A physical training test method and system

Abstract

The application provides a physical training test method and system, the method comprises the following steps: synchronously acquiring motion capture data through a multi-modal data acquisition device, the motion capture data comprises optical action data or surface electromyogram signals; performing data preprocessing on the original motion capture data, the data preprocessing comprises coordinate system normalization processing and timestamp calibration processing; performing multi-level data validity verification processing on the preprocessed motion capture data to obtain a verification result; in response to the fact that the verification result represents valid data, when the motion capture data is optical action data, mapping the motion capture data into joint kinematics parameters based on a biomechanical model, and generating physical action evaluation parameters according to the joint kinematics parameters; the application can improve the scientificity and effectiveness of physical training.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of physical training variable testing, and more particularly to a physical training testing method and system. BACKGROUND

[0002] In the field of physical training variable testing, there are mainly two technical routes: the kinetic analysis scheme based on wearable sensors has the defect that the data verification mechanism is imperfect, when the sensor is out of connection or signal interference, the system cannot effectively identify and filter abnormal data, resulting in that the error data directly enters the calculation link. This kind of scheme also excessively relies on single kinetic parameter evaluation, lacks the analysis ability of key kinematic characteristics such as joint angle deviation, and is difficult to identify the action deformation problems such as knee joint inward buckling and spine hyperextension of athletes in deep squat, bench press and other compound actions.

[0003] The optical motion capture technology based on computer vision can obtain joint motion trajectory, but has obvious defects in multi-modal data fusion. When the system simultaneously collects electromyographic signals and optical data, due to the lack of accurate timestamp synchronization mechanism, the data streams collected by different sensors will have time sequence misalignment phenomenon, which seriously affects the accuracy of cross-modal data analysis. In addition, the existing optical scheme only focuses on the integrity verification of single frame data, and lacks effective detection means for data interruption problems in continuous motion process (such as the barbell stagnation stage in the weight lifting process).

[0004] The common defects of the prior art mainly exist in three aspects: first, the data verification system is too simple, only basic threshold judgment is performed without building a progressive multi-level verification process; second, the biomechanical parameter conversion algorithm is too simple, which cannot realize the accurate mapping from raw data to joint kinematic parameters; finally, the system lacks the deep mining ability of historical evaluation data, and cannot establish the action specification trend early warning mechanism. These problems seriously restrict the scientificity and reliability of the movement training evaluation. SUMMARY

[0005] The purpose of the present application is to provide a physical training testing method and system, which has the advantages of building a progressive multi-level verification process to improve the accuracy of data validity judgment and avoid error data entering the subsequent processing link.

[0006] The application provides a physical training test method, and the technical scheme is as follows: a physical training test method, characterized in that, comprising: synchronously acquiring motion capture data through a multi-modal data acquisition device, the motion capture data comprising optical action data or surface electromyogram signals; performing data preprocessing on the original motion capture data, the data preprocessing comprising coordinate system normalization processing and timestamp calibration processing; performing multi-level data validity verification processing on the preprocessed motion capture data to obtain a verification result; in response to the verification result representing valid data, when the motion capture data is optical action data, mapping the motion capture data into joint kinematics parameters based on a biomechanical model, and generating physical action evaluation parameters according to the joint kinematics parameters; wherein the multi-level data validity verification processing comprises: primary verification: verifying that the total number of data frames of the motion capture data reaches a preset frame number threshold; intermediate verification: verifying that the time interval between adjacent data frames is less than or equal to a preset time threshold; advanced verification: confirming that the proportion of invalid data points in the motion capture data is lower than a preset proportion threshold, and the length of a continuous valid data segment is greater than a preset continuous frame threshold; the multi-level data validity verification processing adopts progressive verification logic, and the subsequent verification is terminated when the current level verification fails.

[0007] Further, the application also provides that generating the physical action evaluation parameters comprises: performing frame processing on the motion capture data to obtain an action frame sequence; for each action frame in the action frame sequence, performing: verifying that the body movement region bounding box coverage rate is greater than a preset coverage rate threshold; performing bone joint positioning processing to obtain a bone joint position coordinate set; generating motion trajectory data based on the bone joint position coordinate set; performing speed analysis processing on the motion trajectory data to obtain an instantaneous speed vector; performing angle deviation calculation processing on the motion trajectory data to obtain a joint angle deviation value; generating a dynamics index according to the instantaneous speed vector and the joint angle deviation value; screening an action frame with a joint angle deviation value less than a preset angle threshold as a key action frame; performing feature extraction processing on the key action frame to obtain motion feature data; and generating the physical action evaluation parameters according to the motion feature data.

[0008] Further, the application also provides that the speed analysis processing comprises: performing time differentiation processing on the motion trajectory data to generate an original speed sequence; performing filtering processing on the original speed sequence to generate a smoothed speed sequence; performing acceleration calculation processing on the smoothed speed sequence to generate an acceleration sequence; extracting a maximum acceleration value of the acceleration sequence as an explosive force index; and determining the smoothed speed sequence as the instantaneous speed vector.

