Multi-dimensional intelligent analysis method for motion process, computer equipment and storage medium

By deeply analyzing imaging parameters and auxiliary information, detecting and compensating invalid data in video data, extracting the pose point sequence and comparing standard action templates, the problems of incomplete data processing and abnormality in intelligent sports training systems are solved, and the motion action recognition is improved accuracy and reliability.

CN118887738BActive Publication Date: 2024-12-06ZHUHAI XINWEI TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411372460.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-29
Publication Date
2024-12-06
Estimated Expiration
2044-09-29

AI Technical Summary

Technical Problem

The existing intelligent sports training system lacks in-depth analysis of imaging parameters in video data processing, and the auxiliary information extraction is not comprehensive enough to effectively handle data missing or abnormalities, resulting in discontinuous motion trajectory and errors in posture recognition, which affects the accuracy of evaluation.

Method used

By collecting student movement video data, analyzing imaging parameters, extracting auxiliary information such as timestamps, site identification and training project types, detecting and marking invalid data, performing data compensation processing, integrating the complete data set, and extracting the pose point sequence and standard action template comparison. Yes, calculate the completion score of the movement technique and generate a comprehensive analysis report.

Benefits of technology

It improves the accuracy and pertinence of action recognition, improves the accuracy and reliability of action recognition, ensures robust processing capabilities in complex scenarios, and enhances the reliability and effectiveness of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118887738B_ABST
    Figure CN118887738B_ABST
Patent Text Reader

Abstract

The present application relates to a multi-dimensional intelligent analysis method for a motion process, a computer device, and a storage medium. The method includes: parsing imaging parameters from video data; extracting auxiliary information from video data; detecting invalid data in video data based on student position, motion trajectory, and second-class identifiers; performing data compensation processing on video clips to be processed to obtain a complete data set; extracting a sequence of posture points during student motion; calculating the completion score of sports technical movements; and generating a comprehensive analysis report based on sports technical movements and completion scores. The present application deeply analyzes imaging parameters during video data processing, thereby improving the accuracy of motion recognition, analyzes auxiliary information, improves the pertinence of motion recognition, processes invalid data, improves the accuracy of motion recognition, and compensates data, thereby improving the reliability and effectiveness of motion recognition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of intelligent data recognition technology, and in particular to a multi-dimensional intelligent analysis method for a motion process, a computer device, and a storage medium. Background Art

[0002] During sports training, students' sports performance is affected by many factors, such as physical fitness, movement techniques, tactical awareness, psychological state, etc. Traditional sports training evaluation often relies on the subjective experience of coaches, or only focuses on single-dimensional data indicators, making it difficult to comprehensively and objectively judge students' sports ability and potential for improvement. With the advancement of computer vision technology, using multiple cameras to collect video data of students' sports processes and performing multi-dimensional analysis through artificial intelligence algorithms can obtain a full range of insights into students' sports performance, providing data support for accurately evaluating students' sports ability and optimizing training plans.

[0003] However, existing intelligent sports training systems still have some limitations. First, in terms of video data processing, many systems lack in-depth analysis and utilization of imaging parameters, which limits the accuracy of subsequent analysis. Second, in terms of auxiliary information extraction, existing systems often only focus on time information and ignore important contextual information such as venue identification and training item type, which reduces the pertinence of analysis results. In addition, in terms of student position recognition and tracking, existing systems usually use simple target detection algorithms and lack robust processing capabilities in complex scenarios, especially when dealing with occlusion, fast motion and other situations.

[0004] Most critically, existing systems often perform poorly when dealing with missing or abnormal data. In actual sports training environments, data loss or quality degradation often occurs due to various factors such as equipment failure, environmental interference, or human occlusion. These data issues may lead to discontinuities in motion trajectories, errors in posture recognition, or deviations in technical action analysis, which seriously affect the accuracy of the overall evaluation. However, existing systems generally lack effective data compensation mechanisms and cannot effectively handle these data anomalies, which greatly limits the reliability and effectiveness of intelligent sports training systems in practical applications. Summary of the invention

[0005] Based on this, it is necessary to provide a multi-dimensional intelligent analysis method, computer equipment and storage medium for motion processes to improve the reliability and effectiveness of motion recognition in response to the above-mentioned technical problems.

[0006] A multi-dimensional intelligent analysis method for student exercise process in a smart sports training area includes the following steps:

[0007] Collect video data of the student's exercise process, parse imaging parameters from the video data, and assign corresponding first-class identifiers to video segments in the video data;

[0008] Extract auxiliary information from the video data based on the imaging parameters and the first type of identifier; the auxiliary information includes frame-by-frame extraction timestamps, site identifiers, and training project types;

[0009] Process the video data according to the auxiliary information, obtain the student position, motion trajectory and the second type of identifier, detect invalid data in the video data according to the student position, motion trajectory and the second type of identifier, and mark the video segment containing invalid data as a pending state;

[0010] Perform data compensation processing on the video clips in the to-be-processed state, and integrate the compensated video clips with the video clips that are not marked as to-be-processed to obtain a complete data set;

[0011] Based on the complete data set, the posture point sequence of the students during the movement is extracted; the posture point sequence includes the spatial position, joint angle and relative position relationship of the human body joints;

[0012] Compare the posture point sequence with the standard action template, identify the sports technical actions in the complete data set, and calculate the completion score of the sports technical actions;

[0013] Generate a comprehensive analysis report based on sports technical movements and completion scores.

[0014] In one embodiment, in the step of extracting auxiliary information from the video data based on the imaging parameters and the first type of identifier, the training item type is obtained based on the following steps:

[0015] Calculate the similarity between the venue identifier and the preset venue template, and use the preset venue template corresponding to the highest similarity as the current venue;

[0016] Determine the scope of training programs based on the current venue;

[0017] A lightweight human body detection algorithm is used to process video data, obtain the human body contour in the video data, and analyze the student's movement pattern based on the human body contour;

[0018] Analyze video data to identify human body distribution and types of sports equipment;

[0019] According to the context of the data, capture the action sequence features within a preset duration in the video data;

[0020] Based on the current venue, movement pattern, human body distribution, type of sports equipment and action sequence characteristics, statistical models are used to determine the type of training program within the training program range.

[0021] In one embodiment, the steps of processing the video data according to the auxiliary information, obtaining the student position, the motion trajectory and the second type of identifier, detecting invalid data in the video data according to the student position, the motion trajectory and the second type of identifier, and marking the video segment containing the invalid data as a pending state include the steps of:

[0022] A boundary model is established according to the site identification, and the student position is compared with the boundary model. If the student position exceeds the boundary model, the data segment corresponding to the student position exceeding the boundary model is recorded as invalid data;

[0023] Calculate the speed and acceleration at each time point according to the motion trajectory, compare the speed and acceleration with the preset human ability range, and if the speed and acceleration exceed the preset human ability range, mark the data segment corresponding to the speed and acceleration exceeding the preset human ability range as invalid data;

[0024] Retrieve whether the second-category identifiers corresponding to the students are consistent. When the number of changes of the second-category identifiers corresponding to the students exceeds a preset number within a preset time period, detect the inducing factors of the change of the second-category identifiers. If the inducing factors do not meet the preset inducing factors, mark the data segments corresponding to the second-category identifiers that do not meet the preset inducing factors as invalid data.

[0025] When the amount of invalid data accumulated in a video segment exceeds a preset threshold, the video segment is marked as being in a pending state.

[0026] In one embodiment, the steps are also included:

[0027] Calculate the student's trajectory curvature, speed change, direction change and posture change based on the motion trajectory;

[0028] Calculate the complexity index of the student's movement based on trajectory curvature, speed change, direction change, and posture change;

[0029] If the complexity index exceeds a preset normal range, it is determined that an error occurs in the acquisition process or the processing process of the video data.

[0030] In one embodiment, the step of performing data compensation processing on the video segments in the to-be-processed state and integrating the compensated video segments with the video segments not marked as to-be-processed state to obtain a complete data set comprises the steps of:

[0031] In terms of time sequence, a data frame is selected as the first reference point in a video segment that is before the video segment in the to-be-processed state and is the closest video segment that is not marked as the to-be-processed state;

[0032] In terms of timing, the first normal data frame after the video clip in the to-be-processed state is selected as the second reference point;

[0033] Calculate the data credibility coefficient according to the length of the video clip in the to-be-processed state, the time distance between the first reference point and the second reference point, the complexity index of the student's movement, and the compensation speed and the original speed of the student's movement;

[0034] Performing weighted calculation on the first reference point, the second reference point and the data credibility coefficient to obtain a compensation estimation value;

[0035] Obtain compensation data according to the compensation estimation value, the length of the video segment to be processed, the length of the video data, and the data credibility coefficient;

[0036] The compensation data is inserted into the video segment in the to-be-processed state to compensate the video segment in the to-be-processed state.

