-time feedback module generates and outputs correction prompts in real time includes: Based on the joint angle deviation value, position deviation distance, and deviation location marker of each athlete's joints in the current frame, the joint angle deviation value of each joint is compared with the preset angle deviation threshold, and the position deviation distance of each joint is compared with the preset position deviation threshold. Based on the comparison results, corresponding correction prompts are generated.
9. The athlete training assistance system based on multi-target tracking according to claim 8, characterized in that, The process by which the report generation module generates a personalized training report includes: After a training session ends, the recorded data for each athlete during that training session is retrieved. For each technical movement of the athlete, based on the start and end timestamps of the movement, as well as the joint angle deviation value and position deviation distance, the data corresponding to all frames within that time period are extracted as the deviation sequence of the technical movement, and then the total deviation value of the technical movement is obtained. Based on the preset full score and total deviation value of the technical action, the action consistency score of the technical action is obtained. The scores of all technical actions are arranged in order of completion time to obtain the action consistency score sequence. Based on the correction prompts recorded during training, the total number of times each athlete triggered correction prompts throughout the entire training process was counted, and the number of triggers for each part was counted according to the deviation part, and the number of triggers for the first level and the second level of correction prompts was counted according to the correction prompt level. For each joint, the total deviation frequency of that joint is obtained based on the number of frames in all frames of all technical actions where the absolute value of the joint angle deviation exceeds a first angle threshold and the number of frames where the position deviation distance exceeds a first position threshold. Sort the total deviation frequencies from high to low, take the top three as the main deviation locations, and label the deviation types. Draw a trend curve of the score change based on the action consistency score sequence. The results are quantified into a scoring sequence, a trend curve of scoring changes, the total deviation frequency of each joint, the main deviation locations and deviation types, and are integrated into a personalized training report for each athlete and each corresponding technical movement.