[0009] Further, the application also provides that the feature extraction processing comprises: performing bone point detection processing on the key action frame to obtain bone point coordinates; performing time sequence alignment processing on the bone point coordinates to generate an aligned bone sequence; performing electromyogram signal integration processing on the aligned bone sequence to generate muscle activation time sequence data; and determining the muscle activation time sequence data as the motion feature data.

[0010] Further, the application further proposes that, when the motion capture data is surface electromyogram, the data validity verification processing includes calculating a ratio of signal power to noise power; the generating physical action evaluation parameter based on the motion capture data includes: performing baseline correction processing and power frequency filtering processing on the surface electromyogram to obtain clean electromyogram; performing feature extraction processing on the clean electromyogram to generate electromyogram feature sequence; performing fatigue calculation processing on the electromyogram feature sequence to obtain muscle fatigue index; and determining the muscle fatigue index as the physical action evaluation parameter.

[0011] Further, the application further proposes that the feature extraction processing further includes: performing lower limb joint detection processing on the key action frame to obtain lower limb bone joint position coordinate set, the lower limb bone joint position coordinate set including hip joint coordinates, knee joint coordinates and ankle joint coordinates; performing cross-frame trajectory alignment processing on the lower limb bone joint position coordinate set to generate lower limb motion trajectory; performing velocity analysis processing on the lower limb motion trajectory: calculating displacement change amount per unit time to generate lower limb instantaneous velocity sequence; performing smoothing processing on the lower limb instantaneous velocity sequence; extracting the ratio of acceleration segment duration to deceleration segment duration; and integrating the explosive strength index, the ratio of acceleration segment duration to deceleration segment duration into motion feature data.

[0012] Further, the application further proposes that, before the acquiring motion capture data, it further includes: receiving a test start instruction containing a geographic position identifier; acquiring current device position coordinates based on a positioning device; calculating the Euclidean distance between the current device position coordinates and a preset training area center coordinate; verifying that the Euclidean distance is less than or equal to a preset area radius threshold; in response to the verification passing, activating a data acquisition device matched with the test type and sending an acquisition parameter configuration instruction.

[0013] Further, the application further proposes that the acquiring motion capture data includes: acquiring three-dimensional acceleration and angular velocity data through an inertial measurement unit; the acquiring motion capture data includes: acquiring three-dimensional coordinate data of infrared reflective marker points through an optical motion capture system; and the acquiring motion capture data includes: generating a two-dimensional motion trajectory based on continuous inter-frame motion vector analysis of a video acquisition device.

[0014] Further, the application further provides a biodynamic training test system, and the system comprises: a data acquisition module configured to acquire motion capture data; a data verification module configured to perform data validity verification processing on the motion capture data to obtain a verification result; and a core processing module configured to generate a physical action evaluation parameter according to the motion capture data in response to the verification result representing that the data is valid.

[0015] Further, the application further provides a biodynamic training test system, and the system comprises: a data acquisition module configured to acquire motion capture data; a data verification module configured to perform data validity verification processing on the motion capture data to obtain a verification result; and a core processing module configured to generate a physical action evaluation parameter according to the motion capture data in response to the verification result representing that the data is valid.

[0016] As can be seen from the above, the biodynamic training test method and system provided by the application have the advantages of improving the accuracy of data validity judgment and avoiding the entry of false data into the subsequent processing link by constructing a progressive verification process through multi-level data validity verification processing and realizing accurate mapping of joint kinematics parameters in combination with a biomechanical model. BRIEF DESCRIPTION OF DRAWINGS

[0017] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which a number of embodiments of the application are illustrated by way of example, in which:

[0018] Figure 1 A flowchart of a biodynamic training test method provided by an embodiment of the application is shown in FIG. 1. DETAILED DESCRIPTION

[0019] The technical solutions in the present application will be described clearly and completely in the present application in combination with the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application. It should be noted that: similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0020] In the conventional existing physical training test system, the data verification mechanism only uses a single level of basic threshold judgment, which cannot effectively identify abnormal interruptions or sensor failure problems in motion capture data, resulting in invalid data segments entering the subsequent calculation link. The multi-modal data collected by different sensors lack a unified space-time reference, resulting in coordinate system deviation and time sequence misplacement when cross-device data fusion, affecting the calculation accuracy of biomechanical parameters. The transformation process from raw data to joint motion parameters relies on simple linear calculation, without establishing a kinematic feature mapping based on biomechanical models, making it difficult to accurately reflect the action specification index.

[0021] For example, when using an optical motion capture system to evaluate deep squat actions, the continuous ten frames of data are lost due to limb occlusion of infrared marker points. The existing verification mechanism only detects whether the total number of frames meets the standard, without verifying the length of the continuous valid data segment, and incorrectly inputs the motion trajectory containing data holes into the calculation module. In the surface electromyography signal acquisition process, poor sensor contact causes a sudden drop in signal power. The traditional method does not calculate the signal noise ratio threshold, resulting in high-frequency interference in the electromyography feature sequence. When multi-modal data fusion, due to the lack of timestamp calibration processing, there is a millisecond-level time sequence deviation between optical data and electromyography signals, causing the correlation analysis of joint angles and muscle activation states to fail.