[0037] In one embodiment, the step of extracting a sequence of posture points during a student's movement based on a complete data set comprises the steps of:

[0038] Use deep learning models to process the complete dataset and identify the students’ body joints;

[0039] According to the calibration parameters of the camera and the geometric information of the site, the spatial position of the human body joints in the three-dimensional space is calculated;

[0040] Calculate joint angles and relative positions based on spatial positions;

[0041] The spatial position, joint angles and relative positions are organized into a sequence of posture points in chronological order.

[0042] In one embodiment, the steps of comparing the posture point sequence with the standard action template, identifying the sports technical action, and calculating the completion score of the sports technical action include the steps of:

[0043] The posture point sequence is aligned with the standard action template using the dynamic time warping algorithm, and the similarity score between the posture point sequence and the standard action template is calculated;

[0044] A dynamic weight system is obtained according to the similarity score, and the video clip that best matches the standard action template in the complete data set is identified based on the dynamic weight system;

[0045] Identify sports technical movements based on video clips, and calculate the completion score of sports technical movements based on similarity scoring, dynamic weighting system and data credibility coefficient.

[0046] In one embodiment, the steps of aligning the posture point sequence with the standard action template using a dynamic time warping algorithm and calculating the similarity score between the posture point sequence and the standard action template include the steps of:

[0047] Calculate a first score based on the original posture point sequence in the video data, the standard action template and the data credibility coefficient;

[0048] Calculate a second score based on the posture point sequence, the standard action template and the data credibility coefficient;

[0049] A similarity score is obtained based on the first score, the second score and the data credibility coefficient.

[0050] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0051] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0052] One of the above technical solutions has the following advantages and beneficial effects:

[0053] The present application includes the following steps: collecting video data of the student's movement process, parsing imaging parameters from the video data, and assigning corresponding first-class identifiers to the video clips in the video data; extracting auxiliary information from the video data based on the imaging parameters and the first-class identifiers; processing the video data according to the auxiliary information, obtaining the student's position, movement trajectory, and second-class identifiers, detecting invalid data in the video data according to the student's position, movement trajectory, and second-class identifiers, and marking the video clips containing invalid data as pending; performing data compensation processing on the video clips in the pending state, and integrating the compensated video clips with the video clips not marked as pending to obtain a complete data set; extracting a posture point sequence during the student's movement based on the complete data set; the posture point sequence includes the spatial position, joint angle, and relative position relationship of the human body's joints; comparing the posture point sequence with the standard action template, identifying the sports technical actions in the complete data set, and calculating the completion score of the sports technical actions; generating a comprehensive analysis report based on the sports technical actions and the completion score. This application conducts in-depth analysis of imaging parameters during video data processing to improve the accuracy of motion recognition, analyzes auxiliary information to improve the pertinence of motion recognition, processes invalid data to improve the accuracy of motion recognition, and compensates for data to improve the reliability and effectiveness of motion recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of the multi-dimensional intelligent analysis method of the student movement process in the smart sports training area in the embodiment of the present application.

[0055] Figure 2 Schematic diagram of the internal structure of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] In order to achieve the above object, in one embodiment, Figure 1 As shown, a multi-dimensional intelligent analysis method for student exercise process in a smart sports training area is provided, including the following steps:

[0058] Step S110 , collecting video data of the student's exercise process, parsing imaging parameters from the video data, and assigning corresponding first-category identifiers to video segments in the video data.

[0059] Among them, imaging parameters include frame rate, resolution and encoding format.

[0060] First, multiple high-definition cameras are deployed in the training area to ensure full coverage of the students' exercise process. The placement of cameras takes into account the spatial layout of the training area, including the activity areas of different sports, the placement of equipment, and possible visual obstructions.

[0061] Once the camera is installed, the system begins to collect video data in real time. During the collection process, the system automatically analyzes key imaging parameters. The first is the frame rate, which determines the smoothness of the video and the accuracy of the motion capture. For most sports analysis, the system uses a frame rate of no less than 30 frames per second, and for high-speed sports, it may use a frame rate of 60 frames per second or higher. The system can identify the frame rate of different cameras and take this factor into account in subsequent analysis.

[0062] The second is resolution, which affects the clarity and detail of the image. The system uses high-definition resolution, such as 1080p or 4K, to provide sufficient visual information. The system is able to process video streams of different resolutions and convert or scale the resolution when necessary to meet the needs of subsequent analysis.

[0063] The third is the encoding format, which is related to the compression method and storage efficiency of video data. The system supports a variety of commonly used encoding formats, such as H.264, H.265, etc. During data processing, the system can identify and process these encoding formats, and select the appropriate format for storage according to storage requirements. When selecting an encoding format, the system will weigh the requirements of image quality and storage space. For important training videos that require fine motion analysis, a higher quality encoding method will be selected; for general surveillance videos, a format with a higher compression rate may be selected to save storage space.

[0064] In order to effectively manage and retrieve large amounts of video data, the system assigns a unique first-class identifier to each video clip. This first-class identifier is a composite structure that contains multiple information elements, such as camera number, shooting date and time, training project type, etc. For example, a first-class identifier may be in the format of "CAM01_20230720_1530_BASKETBALL_001", indicating that this is the first clip of the basketball training video shot by camera No. 1 at 15:30 on July 20, 2023.

[0065] The first type of identifier is designed with a hierarchical structure that is scalable and allows for the addition of new information elements in the future. The system has established an efficient indexing mechanism that supports fast retrieval based on the different elements in the identifier. This design enables the system to quickly locate video clips under specific conditions, such as all basketball training videos on a specific date, or all videos shot by a specific camera in the past week.

[0066] Step S120, extracting auxiliary information from the video data based on the imaging parameters and the first type of identifier; the auxiliary information includes a frame-by-frame extraction timestamp, a venue identifier, and a training item type.

[0067] The timestamp extraction process first uses the frame rate parsed above. The system reads the metadata of the video data and obtains the start time of the video data. Then, the time interval between each frame is calculated based on the frame rate. For example, for a video with a frame rate of 30 frames per second, the time interval between adjacent frames is 1 / 30 second. The system starts from the first frame of the video data and increments the timestamp frame by frame. The specific operation is: convert the start time into a millisecond timestamp, and then for each frame, multiply its serial number by the frame interval (in milliseconds), add the start timestamp, and get the accurate timestamp of the frame.

[0068] During timestamp extraction, the system takes into account possible frame rate fluctuations. It detects whether there is a significant difference by comparing the actual duration of the video data with the theoretical duration calculated based on the number of frames and frame rate. If there is a difference, the system performs a linear interpolation correction on the timestamp to ensure that the timestamp is synchronized with the actual time. This correction mechanism ensures the accuracy of the timestamp in the case of long recordings or unstable frame rates.

[0069] After the timestamp is extracted, the system begins to extract the site identifier. This process combines the first-class identifier assigned above with the video data content analysis. The system first extracts the camera number from the first-class identifier, and then queries the pre-established site mapping database to obtain the training area information covered by the camera. This database contains the location, coverage, and possible training items of each camera.

[0070] In one example, the type of training project is obtained based on the following steps: calculating the similarity between the venue identification and the preset venue template, and taking the preset venue template corresponding to the highest similarity as the current venue; determining the scope of the training project based on the current venue; using a lightweight human body detection algorithm to process video data, obtain the human body contour in the video data, and analyze the student's movement pattern based on the human body contour; analyzing the video data to identify the human body distribution and the type of sports equipment; based on the context of the data, capturing the action sequence features within a preset time length in the video data; and using a statistical model to determine the type of training project within the scope of the training project based on the current venue, movement pattern, human body distribution, type of sports equipment and action sequence features.

[0071] The system performs image analysis on the video data to verify and refine the site identification information. This process uses computer vision algorithms and mainly includes the following steps: First, the system preprocesses the video frames, including denoising, contrast enhancement, and geometric correction. Then, edge detection algorithms (such as the Canny edge detector) are used to identify lines and markings in the site. For the detected lines, the system uses the Hough transform algorithm to extract straight and circular features, which usually correspond to the boundary lines, center lines, or circular markings of the site.

[0072] The system matches the extracted venue logo with the pre-stored preset venue templates. These templates contain the standard layout of different training areas, such as the position of the bars in the single and double bar areas, the take-off line and landing area positions in the long jump area, etc. The matching process uses a feature point matching algorithm (such as SIFT or SURF) to calculate the similarity between the extracted venue logo and the preset venue template. The system selects the template with the highest similarity as the recognition result of the current venue.

[0073] During the recognition process, the system takes into account the change in the camera's viewing angle. It uses a perspective transformation algorithm to convert video data shot at different angles into a standard overhead view, which facilitates accurate comparison with the template. For situations where there may be occlusion or light changes, the system uses a local feature matching strategy, which allows recognition based on the visible part even if part of the site marking is not visible.

[0074] After the venue is identified, the system enters the stage of extracting the training project type. First, the system preliminarily determines the range of possible training projects based on the identified venue type. For example, for a running track, the system will consider various running projects including short-distance running, long-distance running or obstacle running.

[0075] The system then uses a lightweight human detection algorithm, such as a SVM classifier based on HOG features, to quickly detect human contours in the video. Based on the detected human contours, the system performs a simple motion pattern analysis, including detecting the approximate direction and amplitude of motion. For example, continuous linear motion detected on the track may indicate running training, while vertical reciprocating motion detected in the horizontal bar area may suggest pull-up training.

[0076] The system also analyzes the distribution and movement patterns of people in the venue. For example, multiple people evenly distributed on the track may indicate long-distance running training, while gathering at the starting line may indicate short-distance running training. At the same time, the system identifies the presence of specific sports equipment in the venue, such as high jump bars or shot puts, the presence of which can help further narrow down the possible training items.

[0077] Throughout the analysis process, the system considers the continuity and contextual information of the action. It not only focuses on the static features of a single moment, but also captures the overall pattern of action sequence features through simple time series analysis. For example, when identifying long jump training, the system tries to detect the complete sequence of run-up, take-off, aerial posture and landing. This perception of context greatly improves the accuracy of recognition and reduces misjudgment caused by a single feature.

[0078] Finally, the system integrates all the collected information, including the current venue, human body distribution, movement mode, sports equipment type, and action sequence characteristics, and uses statistical models (such as decision trees or naive Bayes classifiers) to infer the most likely training project type. The system outputs a possible training project type sorted by probability to provide preliminary contextual information for subsequent in-depth analysis.

[0079] Step S130, process the video data according to the auxiliary information, obtain the student position, motion trajectory and the second type of identifier, detect invalid data in the video data according to the student position, motion trajectory and the second type of identifier, and mark the video segment containing invalid data as a pending state.

[0080] Optimize video data:

[0081] The system first uses the site identification information extracted above to accurately locate the region of interest (ROI). This process involves analysis and segmentation of the entire video frame. For pull-up training, the system sets the ROI to the horizontal bar area and its surroundings. Specifically, the system identifies the position of the horizontal bar, and then expands a certain pixel range up, down, left, and right with the horizontal bar as the center to form a rectangular area. The size of this area is dynamically adjusted according to the angle and distance of the camera to ensure that the entire movement process of the student can be fully captured.

[0082] During a running session, the system focuses on the track area. This includes identifying the start and end points of the track and the boundaries between the individual tracks. A polygonal ROI is created that accurately covers the entire track area while excluding surrounding spectator seating or other non-relevant areas. For a circular track, a donut-shaped ROI is created that accurately follows the curve of the track.

[0083] Next, the system adjusts the image processing parameters based on the type of training program. In the pull-up training video, the system enhances the contrast of the horizontal bar and the student's upper body area. This process includes local contrast enhancement and edge sharpening. The system applies an adaptive histogram equalization algorithm to the image within the ROI to improve the visibility of dark details. At the same time, unsharp masking technology is used to enhance the edge clarity of key parts such as the horizontal bar and the student's arms and torso.

[0084] For long jump training, the system focuses on improving the clarity of the take-off and landing areas. This involves precise positioning and image enhancement of these two specific areas. The system applies a local contrast enhancement algorithm to the take-off board and bunker area to improve the texture details in these areas. At the same time, a denoising algorithm is used to reduce image noise that may affect accurate measurements.

[0085] The system also determines the current lighting conditions based on the timestamp information and adjusts the brightness and contrast parameters accordingly. For example, if the timestamp shows that it is early morning or late evening, the system will expect the light to be dim, so it will increase the overall brightness and contrast appropriately. If it is midday, the system may reduce the contrast to avoid overexposure. This dynamic adjustment ensures that videos shot at different time periods can maintain optimal visibility.

[0086] To handle blur caused by fast motion, the system uses timestamp information to calculate the possible motion speed, and then applies a deblurring algorithm to the area where blur is detected. Specifically, the system first calculates the time interval between adjacent frames, and estimates the motion speed in combination with the displacement of objects within the ROI. The system then uses this speed information to parameterize the deblurring algorithm. For example, during a running training session, the system pays special attention to the fast-moving leg area. It uses a directional deblurring filter, in which the direction and strength of the filter are dynamically adjusted based on the estimated motion direction and speed.

[0087] Finally, the system performs spatiotemporal resolution optimization. By analyzing the timestamp information, the system can accurately calculate the actual frame rate of the video. If the frame rate is not enough to capture fast action (for example, below 30fps), the system will perform frame interpolation. This process uses an optical flow estimation algorithm to create smooth transitions between adjacent actual frames. For example, if the original video is 25fps, the system may increase it to 50fps, inserting a calculated intermediate frame between every two original frames.

[0088] The student position identification, motion trajectory generation and second identifier allocation are performed based on the following steps:

[0089] First, the system processes each frame of the optimized video data using advanced deep learning object detection algorithms, such as YOLOv5 or Faster R-CNN. These algorithms are specially trained to accurately identify students in complex sports training scenarios. The algorithms not only consider the students' appearance features, but also recognize the students' shapes in different sports postures. During the detection process, the system generates a bounding box for each identified student, accurately locating the student's position and size in the image.

[0090] In order to obtain a more accurate student position, the system calculates the coordinates of the center point of each bounding box. This center point (x, y) is regarded as the representative of the student's position in the frame. The coordinate system is based on image pixels, usually with the upper left corner as the origin (0, 0), the right as the positive direction of the x-axis, and the bottom as the positive direction of the y-axis. For example, if the coordinates of the upper left corner of a student's bounding box are (100, 150) and the coordinates of the lower right corner are (200, 250), then the coordinates of the student's position will be calculated as ((100+200) / 2, (150+250) / 2), that is, (150, 200).

[0091] At this stage, the system introduces the concept of position information Pt. Pt represents the three-dimensional position coordinates of the student at time t, defined as:

[0092] Pt = (x, y, z)

[0093] Among them, x and y are the two-dimensional coordinates obtained by the above calculation, and z is the depth information obtained by multi-view triangulation or depth estimation algorithm. In this way, the system can accurately locate the student's position in three-dimensional space.

[0094] The system also uses the extracted site identification information to improve the accuracy of position recognition at this stage. For example, during running training, the system will compare the detected student position with the known track position. If the detected position is significantly off track, the system will mark the detection result as suspicious and give special attention in subsequent processing. Similarly, during pull-up training, the system will pay special attention to the area around the horizontal bar, improve the detection sensitivity in this area, and ensure that the student can be accurately located even in the case of partial occlusion.

[0095] Next, the system uses the timestamp information to connect the position coordinates of the same student in consecutive frames to form a motion trajectory. This process is not just a simple point connection, but involves complex mathematical calculations and kinematic analysis. The system first calculates the displacement vector between adjacent frames. Assuming the position at time t1 is (x1, y1, z1) and the position at time t2 is (x2, y2, z2), then the displacement vector will be (x2-x1, y2-y1, z2-z1).

[0096] Based on this displacement vector and the time interval between frames (t2-t1), the system calculates the instantaneous velocity. The velocity is calculated as: v = (x2-x1, y2-y1, z2-z1) / (t2-t1). This velocity is a three-dimensional vector that contains velocity components in the horizontal, vertical, and depth directions. The system also calculates the scalar magnitude of the velocity, sqrt((x2-x1)^2 + (y2-y1)^2+ (z2-z1)^2) / (t2-t1), which represents the absolute speed of the student's movement.