[0022] If the above problems are not solved, the invalid data segment will cause peak error in joint kinematics parameter calculation, making the motion evaluation parameter deviate from the true biomechanical state. The non-uniformity of cross-modal data space-time reference will cause distortion in multi-source information fusion, reducing the credibility of physical fitness test results. The lack of biomechanical model support in parameter mapping cannot capture the deformation characteristics of knee joint in buckling and other actions, affecting the accuracy of the motion injury warning mechanism. Insufficient data verification granularity will also lead to continuous processing of low-quality data by the system, increasing the consumption of computing resources and reducing the efficiency of real-time feedback.

[0023] To this end, the present application provides a physical fitness training test method, comprising:

[0024] S1, synchronously acquiring motion capture data through a multi-modal data acquisition device, the motion capture data including optical action data or surface electromyogram signals;

[0025] S2, performing data preprocessing on the original motion capture data, the data preprocessing including coordinate system normalization processing and timestamp calibration processing;

[0026] S3, performing multi-level data validity verification processing on the preprocessed motion capture data to obtain a verification result;

[0027] S4, in response to the verification result representing data validity, when the motion capture data is optical action data, mapping the motion capture data to joint kinematics parameters based on a biomechanical model, and generating physical fitness motion evaluation parameters according to the joint kinematics parameters;

[0028] Wherein, the multi-level data validity verification processing includes: primary verification: verifying that the total number of data frames of the motion capture data reaches a preset frame number threshold; intermediate verification: verifying that the time interval between adjacent data frames is less than or equal to a preset time threshold; advanced verification: confirming that the proportion of invalid data points in the motion capture data is lower than a preset proportion threshold, and the length of the continuous valid data segment is greater than a preset continuous frame threshold; the multi-level data validity verification processing uses progressive verification logic, and terminates subsequent verification when the current level verification fails.

[0029] The multi-modal data acquisition device refers to a hardware device capable of simultaneously acquiring different types of motion data, and can be implemented by combining an inertial measurement unit, an optical motion capture system, or a surface electromyogram sensor, for example, to synchronously acquire optical motion data or surface electromyogram signals, and solve the problem of insufficient dimensions of a single data source. The coordinate system normalization processing refers to converting data collected by different sensors into a unified spatial coordinate system, which can be implemented by using a homogeneous coordinate transformation matrix, and eliminates positioning deviations caused by differences in coordinate systems of different devices. The timestamp calibration processing refers to aligning the time series of multi-source data, which can be implemented by using a network time protocol or a hardware synchronization signal trigger, and ensures the time sequence consistency of cross-modal data. The multi-level data validity verification processing refers to a hierarchical data quality detection mechanism that is executed according to a preset logical sequence, which can be implemented by using a progressive process of primary verification to judge the total amount of data, intermediate verification to detect sampling intervals, and advanced verification to evaluate data continuity, to prevent invalid data from entering subsequent calculation links. The biomechanical model refers to a joint kinematics parameter calculation model established based on human anatomical structure, which can be implemented by using an inverse kinematics algorithm or a rigid body dynamics equation, and maps the original motion data to biomechanical parameters such as joint angles and angular velocities.

[0030] The present application combines a progressive multi-level data verification mechanism with a biomechanical model to build a full-link quality control system from raw data acquisition to biomechanical parameter generation. The multi-level verification mechanism filters out invalid data segments layer by layer through the progressive screening of primary, intermediate, and advanced verification, solving the problem of coarse verification granularity in existing technologies. At the same time, the application of the biomechanical model realizes the accurate conversion of raw data to joint kinematics parameters, making up for the evaluation dimension loss defect caused by the dependence of traditional schemes on simple calculation.

[0031] The present application synchronously acquires motion capture data, including optical motion data or surface electromyogram signals, through a multi-modal data acquisition device. The raw motion capture data is preprocessed, including coordinate system normalization processing and timestamp calibration processing. The coordinate system normalization processing converts data collected by different devices into a unified coordinate system, eliminating positioning deviations caused by differences in coordinate systems. The timestamp calibration processing realizes the alignment of cross-device data time sequence, overcoming the multi-source data fusion failure problem.

[0032] A multi-stage data validity verification process is performed on the pre-processed motion capture data. The primary verification ensures that the data acquisition amount meets the analysis requirements by verifying whether the total number of data frames reaches a preset frame number threshold. The intermediate verification checks whether the time interval between adjacent data frames is less than or equal to a preset time threshold, excluding abnormal sampling rate problems. The advanced verification confirms whether the invalid data point proportion is lower than a preset proportion threshold, and whether the length of the continuous valid data segment is greater than a preset continuous frame threshold, effectively identifying invalid data segments caused by sensor disconnection or motion interruption. The multi-stage verification adopts a progressive logic, terminating subsequent verification when the current stage fails, avoiding invalid data entering the calculation link.

[0033] When the verification result represents data validity, and the motion capture data is optical action data, the motion capture data is mapped to joint kinematics parameters based on a biomechanical model. The biomechanical model considers the multi-joint linkage effect, achieving accurate conversion from raw data to biomechanical parameters. The physical action evaluation parameters are generated according to the joint kinematics parameters, integrating kinematic features such as joint angle deviation, which can detect action deformation problems.