[0097] Furthermore, the system calculates acceleration by comparing the velocity changes at adjacent time points. The acceleration calculation formula is: a = (v2 - v1) / (t2-t1), where v1 and v2 are velocity vectors at two adjacent time points.

[0098] In the process of generating motion trajectories, the system records the position, velocity, and acceleration information at each time point to form a continuous motion trajectory data set.

[0099] At this stage, the system also introduces the concept of posture information A(t). A(t) represents the posture information of the student at time t and is defined as:

[0100] A(t)= {θ1, θ2, ..., θn}

[0101] Among them, θi represents the angle of the i-th key joint. These joint angles are obtained by pose estimation of the student image, using a human pose estimation algorithm such as OpenPose. The introduction of At enables the system to not only track the position of the student, but also analyze its action details.

[0102] Finally, in order to accurately track each student throughout the video sequence, the system implements a complex strategy for assigning and maintaining student second-class identifiers. When a new student target is first detected, the system assigns it a unique second-class identifier, such as "Student 1", "Student 2", etc. The second-class identifier is not just a simple number, but is associated with multiple features of the student to form a multi-dimensional feature vector. This vector contains the student's appearance features, initial position, motion features, and body shape information. Appearance features are visual features extracted by a deep learning model, the initial position records the 3D coordinates of the student when it is first detected, and the motion features include the initial velocity and acceleration estimates.

[0103] In one example, the steps of processing video data according to auxiliary information, obtaining a student position, a motion trajectory, and a second type of identifier, detecting invalid data in the video data according to the student position, the motion trajectory, and the second type of identifier, and marking a video segment containing invalid data as a pending state include the steps of:

[0104] A boundary model is established according to the site identification, and the student position is compared with the boundary model. If the student position exceeds the boundary model, the data segment corresponding to the student position exceeding the boundary model is recorded as invalid data;

[0105] Calculate the speed and acceleration at each time point according to the motion trajectory, compare the speed and acceleration with the preset human ability range, and if the speed and acceleration exceed the preset human ability range, mark the data segment corresponding to the speed and acceleration exceeding the preset human ability range as invalid data;

[0106] Retrieve whether the second-category identifiers corresponding to the students are consistent. When the number of changes of the second-category identifiers corresponding to the students exceeds a preset number within a preset time period, detect the inducing factors of the change of the second-category identifiers. If the inducing factors do not meet the preset inducing factors, mark the data segments corresponding to the second-category identifiers that do not meet the preset inducing factors as invalid data.

[0107] When the amount of invalid data accumulated in a video segment exceeds a preset threshold, the video segment is marked as being in a pending state.

[0108] First, the system uses the site identification information to establish an accurate boundary model. This model includes not only the outer boundaries of the site, but also the internal functional area division. For example, in running training, the model will accurately define the boundaries of each track; in pull-up training, the horizontal bar area and the safety range around it will be marked in detail. The system compares the coordinates of each student's position with this boundary model. If the detected student position is beyond the area where it should be, the system will immediately mark it as potential invalid data. This check is not just a simple boundary judgment, but also takes into account the characteristics of different training items. For example, in long jump training, the system will allow students to leave the take-off area briefly, but if it is detected that the student's position is in a non-training area for a long time, it will be marked as an abnormality.

[0109] Secondly, the system conducts an in-depth analysis of the continuity of the generated motion trajectory. This analysis process uses timestamp information to not only check the changes between adjacent frames, but also evaluate the overall trend of the movement over a longer time window. The system sets specific motion models for different training items. For example, in pull-up training, the system expects to see periodic up and down motion; in running training, it expects to see continuous forward motion. If the detected motion trajectory deviates significantly from these preset models, the system will perform a more in-depth analysis.

[0110] Specifically, the system calculates the student's speed and acceleration at each time point and compares it with the preset human ability range. If a speed or acceleration change that exceeds the normal human ability range is detected in a short period of time, the system will mark it as potentially invalid data. For example, in a 100-meter sprint training, if the system detects that the student's instantaneous speed suddenly jumps from 10m / s to 0m / s, and this change occurs within 0.1 seconds, it is obviously beyond the possible deceleration range of the human body, and the system will mark this data as abnormal.

[0111] The system also pays special attention to the consistency of the assigned second-category identifiers. This checking process is not just a simple monitoring of whether the identifier has changed, but also a comprehensive judgment based on other information. The system maintains a time window (for example, 5 seconds) within which to track the changes of each second-category identifier. If a student's second-category identifier changes more than a preset number of times (for example, 3 times) within this time window, the system will trigger a deeper check.

[0112] During the in-depth inspection, the system analyzes possible reasons (causative factors) for the change in the identifier. For example, it checks if there are other students nearby, if there is occlusion, or if the student has made an action that may cause a significant change in appearance (such as a quick turn). If none of these triggers are true, the system will mark this data as potentially invalid. At the same time, the system will try to infer the correct identifier by analyzing the information of the previous and next frames to maintain the continuity of tracking.

[0113] When invalid data accumulates, the system uses a weighted calculation method. Eventually, when the amount of accumulated invalid data exceeds a preset threshold (for example, the weighted total score reaches 30 points, with a full score of 100 points), the system will mark the unique identifier of the video clip as pending.

[0114] In one example, the method further includes the steps of: calculating the curvature of the student's trajectory, speed change, direction change and posture change based on the motion trajectory; calculating the complexity index of the student's movement based on the trajectory curvature, speed change, direction change and posture change; if the complexity index exceeds a preset normal range, determining that an error has occurred in the acquisition process or the processing process of the video data.

[0115] At this stage, the system introduces the motion complexity index K = w1 * (C / Cmax) + w2 * (V / Vmax) + w3 * (D / Dmax) + w4 * (P / Pmax)

[0116] in:

[0117] - Curvature of trajectory , n is the number of sampling points, θ i is the tangent angle at the ith point, and Δs is the arc length between adjacent points.

[0118] - Speed ​​change , vi is the speed at the ith time point, is the average speed.

[0119] - Direction change , is the angle of the motion direction at the i-th time point, and Δt is the time interval.

[0120] - Posture changes , m is the number of joint points, α j,i is the angle of the j-th joint at the i-th time point.

[0121] w1, w2, w3, and w4 are weight coefficients, and their sum is equal to 1. Cmax, Vmax, Dmax, and Pmax are the preset maximum values ​​of C, V, D, and P, respectively, which are used to normalize each component to the range of [0, 1].

[0122] The system normalizes the K value to a range of 0 to 10, where 0 represents the simplest straight-line uniform motion and 10 represents an extremely complex motion pattern. By calculating the K value, the system can identify abnormally complex motion patterns (complexity index exceeds the preset normal range), which may represent errors in the data acquisition or processing process.

[0123] Step S140 , performing data compensation processing on the video segments in a to-be-processed state, and integrating the compensated video segments with the video segments not marked as to-be-processed states to obtain a complete data set.

[0124] In one example, a step of performing data compensation processing on a video clip in a to-be-processed state, and integrating the compensated video clip and the video clip not marked as being in a to-be-processed state to obtain a complete data set includes the following steps: in terms of timing, selecting a data frame as a first reference point in the video clip that is before and closest to the video clip in a to-be-processed state; in terms of timing, selecting the first normal data frame after the video clip in a to-be-processed state as a second reference point; calculating a data credibility coefficient according to the length of the video clip in a to-be-processed state, the time distance between the first reference point and the second reference point, the complexity index of the student's movement, and the compensated speed and the original speed of the student's movement; performing weighted calculation on the first reference point, the second reference point and the data credibility coefficient to obtain a compensation estimate; obtaining compensation data according to the compensation estimate, the length of the video clip in a to-be-processed state, the length of the video data and the data credibility coefficient; inserting the compensation data into the video clip in a to-be-processed state to compensate for the video clip in a to-be-processed state.

[0125] First, find the nearest data frame (i.e., valid data frame) that is not marked as a pending state in the front part of the video clip to be processed as a reference point. This reference point contains reliable position coordinates, speed, acceleration, student identifier and other information. It is selected because it is closest to the interval to be compensated in time and its data is most likely to reflect the actual situation of the interval to be compensated. For example, in a 100-meter sprint training, if the system detects an anomaly at 60 meters, it will select the data frame closest to 60 meters, located before 60 meters and not marked as an anomaly as a reference point. This reference point may be located at 59.8 meters.