[0034] As a preferred embodiment, the scheme of the present application is implemented as follows:

[0035] The multi-modal data acquisition device includes an optical action capture system and a surface electromyography acquisition device. The optical action capture system is composed of multiple high-speed cameras, which obtain three-dimensional motion data by tracking reflective markers attached to key points of the human body. The surface electromyography acquisition device is composed of multiple channel electromyography sensors, which acquire muscle surface potential signals.

[0036] In the data preprocessing stage, coordinate system normalization processing is first performed. The marker point coordinates collected by the optical system are converted from the camera coordinate system to the global coordinate system, eliminating the differences in the viewing angles of different cameras. Spatial filtering is performed on the surface electromyography signals to remove baseline drift caused by changes in skin impedance. Then, timestamp calibration processing is performed to align the sampling times of optical data and electromyography signals according to the system clock, ensuring the time sequence consistency of multi-source data.

[0037] The multi-stage data validity verification process is divided into three stages. The primary verification checks whether the total number of data frames meets the preset minimum frame number requirement, for example, no less than 1000 frames. The intermediate verification calculates the time interval between adjacent frames and verifies whether it is less than or equal to the preset threshold, such as 10 milliseconds. The advanced verification first counts the number of invalid data points in each frame and calculates whether the invalid point proportion is lower than the preset threshold, such as 5%. Then, it detects whether the length of the continuous valid data segment is greater than the preset continuous frame threshold, such as 100 frames. The verification process adopts a progressive logic, terminating subsequent verification and returning a data invalid result when any level fails.

[0038] When the verification passes and the data type is optical action data, parameter mapping is performed based on a biomechanical model. First, a human skeletal chain model is constructed, and joint degrees of freedom and motion constraints are defined. Then, inverse kinematics algorithm is used to convert the three-dimensional coordinates of the marker points into a joint angle sequence. Based on the joint angle sequence, angular velocity and angular acceleration are calculated to obtain a complete set of joint kinematics parameters. Finally, based on the kinematics parameters, physical action evaluation indicators such as joint range of motion and coordination are generated.

[0039] Through the above scheme, the application constructs a multi-level data verification system, effectively identifying abnormal data at different levels. The progressive verification mechanism can distinguish between incidental noise and systematic failure, avoiding invalid data segments from entering the subsequent calculation link. The coordinate system unification and timestamp calibration in the data preprocessing stage ensure the accuracy of multi-modal data fusion, overcoming the problem of non-uniform space-time reference of cross-device data. The parameter mapping based on the biomechanical model breaks through the limitations of traditional linear calculation methods and can accurately reflect the joint kinematics characteristics, providing a reliable basis for detecting action deformation. The overall scheme improves the data quality and biomechanical parameter conversion accuracy of physical training tests, providing more reliable technical support for action evaluation.

[0040] The application further proposes generating physical action evaluation parameters, including performing frame processing on the motion capture data to obtain an action frame sequence; for each action frame in the action frame sequence, verifying that the human motion region bounding box coverage rate is greater than a preset coverage rate threshold; performing bone joint positioning processing to obtain a bone joint position coordinate set; generating motion trajectory data based on the bone joint position coordinate set; performing velocity analysis processing on the motion trajectory data to obtain an instantaneous velocity vector; performing angle deviation calculation processing on the motion trajectory data to obtain a joint angle deviation value; generating a dynamics indicator according to the instantaneous velocity vector and the joint angle deviation value; filtering action frames with joint angle deviation values less than a preset angle threshold as key action frames; performing feature extraction processing on the key action frames to obtain motion feature data; and generating physical action evaluation parameters based on the motion feature data.

[0041] In the formula, the frame processing divides the continuous motion data into a sequence of discrete motion frames with time correlation, and the preset coverage threshold can be set to be that the area ratio of the bounding box covering the human body region is not less than 85%, and the effective pixel ratio in the bounding box is calculated by an image recognition algorithm to realize verification. The bone joint positioning processing adopts a deep neural network model to output a three-dimensional coordinate set containing key points such as hips, knees and ankles, and the coordinate accuracy is controlled within ±2 mm. The motion trajectory data generation calculates the Euclidean distance change of joint coordinates between adjacent frames to form a time-space displacement curve. The velocity analysis processing calculates the instantaneous velocity by using the central difference method, for example, the displacement change rate is calculated with a time window of 10 ms, and the high-frequency noise is removed by a Butterworth low-pass filter. The angle deviation calculation processing compares the actual joint angle with the ideal angle in the standard motion database to generate a deviation value by using a cosine similarity algorithm, and the preset angle threshold can be set to 5 degrees. The key motion frame screening retains the frame data with a deviation value below the threshold by traversing the joint angle deviation values of all motion frames, for example, abnormal frames with excessive knee inward bending angle are removed in the deep squat motion. The feature extraction processing performs principal component analysis on the retained key frames to extract feature vectors such as motion trajectory curvature, velocity change rate and muscle activation timing.

[0042] When the motion capture data passes through multi-level validity verification, the frame processing converts the continuous data stream into a sequence of discrete motion frames, ensuring that the subsequent processing has a clear time reference. Each motion frame first verifies its human body region coverage integrity, excluding invalid frames caused by target deviation. After establishing a three-dimensional space reference through high-precision bone joint positioning, the motion trajectory data generation module captures the continuous displacement change of the joint nodes to form spatial motion features. The velocity analysis module and the angle deviation calculation module process the motion trajectory data in parallel, the former quantifies the instantaneous dynamic characteristics of the motion, and the latter detects kinematic abnormalities of the joint. The kinetic indicators are generated by fusing the speed vector module length and the angle deviation value, for example, the explosive force indicator is calculated as the product of the speed peak value and the joint stability coefficient. The key motion frame screening mechanism dynamically adjusts the effective data range based on the angle deviation threshold to eliminate noise interference caused by motion deformation. The final feature extraction module integrates kinematic and kinetic multi-dimensional features to generate comprehensive evaluation parameters including motion specification, explosive force intensity and stability. The scheme effectively improves the comprehensiveness and accuracy of the evaluation parameters through space-time feature fusion and data cleaning mechanism.

[0043] As a preferred embodiment, the scheme of the application is implemented as follows:

[0044] Frame processing is performed on the motion capture data to obtain a sequence of motion frames. For each motion frame in the sequence of motion frames, first, verify whether the bounding box coverage rate of the human motion region is greater than the preset coverage threshold.

[0045] Specifically, a deep learning model can be used to identify human key points and extract three-dimensional coordinates of joints including the head, shoulders, elbows, wrists, hips, knees, and ankles.

[0046] Motion trajectory data is generated based on the set of joint position coordinates. By connecting corresponding joints of adjacent frames, a continuous motion trajectory curve is formed.

[0047] Velocity analysis processing is performed on the motion trajectory data to obtain instantaneous velocity vectors. The displacement change between adjacent frames of joint positions is divided by the time interval to obtain the instantaneous velocity of each joint.

[0048] Angle deviation calculation processing is performed on the motion trajectory data to obtain joint angle deviation values. Based on the standard motion posture, the difference between the actual motion angle and the standard angle is calculated.

[0049] Kinetic indicators are generated based on the instantaneous velocity vectors and joint angle deviation values. For example, the peak value of the instantaneous velocity can be combined with the average value of the angle deviation to form a comprehensive scoring indicator.

[0050] Key motion frames with joint angle deviation values less than a preset angle threshold are selected as key motion frames. The angle threshold can be set to 10 degrees to retain key frames with more standard postures.

[0051] Feature extraction processing is performed on the key motion frames to obtain motion feature data. Time series features such as joint positions, velocities, and accelerations in the key frames are extracted.

[0052] Physical action evaluation parameters are generated based on the motion feature data. Machine learning algorithms can be used to train evaluation models based on historical data to output quantitative indicators such as motion standardization, explosive force, and stability.

[0053] Through the above technical solutions, the present application realizes multi-dimensional action feature extraction and evaluation. By introducing instantaneous velocity vector analysis, the change characteristics of explosive force are captured. Combined with joint angle deviation calculation, the standardization of the action is fully reflected. By selecting key motion frames, the interference of invalid data on the evaluation results is effectively reduced. The final physical action evaluation parameters comprehensively consider the standardization, explosive force, and stability of the action, improving the accuracy and comprehensiveness of the evaluation.

[0054] The present application further proposes a velocity analysis processing including: performing time differentiation processing on the motion trajectory data to generate an original velocity sequence; performing filtering processing on the original velocity sequence to generate a smoothed velocity sequence; performing acceleration calculation processing on the smoothed velocity sequence to generate an acceleration sequence; extracting the maximum acceleration value of the acceleration sequence as an explosive force indicator; and determining the smoothed velocity sequence as the instantaneous velocity vector.

[0055] The time differential processing generates a raw speed sequence by calculating the ratio of displacement difference between adjacent time points and time interval, for example, using central difference method or backward difference method, and the time interval can be set to 0.01-0.05 seconds. The filtering processing can use a low-pass filter or a moving average filter, and the cut-off frequency is preferably 10-15 Hz, for suppressing high-frequency noise and retaining effective motion characteristics. The acceleration calculation processing is implemented by performing time differential on the smoothed speed sequence again, for example, using a second-order difference algorithm, and the time differential interval is consistent with the speed sequence generation interval. The extraction of the maximum acceleration value is completed by traversing the peak value of the acceleration sequence, for example, using a sliding window method to identify local maximum values, and the window length is set to 5-10 data points. The smoothed speed sequence is consistent with the raw speed sequence in data length as an instantaneous speed vector, ensuring the timing correspondence.