[0126] Next, the system determines the second reference point for compensation calculations, called the "next valid frame". This is the first data frame that is not marked as abnormal (i.e., the first normal data frame) after the video segment in the pending state. The purpose of selecting this frame is to provide a future reference point for compensation calculations in order to more accurately estimate the data change trend within the interval to be compensated. When determining the next valid frame, the system takes into account the periodicity of the movement. For sports with obvious periodicity, such as running, the system will try to select a frame that is in a similar motion phase as the reference point, usually by analyzing the student's step frequency. For example, if the system finds that the student's average step frequency is 4 steps per second, then each complete step cycle takes 0.5 seconds. The system will try to move forward in time by 0.5 seconds or an integer multiple of it to find the next valid frame to ensure that the selected frame represents the beginning of the same motion phase.

[0127] With the data of these two key frames, the system begins the core calculation of data compensation. In this process, the system introduces the data credibility coefficient C(t) to quantify the reliability of the compensation data. The calculation of C(t) is based on the length of the video clip to be processed, the time distance between the first reference point and the second reference point, the complexity index of the movement, and the compensated speed and original speed of the student's movement. The calculation formula of C(t) is as follows:

[0128] C(t) = exp(-a * L / L_max - b * D / D_max - c * K / K_max) * (1 - d * |V_comp - V_orig| / V_max)

[0129] in:

[0130] L is the length of the interval to be compensated, and L_max is the preset maximum acceptable compensation interval length;

[0131] D is the time distance between the reference point and the next valid frame, and D_max is the preset maximum acceptable time distance;

[0132] K is the motion complexity index, which is obtained from the motion trajectory analysis in step 3, and K_max is the preset maximum complexity value;

[0133] V_comp is the compensated speed (compensated speed), V_orig is the original speed, and V_max is the preset maximum speed difference;

[0134] a, b, c, d are adjustment coefficients used to balance the influence of various factors;

[0135] The value range of C(t) is [0, 1]. The larger the value, the higher the credibility. The system directly incorporates C(t) into the calculation process of compensation data to adjust the credibility of the compensation result.

[0136] When calculating compensation data, the system uses a weighted average method while taking into account the influence of C(t). The revised calculation formula is as follows:

[0137] Compensation estimate = (reference point data * (1-C(t)) * first weight coefficient) + (next valid frame data * C(t) * second weight coefficient)

[0138] Here, C(t) directly affects the weight of the reference point data and the next valid frame data. When C(t) is small, the system will rely more on the reference point data; when C(t) is large, the system will consider the next valid frame data more. This method ensures that when the data credibility is low, the compensation result will tend to use more reliable historical data.

[0139] For non-numeric data such as student identifiers, the system first converts the identifier to a numeric value (for example, through a hash function) and then applies the weighted average method described above. After the calculation is completed, the system rounds the result to the nearest valid identifier to ensure that the final compensation result is a meaningful student identifier.

[0140] However, this estimate represents the average of the entire motion cycle, and the video clip to be processed may only be a part of the entire cycle. Therefore, the system needs to further adjust this estimate to make it suitable for the actual size of the interval to be compensated. The final compensation data is calculated as follows:

[0141] Compensation data = (compensation estimate * (length of interval to be compensated / length of complete motion cycle)) * C(t)

[0142] In this formula, C(t) is applied again to ensure that the final compensation data reflects its credibility. This means that for data with lower credibility, the system will correspondingly reduce its influence in the compensation result.

[0143] The system inserts the calculated compensation data into the video clip in the pending state, using interpolation techniques to ensure data continuity and smoothness. For continuous data such as position and speed, the system may generate multiple intermediate points in the video clip in the pending state to ensure a smooth transition from the last valid data point before the abnormal interval to the first valid data point after the abnormal interval. For student identifiers, the system rounds the calculation result to the nearest valid identifier and maintains the consistency of this identifier throughout the compensation interval.

[0144] After the data is inserted, the system performs a final check to ensure that the compensated data remains continuous and reasonable overall. This check process includes verifying whether the position change conforms to the laws of physics, whether the speed and acceleration are within a reasonable range, whether the student identifiers are consistent, etc. If all checks pass, the system updates the status of these video clips to "processed" and saves the corresponding C(t) values ​​for use in subsequent analysis.

[0145] In the step of integrating the compensated video clips and the video clips that are not marked as pending, obtaining a complete data set:

[0146] The system first ensures that all data fragments are correctly sorted according to their timestamps. This includes the original unlabeled data and the compensated data. The accuracy of the timestamp determines the temporal order of the action sequence.

[0147] After the sorting is completed, the system splices the video segments that are not marked as pending and the compensated video segments in chronological order. At the splicing point, the system performs processing to avoid sudden changes. This processing may involve interpolation algorithms to ensure that the transition of data at the splicing point is gradual. For example, if there is a clear jump in the position data at the splicing point, the system may use linear interpolation or spline interpolation to create a transition. This processing is critical to maintaining the continuity of the data because it ensures that subsequent analysis will not be affected by sudden changes at the data splicing point.

[0148] After integration, the system performs a comprehensive consistency check to ensure that there are no duplicate, missing or obvious discontinuous points in the data. This check process includes verifying that there is only one unique set of data at each time point, checking whether there are abnormal intervals in the time series, and ensuring that the changes in key parameters (such as position, velocity, acceleration) throughout the series are continuous. If any anomalies are found, the system will mark these points.

[0149] Step S150, based on the complete data set, extract the posture point sequence of the student during the movement process, wherein the posture point sequence includes the spatial position, joint angle and relative position relationship of the human body joint points.

[0150] In one example, the steps of extracting a sequence of posture points during a student's movement based on a complete data set include the steps of: processing the complete data set using a deep learning model to identify the student's body joints;

[0151] According to the calibration parameters of the camera and the geometric information of the site, the spatial position of the human body joints in the three-dimensional space is calculated; according to the spatial position, the joint angle and relative position are calculated; the spatial position, joint angle and relative position are combined into a posture point sequence in chronological order.

[0152] After completing the data integration, the system updates the metadata of the integrated dataset, including the total number of frames, duration, student IDs involved, and other information. Next, based on the integrated complete dataset, the system begins to extract the detailed sequence of posture points during the student's movement. The system uses deep learning models (such as OpenPose or HRNet) to identify the main joints of the students in each frame. These models have been trained with a large amount of human posture data and can accurately locate joints in various complex backgrounds and postures.

[0153] The identified joints usually include the head (top of the head, nose tip), torso (neck, shoulder center, chest center, waist center), upper limbs (left and right shoulders, left and right elbows, left and right wrists) and lower limbs (left and right hips, left and right knees, left and right ankles). These points are selected based on the needs of human motion analysis and can fully describe the posture changes of the human body during motion.

[0154] For each identified joint point, the system calculates its spatial position in 3D space. This requires combining the calibration parameters of the camera and the geometric information of the venue. Camera calibration is a key step that establishes the mapping relationship between 2D image coordinates and 3D world coordinates. Each joint point is represented by (x, y, z) coordinates, where x and y represent the plane position and z represents the depth information.

[0155] For example, for a student running on a 100-meter track, the system might calculate the position of the left ankle as (23.5m, 1.2m, 0.1m), meaning it is 23.5 meters from the start of the track, 0.1 meters above the ground, and 1.2 meters laterally offset. This precise three-dimensional positioning enables subsequent motion analysis to take into account actual distances and positions in space, not just projections on the image plane.

[0156] After obtaining the spatial position of the joint points, the system calculates the joint angles of the main joints. This involves vector calculations and trigonometric functions. For example, the knee joint angle is obtained by calculating the angle between the thigh vector and the calf vector. Specifically, the system first determines the vector from the hip to the knee and the vector from the knee to the ankle, and then calculates the angle between these two vectors using the dot product formula. Similarly, the elbow joint angle is calculated by the angle between the upper arm vector (shoulder to elbow) and the forearm vector (elbow to wrist). These angles are usually expressed in degrees, ranging from 0 to 360 degrees.

[0157] In addition to spatial position and joint angles, the system also calculates the relative positional relationships between joint points. These relative positional relationships can provide more information about movement posture and skills. For example, the system calculates the arm swing amplitude, which can be quantified by calculating the change in the position of the wrist relative to the hip. Stride is another important relative position indicator, which can be measured by calculating the maximum distance between the two feet. The torso inclination can be determined by calculating the angle between the line connecting the center of the shoulder to the center of the hip and the vertical line.