[0056] Specifically, the motion trajectory data generates a raw speed sequence after time differential processing, and at this time, the data may contain high-frequency components caused by sensor noise or limb jitter. Through low-pass filtering processing, noise components higher than the preset cut-off frequency are filtered out, for example, a Butterworth filter is used to process speed data with a sampling rate of 100 Hz at a cut-off frequency of 12 Hz, to eliminate abnormal fluctuations. The smoothed speed sequence generates an acceleration sequence after second-order differentiation, and since the original speed fluctuations have been suppressed, no mutation error will be introduced in the acceleration calculation process. The extraction of the maximum acceleration value focuses on the peak phase of the power output in the motion process, for example, capturing the acceleration extreme value at the moment of take-off in the vertical jump action. Finally, the filtered speed sequence is output as an instantaneous speed vector, which not only retains the dynamic characteristics of the motion trajectory, but also provides a stable input for subsequent joint angle deviation calculation. The processing flow cooperates with the front data validity verification to ensure that only continuous valid data segments are analyzed, avoiding the filtering boundary effect caused by data interruption, thereby systematically improving the accuracy and calculation stability of the explosive strength index.

[0057] As a preferred embodiment, time differential processing is performed on the motion trajectory data to generate a raw speed sequence. Specifically, the central difference method can be used to calculate the displacement change rate between adjacent time points to obtain a speed vector sequence containing x, y, and z direction components.

[0058] Further, filtering processing is performed on the raw speed sequence to generate a smoothed speed sequence. For example, a Butterworth low-pass filter can be applied, with a cut-off frequency set to 10 Hz, to eliminate high-frequency noise and retain the main characteristics of the motion.

[0059] Thus, acceleration calculation processing is performed on the smoothed speed sequence to generate an acceleration sequence. In specific implementation, a second-order central difference method can be used to calculate the change rate of speed at adjacent time points to obtain an acceleration sequence.

[0060] wherein the maximum acceleration value of the extracted acceleration sequence is taken as the explosive force indicator. This can be achieved by traversing the acceleration sequence, comparing and recording the maximum value.

[0061] Finally, the smoothed speed sequence is determined as the instantaneous speed vector. This step directly uses the filtered speed data without additional calculation.

[0062] Through the above technical solutions, the application can effectively eliminate high-frequency noise and abnormal fluctuations in the original speed sequence, improve the accuracy of the instantaneous speed vector. At the same time, the acceleration is calculated based on the smoothed speed sequence, which reduces the acceleration mutation error caused by data noise, thereby improving the reliability of the explosive force indicator. In addition, through multi-stage data processing, the scheme not only retains the timing characteristics of the motion trajectory, but also improves the stability of the subsequent dynamics indicator calculation, providing a more accurate and reliable data basis for physical training evaluation.

[0063] The application further proposes a feature extraction process, which includes: performing a skeleton point detection process on the key action frame to obtain skeleton point coordinates; performing a timing alignment process on the skeleton point coordinates to generate an aligned skeleton sequence; performing an electromyographic signal integration process on the aligned skeleton sequence to generate muscle activation timing data; and determining the muscle activation timing data as the motion feature data.

[0064] wherein the skeleton point detection process can use a pose estimation algorithm based on a convolutional neural network to locate the two-dimensional or three-dimensional coordinates of the hip joint and the knee joint by regressing the key point heat map. The timing alignment process eliminates the inter-frame time offset of the skeleton point coordinates through a dynamic time warping algorithm, for example, limiting the change in the displacement of the skeleton points between adjacent frames within a preset time window length. The electromyographic signal integration process takes the aligned skeleton sequence as the time reference and performs sliding window integration calculation on the synchronously collected electromyographic signals, with the integration period set to 50 milliseconds to 200 milliseconds, quantifying the activation intensity of the rectus femoris muscle or the vastus lateralis muscle. The time references of the skeleton point coordinates and the electromyographic signals achieve data synchronization through a unified timestamp, ensuring that the muscle activation state and the timing of the skeletal motion trajectory match during the integration process.

[0065] The skeleton point detection processing extracts the coordinate data of the joints of the lower limbs of the human body from the key action frames, to provide a spatial positioning basis for subsequent analysis. The time sequence alignment processing corrects the time sequence deviation of the skeleton point coordinates by interpolation compensation or a kinematics constraint model, for example, using cubic spline interpolation to fill in the coordinate data of the missing frames, so that the cross-frame motion trajectory remains continuous. The electromyographic signal integration processing combines the skeleton motion trajectory with the muscle electrophysiological signal, calculates the area integral value of the electromyographic signal within a preset time window, and generates a time sequence curve reflecting the muscle activation degree. The muscle activation time sequence data simultaneously contains the spatial features of the skeleton motion trajectory and the time domain features of muscle synergistic activation, for example, in the squatting action, the knee joint angle change and the quadriceps femoris muscle activation intensity show a positive correlation in time sequence. The physical action evaluation parameters generated thereby can accurately quantify the action standardization, detect the deformation action such as knee joint inward buckling, and improve the biomechanical reliability of the evaluation results.