[0158] During the entire extraction process, the system organizes the posture data of each frame into a time series. This includes the movement trajectory of each joint point over time, the change curve of the joint angle, and the dynamic changes of the relative position relationship. For example, in a long jump attempt, the system can record the position changes of each joint point, the changes in key joint angles, and the movement trajectory of the body's center of gravity from the start of the run-up to the landing.

[0159] Finally, the system organizes all the extracted posture data into a structured posture point sequence dataset. This dataset contains detailed posture information for each frame, including:

[0160] 1. Spatial coordinates of all joint points

[0161] 2. Main joint angles

[0162] 3. Key relative position relationships

[0163] 4. Corresponding timestamp information

[0164] Step S160, comparing the posture point sequence with the standard action template, identifying the sports technical actions in the complete data set, and calculating the completion score of the sports technical actions.

[0165] In one example, the steps of comparing a posture point sequence with a standard action template, identifying sports technical actions, and calculating the completion score of the sports technical actions include: using a dynamic time warping algorithm to align the posture point sequence with the standard action template, and calculating the similarity score between the posture point sequence and the standard action template; obtaining a dynamic weight system based on the similarity score, and identifying a video clip in the complete data set that best matches the standard action template based on the dynamic weight system; identifying sports technical actions based on the video clips, and calculating the completion score of the sports technical actions based on the similarity score, the dynamic weight system and the data credibility coefficient.

[0166] The system first loads a predefined library of standard action templates, which contains standard posture sequences of key technical actions in various sports. Each standard action template consists of a series of key frames, including joint positions, angles, and relative position relationships in an ideal state. These templates are developed by professional coaches and students through a lot of practice and research, and represent the best execution methods for various sports techniques. The template library includes not only static postures, but also dynamic motion sequences to capture subtle changes throughout the action process.

[0167] Next, the system uses an improved dynamic time warping (DTW) algorithm to align the pose point sequence with the standard action template. The DTW distance function can be expressed as: D(t1, t2) = λ1 * ||A(t1) - A(t2)|| + λ2 * ||P(t1) - P(t2)||. Among them, λ1 and λ2 are weight coefficients for balancing pose similarity and position consistency, P(t) is the student position at time t, and A(t) is the pose information at time t.

[0168] In one example, a posture point sequence is aligned with a standard action template using a dynamic time warping algorithm, and the steps of calculating a similarity score between the posture point sequence and the standard action template include the steps of: calculating a first score based on the original posture point sequence in the video data, the standard action template, and the data credibility coefficient; calculating a second score based on the posture point sequence, the standard action template, and the data credibility coefficient; and obtaining a similarity score based on the first score, the second score, and the data credibility coefficient.

[0169] During the alignment process, the system calculates the similarity between the student's posture point sequence and the standard action template. Here, an innovative "dynamic credibility weighting" mechanism is introduced, which uses the data credibility coefficient C(t) to adjust the credibility of the similarity calculation. For each time point t, the system calculates the similarity score of the original data and the compensated data separately:

[0170] First score S_orig(t) = Similarity(A_student_orig(t), A_template(t))

[0171] Second score S_comp(t) = C(t) * Similarity(A_student_comp(t), A_template(t))

[0172] Where A_student_orig(t) is the student posture in the original unlabeled data, A_student_comp(t) is the student posture in the compensated data, A_template(t) is the posture of the standard action template, and C(t) is the data credibility coefficient calculated in step 4. The Similarity() function calculates the similarity between the student posture A_student(t) and the template posture A_template(t), using the Similari cosine similarity calculation method, represents the posture as a high-dimensional vector, and then calculates the cosine value of the angle between the two vectors.

[0173] The final similarity score S(t) = (S_orig(t) + C(t) * S_comp(t)) / (1 + C(t))

[0174] By introducing C(t), the system can properly handle data segments that may have problems to ensure the reliability of the score. For example, if a brief occlusion or sensor failure occurs during the execution of an action, resulting in low data credibility for some frames, the system will correspondingly reduce the impact of these frames in the overall score, thereby avoiding score deviations caused by data quality issues.

[0175] On this basis, the system introduces a dynamic weight system Wi(t). Wi(t) is directly related to S(t) and reflects the relative importance of the i-th evaluation factor at the current moment t, such as posture accuracy, range of motion, speed control, balance, etc. For example, when evaluating a pull-up, i=1 may represent the pull-up strength, i=2 may represent the smoothness of the movement, i=3 may represent body stability, and so on.

[0176] Wi(t) is calculated using the sigmoid function, which is an S-shaped nonlinear function that can provide smooth weight adjustment when the similarity changes. The specific calculation formula is: Wi(t) = 1 / (1 + e^(-k(S(t) - θi))). Where k is the parameter that controls the steepness of the function, and θi is the threshold parameter for the i-th evaluation factor. The characteristics of this function are that when S(t) is much smaller than θi, Wi(t) is close to 0; when S(t) is much larger than θi, Wi(t) is close to 1; when S(t) is close to θi, Wi(t) changes rapidly. This design allows the system to dynamically adjust the importance of each evaluation factor based on the execution of the current action. For example, in the pull-up action, when the student's posture point sequence is highly similar to the standard action template (that is, S(t) is large), the system will increase the weight of more advanced factors such as power output and movement stability because the basic posture has reached the standard; when the posture similarity is low, the system will pay more attention to the correctness of the basic posture to establish a good foundation for the action. This adjustment reflects the priorities in actual training: get the basics right first, then focus on more advanced performance factors.

[0177] Using this dynamic weighting system, the system can more accurately identify the segments in the student's pull-up sequence that best match the standard action template, and these segments are marked as specific sports technical movements. The system can not only identify the traditional key movements, but also capture some subtle but critical transition movements. For example, in the pull-up movement, the system can not only identify the key posture of pulling up to the highest point, but also capture the contraction of the shoulder blades when starting to exert force, the gradual process of the arms from straightening to bending during the pull-up process, the tiny stabilization adjustments of the body when reaching the highest point, and the speed control and eccentric contraction of the muscles during the lowering process.

[0178] For each identified technical action, the system calculates its completion score. This score reflects the degree of closeness of the action performed by the student to the standard action template. The calculation of the completion score uses a nonlinear function, which can better reflect the mutual influence between different factors and the dynamic changes of weights. Let S be the final completion score, Si be the score of the i-th evaluation factor, then: S = G(W1(t)S1, W2(t)S2, ..., Wn(t)Sn, C(t)). Among them, G is a nonlinear function, and C(t) is the overall data credibility at time t. This nonlinear calculation method avoids the distortion that may be caused by traditional linear weighting, making the scoring results closer to the judgment of professional coaches.

[0179] Specifically, the G function takes the form of power average: . Where p is a parameter greater than 1, which is used to adjust the degree of nonlinearity. Through a large number of experiments and expert evaluations, the value of p was set to 2.5. This value can maintain the sensitivity of the score while appropriately emphasizing the impact of high-scoring items. When p=2.5, this formula can well simulate the scoring logic of human coaches, taking into account the comprehensive performance of various evaluation factors and highlighting the importance of key movements. For example, when evaluating a set of pull-ups, even if other aspects perform well, if there is obvious swinging or leveraging in the key pull-up phase, the overall score will be significantly affected. This is consistent with the actual pull-up training evaluation criteria, because the quality of the pull-up phase directly reflects the core strength and purity of the technique.

[0180] Step S170, generating a comprehensive analysis report based on the sports technical movements and completion scores.

[0181] The first point of evaluation is the accuracy of the movement. This is directly derived from the completion score of each sports technical movement. The system will analyze the scores of each key movement stage in detail, indicating the high and low scores. The evaluation of movement accuracy not only focuses on the posture at a single moment, but also takes into account the continuous changes throughout the movement. For example, when evaluating the pull-up movement, the system not only analyzes the posture score at the highest point of the pull-up, but also evaluates the coordination of various parts of the body during the entire pull-up process. The system may find that the score at the beginning of the pull-up is high (such as 95 points), indicating that the technical movement in the starting stage is standard; while the score near the top is low (such as 75 points), which may indicate that there are technical problems in the second half of the movement, such as excessive elbow abduction or insufficient shoulder internal rotation.

[0182] In the running action assessment, the system analyzes the entire process from start to sprint. It may find that the student's mid-running posture scores the highest (such as 92 points), indicating that the technical action is most ideal in the stable speed stage; while the start stage scores are lower (such as 78 points), suggesting that the start technique training needs to be strengthened. Through this detailed score analysis, coaches and students can accurately locate the specific action links that need to be improved, so as to make targeted technical adjustments and strengthen training.

[0183] The second key evaluation point is movement consistency. The system analyzes the completion scores of multiple consecutive movement cycles and calculates their standard deviation and coefficient of variation. This indicator reflects the student's ability to maintain stable technical movements throughout the movement, which is crucial for evaluating the student's technical proficiency and physical condition. A small coefficient of variation indicates that the student is able to perform high-quality movements consistently and stably, which is usually a characteristic of high-level students. A large coefficient of variation may mean unstable technology or insufficient physical fitness, which requires further analysis and improvement.

[0184] For example, when evaluating long-distance running, the system will analyze the changes in stride and cadence throughout the entire process. If the stride and cadence completion scores are found to drop significantly and fluctuate in the second half, the system will point out that this may be due to a technical breakdown caused by insufficient endurance. In this case, the report may recommend adding special endurance training or adjusting the competition strategy to better allocate physical energy. When evaluating repetitive movements, such as continuous pull-ups or push-ups, analysis of movement consistency can reveal the impact of fatigue on technical movements. If the system finds that movement consistency drops significantly in the last few repetitions, it may recommend adding strength endurance training or paying more attention to maintaining technology under fatigue during training.

[0185] Movement fluidity is the third important evaluation point. This is mainly assessed by analyzing the completion scores between adjacent movements. A high completion score means that the movements flow naturally without obvious pauses or incoordination. Movement fluidity not only affects the aesthetics of the movement, but is also directly related to the efficiency and performance of the movement. Fluency analysis is particularly important when evaluating continuous movements, such as skipping or rhythm maintenance in long-distance running.

[0186] Taking long jump as an example, the system will evaluate the entire process from run-up to take-off, to aerial movements and landing. If the score is low in the transition phase from run-up to take-off, it may indicate that the student is hesitant or the rhythm is interrupted at this critical moment, which will directly affect the effect of the take-off. The system will specifically point out whether the step frequency is improperly adjusted or the center of gravity is not smooth, thus providing clear direction for technical improvement.

[0187] Movement efficiency is the fourth evaluation focus. The system evaluates efficiency by analyzing specific sports technical movements and their corresponding completion scores. Efficient movements usually receive higher completion scores because they are often closer to the standard technical movement model.

[0188] When evaluating running movement efficiency, the system will focus on the identified sports technical movements, such as arm swing, leg lift, and ground push, and analyze the completion scores of these movements. For example, if it is shown that the student's arm swing movement score is low (such as 70 points) and the leg lift movement score is high (such as 90 points), the system will point out that there may be problems with the coordination of the upper and lower limbs, which may lead to lower overall running efficiency. The report may recommend strengthening the coordination training of the upper and lower limbs to improve the overall movement efficiency.

[0189] For the rope skipping event, the system will analyze the completion scores of the identified jumping and rope swinging movements. If the jumping movement score is high (such as 95 points), but the rope swinging movement score is low (such as 75 points), the system will point out that although the lower limb movement is efficient, the upper limb movement may have unnecessary large swings, reducing the overall efficiency. Based on this analysis, the report may recommend optimizing the wrist rotation technique to achieve higher movement efficiency.

[0190] When evaluating the efficiency of the pull-up, the system will focus on analyzing the completion scores of the two main phases identified, the pull-up and the lowering phase. For example, if the scores of the pull-up phase are generally higher than the lowering phase (such as 90 points for the pull-up and 70 points for the lowering), the system will indicate that the student may have efficiency issues in eccentric contraction control. This may lead to additional energy consumption and reduce the ability to perform multiple pull-ups in a row. Based on this analysis, the system may recommend strengthening the training of eccentric contraction control to improve overall movement efficiency.

[0191] In addition, the system also analyzes the completion scores between consecutive movements. Efficient movements are usually smooth and coherent, so if the score in the transition phase is low, it may mean that there is a problem with the efficiency of the movement. For example, in a continuous pull-up movement, if the completion score from the end of one movement to the beginning of the next is low, the system will point out that this may lead to unnecessary energy loss, and it is recommended to focus on improving the continuity of the movement to improve overall efficiency.

[0192] Finally, the system generates a comprehensive score based on these evaluation points. This score is a weighted average of the scores of each evaluation point, and the weights vary depending on the characteristics of the sport. For example, when evaluating sprinting, the weight of movement accuracy and efficiency may be higher; when evaluating long-distance running, the weight of movement consistency and fluency may be greater. This flexible scoring mechanism ensures that the evaluation results can accurately reflect the specific requirements of different sports.

[0193] The final section of the report provides specific training recommendations. These recommendations are based directly on the above assessment results and provide clear training directions for each identified problem or area for improvement. The recommendations will cover multiple aspects such as technical training, physical training, and psychological training to form a comprehensive improvement plan.

[0194] For example, if the system finds that a student excels in movement accuracy but has problems with movement consistency, it may recommend adding endurance training and practicing technical movements in a fatigued state to improve the ability to maintain correct movements for a long time. If movement fluidity is the main problem, the system may recommend adding rhythm training and movement continuity exercises, and may also recommend some auxiliary coordination training. For problems with movement efficiency, recommendations may include fine-tuning of technical movements and targeted strength training to optimize force application.

[0195] The present application includes the following steps: collecting video data of the student's movement process, parsing imaging parameters from the video data, and assigning corresponding first-class identifiers to the video clips in the video data; extracting auxiliary information from the video data based on the imaging parameters and the first-class identifiers; processing the video data according to the auxiliary information, obtaining the student's position, movement trajectory, and second-class identifiers, detecting invalid data in the video data according to the student's position, movement trajectory, and second-class identifiers, and marking the video clips containing invalid data as pending; performing data compensation processing on the video clips in the pending state, and integrating the compensated video clips with the video clips not marked as pending to obtain a complete data set; extracting a posture point sequence during the student's movement based on the complete data set; the posture point sequence includes the spatial position, joint angle, and relative position relationship of the human body's joints; comparing the posture point sequence with the standard action template, identifying the sports technical actions in the complete data set, and calculating the completion score of the sports technical actions; generating a comprehensive analysis report based on the sports technical actions and the completion score. This application conducts in-depth analysis of imaging parameters during video data processing to improve the accuracy of motion recognition, analyzes auxiliary information to improve the pertinence of motion recognition, processes invalid data to improve the accuracy of motion recognition, and compensates for data to improve the reliability and effectiveness of motion recognition.

[0196] This application integrates high-definition video data, imaging parameters, and contextual information such as venue identification and training project types to achieve comprehensive multi-dimensional data fusion. The core innovation lies in the introduction of the data credibility coefficient C(t) and the establishment of a dynamic data compensation mechanism. This mechanism takes into account multiple factors, such as the length of the compensation interval L, the time distance D, the motion complexity K, etc., and effectively solves the problems of data missing, occlusion and anomalies in complex training environments. Through the dynamic calculation of C(t), the system can adaptively adjust the compensation intensity according to the data reliability, significantly improving the applicability and analysis accuracy in various sports.

[0197] This application proposes a motion complexity index K and a dynamic weight system Wi(t) to construct an adaptive motion evaluation system. K comprehensively considers trajectory curvature C, speed change V, direction change D, and posture change P, providing a unified quantitative standard for different types of motion. Wi(t) can adjust the importance of evaluation factors in real time according to the characteristics of different motion stages, and is closely related to the similarity score S(t). This innovation enables the system to accurately capture the characteristics of various types of motion, from simple repetitive movements (such as skipping) to complex multi-stage movements (such as pull-ups), and flexibly switch evaluation criteria between different items such as running, skipping, and pull-ups.

[0198] This application establishes a full-process quality control mechanism from data collection to scoring, which significantly improves the system's ability to process actual training data. The innovatively introduced concept of "pending state" and complex invalid data detection algorithm can effectively identify and handle problems such as position anomalies, motion trajectory continuity anomalies, and student identifier anomalies. The dynamic credibility weighting mechanism is adopted in the scoring stage. Through S(t), the original data S_orig(t) and the compensation data S_comp(t) are cleverly combined to ensure the accuracy of the score when the data quality fluctuates. The introduction of the nonlinear scoring function G, coupled with the adjustable parameter p, enables the system to better simulate the judgment logic of professional coaches and adapt to the characteristics of different sports. This system not only improves the accuracy and consistency of action evaluation, but also enhances the ability to analyze the fluency and efficiency of actions.