[0066] Preferably, the feature extraction processing includes the following steps: first, performing skeleton point detection processing on the key action frames to obtain skeleton point coordinates. Specifically, a deep learning model can be used to locate the skeleton points of the key action frames, and two-dimensional or three-dimensional coordinate information of the main joint points of the human body can be extracted. Second, performing time sequence alignment processing on the skeleton point coordinates to generate an aligned skeleton sequence. For example, a dynamic time warping algorithm can be used to align the skeleton point coordinates at different time points, eliminating the time sequence inconsistency problem caused by differences in sampling frequency or action speed. Third, performing electromyographic signal integration processing on the aligned skeleton sequence to generate muscle activation time sequence data. Specifically, the surface electromyographic signals collected synchronously can be time-matched with the aligned skeleton sequence, and then rectified and smoothed, and the integral value of the envelope line is calculated as a quantitative indicator of muscle activation degree. Finally, the muscle activation time sequence data is determined as the motion feature data. In this way, the obtained motion feature data contains not only the spatial information of the skeleton motion, but also the time sequence features of muscle contraction.

[0067] Through the above technical solutions, the present application solves the problems of time sequence consistency and muscle activation state quantification in motion feature data extraction. Through skeleton point detection, time sequence alignment and electromyographic signal integration processing, dynamic correlation analysis of the skeleton motion trajectory and muscle activation state is realized. This multi-dimensional biomechanical feature extraction method improves the accuracy and comprehensiveness of the physical action evaluation parameters, and provides a more reliable data basis for subsequent action quality evaluation and training guidance.

[0068] The application further proposes that when the motion capture data is surface electromyography signals, the data validity verification processing includes calculating the ratio of signal power to noise power; generating the physical action evaluation parameter from the motion capture data includes performing baseline correction processing and power frequency filtering processing on the surface electromyography signals to obtain clean electromyography signals, performing feature extraction processing on the clean electromyography signals to generate an electromyography feature sequence, performing fatigue calculation processing on the electromyography feature sequence to obtain a muscle fatigue index, and determining the muscle fatigue index as the physical action evaluation parameter.

[0069] In the data validity verification stage, the ratio of signal power to noise power can be calculated using a sliding window method, for example, a local signal-to-noise ratio is calculated with a window length of 200 milliseconds, and when the ratio is lower than a preset threshold, it is determined that the signal quality is not up to standard. The baseline correction processing eliminates device zero drift through a moving average filter, for example, a high-pass filter with a time constant of 0.5 seconds is used. The power frequency filtering processing can be configured as a notch filter of 50Hz or 60Hz, and the specific selection is based on the local power system frequency. The feature extraction processing can use the root mean square value calculation method, for example, the root mean square value sequence of the electromyography signal is calculated within a 500 millisecond time window. The fatigue calculation processing can establish a model based on the median frequency shift, for example, when the amplitude of the median frequency drop exceeds 15% of the reference value, it is determined that the muscle enters a state of fatigue.

[0070] Specifically, the surface electromyography signals first pass through the signal-to-noise ratio verification, and when the ratio of signal power to noise power reaches the preset threshold, it enters the data processing flow. The baseline correction processing eliminates the direct current offset component in the signal to ensure that the electromyography signal fluctuates around the zero baseline. The power frequency filtering processing performs frequency domain filtering on the power system interference, and retains the effective components of the electromyography signal in the frequency range of 20-450Hz. The clean electromyography signal generates a feature sequence representing muscle activation intensity after time domain feature extraction, and after inputting the sequence into the fatigue calculation model, a muscle fatigue index is generated by analyzing the amplitude attenuation slope or frequency spectrum shift. The index can quantitatively reflect the dynamic changes of muscle contraction ability during exercise, for example, when the fatigue index reaches 0.7, it indicates that the muscle enters a compensatory contraction stage. The whole process ensures the quality of the signal, and realizes the accurate evaluation of the muscle function state through multi-stage signal processing and feature modeling.

[0071] As a preferred embodiment, when the motion capture data is surface electromyography signals, first calculate the ratio of signal power to noise power to perform data validity verification processing. Specifically, the short-time Fourier transform method can be used for time-frequency analysis of the original electromyography signal, and the power spectral density of the signal main frequency band and the background noise frequency band is extracted, and the ratio of the two is calculated as the signal-to-noise ratio index. For example, the frequency range of 20-450Hz can be taken as the signal main frequency band, and the frequency ranges of 0-20Hz and 450-500Hz can be taken as the background noise frequency band. When the signal-to-noise ratio index is higher than the preset threshold, it is determined that the data is valid.

[0072] For verifying the valid surface electromyography signal, further baseline correction processing and power frequency filtering processing are performed to obtain a clean electromyography signal. The baseline correction processing can remove low-frequency drift by using a high-pass filter, and the cutoff frequency can be set to 10 Hz. The power frequency filtering processing uses a notch filter to eliminate 50 Hz / 60 Hz power frequency interference.

[0073] The feature extraction processing is performed on the clean electromyography signal to generate an electromyography feature sequence. The sliding window method can be used, the window length is 256 ms, the overlap rate is 50%, and the time domain features (such as root mean square value, average absolute value) and frequency domain features (such as median frequency, average power frequency) are extracted in each window.

[0074] The fatigue calculation processing is performed on the electromyography feature sequence to obtain a muscle fatigue index. The fatigue evaluation model can be established based on the change trend of the median frequency, for example, the linear regression method is used to fit the slope of the change of the median frequency with time, and the absolute value of the slope is taken as the muscle fatigue index.