[0199] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0200] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 2 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as video data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a multi-dimensional intelligent analysis method for the student movement process in a smart sports training area is implemented.

[0201] Those skilled in the art will understand that Figure 2The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0202] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0203] Collect video data of the student's exercise process, parse imaging parameters from the video data, and assign corresponding first-class identifiers to video segments in the video data;

[0204] Extract auxiliary information from the video data based on the imaging parameters and the first type of identifier; the auxiliary information includes frame-by-frame extraction timestamps, site identifiers, and training project types;

[0205] Process the video data according to the auxiliary information, obtain the student position, motion trajectory and the second type of identifier, detect invalid data in the video data according to the student position, motion trajectory and the second type of identifier, and mark the video segment containing invalid data as a pending state;

[0206] Perform data compensation processing on the video clips in the to-be-processed state, and integrate the compensated video clips with the video clips that are not marked as to-be-processed to obtain a complete data set;

[0207] Based on the complete data set, the posture point sequence of the students during the movement is extracted; the posture point sequence includes the spatial position, joint angle and relative position relationship of the human body joints;

[0208] Compare the posture point sequence with the standard action template, identify the sports technical actions in the complete data set, and calculate the completion score of the sports technical actions;

[0209] Generate a comprehensive analysis report based on sports technical movements and completion scores.

[0210] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0211] Collect video data of the student's exercise process, parse imaging parameters from the video data, and assign corresponding first-class identifiers to video segments in the video data;

[0212] Extract auxiliary information from the video data based on the imaging parameters and the first type of identifier; the auxiliary information includes frame-by-frame extraction timestamps, site identifiers, and training project types;

[0213] Process the video data according to the auxiliary information, obtain the student position, motion trajectory and the second type of identifier, detect invalid data in the video data according to the student position, motion trajectory and the second type of identifier, and mark the video segment containing invalid data as a pending state;

[0214] Perform data compensation processing on the video clips in the to-be-processed state, and integrate the compensated video clips with the video clips that are not marked as to-be-processed to obtain a complete data set;

[0215] Based on the complete data set, the posture point sequence of the students during the movement is extracted; the posture point sequence includes the spatial position, joint angle and relative position relationship of the human body joints;

[0216] Compare the posture point sequence with the standard action template, identify the sports technical actions in the complete data set, and calculate the completion score of the sports technical actions;

[0217] Generate a comprehensive analysis report based on the sports technical actions and completion scores. A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it may include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0218] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0219] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims

1. A multi-dimensional intelligent analysis method for student exercise process in a smart sports training area, characterized in that: The following steps are involved: Collecting video data of the student's exercise process, parsing imaging parameters from the video data, and assigning corresponding first-category identifiers to video segments in the video data; Extracting auxiliary information from the video data based on the imaging parameters and the first type of identifier; the auxiliary information includes a frame-by-frame extraction timestamp, a venue identifier, and a training item type; Process the video data according to the auxiliary information, obtain the student position, motion trajectory and second type of identifier, detect invalid data in the video data according to the student position, motion trajectory and second type of identifier, and mark the video segment containing the invalid data as a pending state; Performing data compensation processing on the video clips in a to-be-processed state, and integrating the compensated video clips with the video clips not marked as in a to-be-processed state to obtain a complete data set; Based on the complete data set, extract the posture point sequence of the student during the movement; the posture point sequence includes the spatial position, joint angle and relative position relationship of the human body joint point; Comparing the posture point sequence with a standard action template, identifying the sports technical actions in the complete data set, and calculating the completion score of the sports technical actions; Generate a comprehensive analysis report based on the sports technical actions and the completion scores; Wherein, in the step of extracting the auxiliary information in the video data based on the imaging parameters and the first type identifier, the training item type is obtained based on the following steps: Calculating the similarity between the venue identifier and a preset venue template, and taking the preset venue template corresponding to the highest similarity as the current venue; Determine the scope of the training program based on the current venue; Processing the video data using a lightweight human body detection algorithm to obtain a human body contour in the video data, and analyzing the student's movement pattern based on the human body contour; Analyzing the video data to identify the distribution of human bodies and types of sports equipment; capturing action sequence features within a preset duration in the video data according to the context of the data; Determine the type of the training item within the training item range using a statistical model according to the current venue, the movement pattern, the human body distribution, the type of sports equipment and the action sequence characteristics; The step of processing the video data according to the auxiliary information, obtaining the student position, motion trajectory and second-type identifier, detecting invalid data in the video data according to the student position, motion trajectory and second-type identifier, and marking the video segment containing the invalid data as a pending state includes the following steps: A boundary model is established according to the site identification, and the student position is compared with the boundary model. If the student position exceeds the boundary model, the data segment corresponding to the student position mark exceeding the boundary model is recorded as the invalid data; Calculating the speed and acceleration at each time point according to the motion trajectory, comparing the speed and acceleration with a preset human ability range, and if the speed and acceleration exceed the preset human ability range, marking the data segment corresponding to the speed and acceleration exceeding the preset human ability range as the invalid data; Retrieve whether the second-category identifier corresponding to the student is consistent, and when the number of changes of the second-category identifier corresponding to the student exceeds a preset number within a preset time period, detect the inducing factor of the change of the second-category identifier, and if the inducing factor does not meet the preset inducing factor, mark the data segment corresponding to the second-category identifier that does not meet the preset inducing factor as the invalid data; When the amount of the invalid data accumulated in the video clip exceeds a preset threshold, marking the video clip as the pending state; The step of performing data compensation processing on the video clips in the to-be-processed state and integrating the compensated video clips with the video clips not marked as to-be-processed state to obtain the complete data set comprises the following steps: In terms of time sequence, a data frame is selected as a first reference point in the video segment that is before the video segment in the to-be-processed state and is the closest to the video segment that is not marked as the to-be-processed state; In terms of timing, the first normal data frame after the video segment in the to-be-processed state is selected as the second reference point; Calculate the data credibility coefficient according to the length of the video segment in the to-be-processed state, the time distance between the first reference point and the second reference point, the complexity index of the student's movement, and the compensation speed and the original speed of the student's movement; Performing weighted calculation on the first reference point, the second reference point and the data credibility coefficient to obtain a compensation estimation value; Acquire compensation data according to the compensation estimation value, the length of the video segment to be processed, the length of the video data and the data credibility coefficient; Inserting the compensation data into the video segment in a to-be-processed state to compensate the video segment in the to-be-processed state; The step of extracting the posture point sequence of the student during the movement based on the complete data set includes the following steps: Process the complete data set using a deep learning model to identify the body joints of the students; Calculating the spatial position of the human body joints in three-dimensional space according to the calibration parameters of the camera and the geometric information of the site; Calculating the joint angle and the relative position according to the spatial position; Combining the spatial position, the joint angle and the relative position into the posture point sequence in chronological order; The step of comparing the posture point sequence with the standard action template, identifying the sports technical action, and calculating the completion score of the sports technical action includes the steps of: Using a dynamic time warping algorithm to align the posture point sequence with the standard action template, and calculating a similarity score between the posture point sequence and the standard action template; Acquire a dynamic weight system according to the similarity score, and identify the video clip in the complete data set that best matches the standard action template based on the dynamic weight system; The sports technical action is identified based on the video clip, and the completion score of the sports technical action is calculated according to the similarity score, the dynamic weight system and the data credibility coefficient.

2. The multi-dimensional intelligent analysis method of student exercise process in the smart sports training area according to claim 1 is characterized in that: Also includes the steps: Calculating the student's trajectory curvature, speed change, direction change, and posture change based on the motion trajectory; Calculating a complexity index of the student's motion according to the trajectory curvature, the speed change, the direction change, and the posture change; If the complexity index exceeds a preset normal range, it is determined that an error occurs in the acquisition process or the processing process of the video data.

3. The multi-dimensional intelligent analysis method of student exercise process in the smart sports training area according to claim 1 is characterized in that: The step of aligning the posture point sequence with the standard action template by using a dynamic time warping algorithm and calculating a similarity score between the posture point sequence and the standard action template comprises the following steps: Calculating a first score according to the original posture point sequence in the video data, the standard action template and the data credibility coefficient; Calculating a second score according to the posture point sequence, the standard action template and the data credibility coefficient; The similarity score is obtained according to the first score, the second score and the data credibility coefficient.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.