[0075] Finally, the calculated muscle fatigue index is determined as the physical action evaluation parameter, which is used for subsequent sports performance analysis and training guidance.

[0076] Through the above technical solutions, the application realizes a specialized processing flow for surface electromyography signals. The signal quality degradation problem is effectively identified by signal-to-noise ratio calculation, baseline correction and power frequency filtering ensure that high-quality electromyography signals are obtained. Feature extraction and fatigue calculation make full use of the time-frequency characteristics of electromyography signals to generate evaluation parameters that can objectively reflect the muscle function state. This multi-dimensional signal processing and feature analysis method significantly improves the accuracy and reliability of physical action evaluation based on electromyography signals, and provides more detailed and personalized guidance for sports training.

[0077] The application further proposes that the key action frame is subjected to lower limb joint detection processing to obtain a lower limb bone joint position coordinate set, the lower limb bone joint position coordinate set includes hip joint coordinates, knee joint coordinates and ankle joint coordinates; the cross-frame trajectory alignment processing is performed on the lower limb bone joint position coordinate set to generate a lower limb motion trajectory; the speed analysis processing is performed on the lower limb motion trajectory: the displacement change amount per unit time is calculated to generate a lower limb instantaneous speed sequence; the smoothing processing is performed on the lower limb instantaneous speed sequence; the ratio of the acceleration segment duration to the deceleration segment duration is extracted; the explosive strength index, the ratio of the acceleration segment duration to the deceleration segment duration are integrated into motion feature data.

[0078] The lower limb joint detection processing can adopt a joint point regression model based on deep learning, such as identifying the two-dimensional or three-dimensional coordinates of the hip, knee, and ankle joints from key action frames through a convolutional neural network. The cross-frame trajectory alignment processing can adopt a time series interpolation algorithm to perform linear or nonlinear interpolation on the coordinate data between adjacent frames, eliminating trajectory breaks caused by sampling rate fluctuations. In the speed analysis processing, the unit time can be set to 10 milliseconds to 50 milliseconds, and the displacement change is obtained by calculating the Euclidean distance difference between the coordinates of the adjacent two frames. The smoothing processing can use a sliding average filter with a window length of 3 to 5 data points. The ratio of the duration of the acceleration segment to the duration of the deceleration segment can be segmented by the sign change point of the first derivative of the speed curve, for example, when the acceleration changes from positive to negative, it is determined as the starting point of the deceleration segment.

[0079] In the action frame sequence obtained by frame processing, frames with joint angle deviation values less than a preset angle threshold are selected as key action frames. The coordinates of the hip, knee, and ankle joints are extracted through lower limb joint detection processing to form a coordinate set containing spatial position information. The cross-frame trajectory alignment processing connects the discrete joint coordinates in time sequence to generate a continuous lower limb motion trajectory curve. In the speed analysis processing, the displacement difference between adjacent frames and the time interval are calculated to generate an original instantaneous speed sequence, and then Gaussian filtering is performed to eliminate speed mutations caused by sensor noise. The smoothed speed sequence determines the demarcation points of the acceleration and deceleration stages by finding local maximum points. For example, in a vertical vertical jump action, the ratio of the duration of the take-off acceleration segment to the duration of the landing buffer deceleration segment can be quantified as 1.5:1 to 2.5:1. Finally, the explosive force index and the ratio of the durations of the acceleration and deceleration stages are input into the evaluation model to generate a composite feature vector reflecting the lower limb explosive force and motion coordination. By quantifying the ratio of the durations of the acceleration and deceleration stages, the asymmetry in action control can be identified, such as the difference in the synergistic working state of the hip joint extensor group during take-off and the knee joint flexor group during landing in basketball, thereby improving the biomechanical representation ability of the motion feature data.

[0080] As a preferred embodiment, the feature extraction processing further includes performing lower limb joint detection processing on the key action frames. This processing can be implemented through a deep learning model, such as using a convolutional neural network to perform semantic segmentation on images to identify the positions of the hip joint, knee joint, and ankle joint. The obtained lower limb bone joint position coordinate set contains the three-dimensional spatial coordinates of these three joints.

[0081] The cross-frame trajectory alignment processing is performed on the lower limb bone joint position coordinate set. This step can use a Kalman filter algorithm to predict the possible position of each joint in the next frame and fuse it with the actual detection result to generate a smooth and continuous lower limb motion trajectory.

[0082] Perform speed analysis processing on the generated lower limb motion trajectory. First, calculate the displacement change in unit time to generate the lower limb instantaneous speed sequence. 10 milliseconds can be selected as the unit time interval, and the instantaneous speed is obtained by dividing the position difference between the adjacent two time points by the time interval.

[0083] Then perform smoothing processing on the lower limb instantaneous speed sequence. A sliding average filter can be used, and a window size of 5 data points is selected. The average value of the speed in the window is calculated as the smoothed speed at the current time.

[0084] Extracting the ratio of the duration of the acceleration segment and the duration of the deceleration segment can be achieved by judging the change trend of the smoothed speed sequence. The stage of speed rising is defined as the acceleration segment, and the stage of speed falling is defined as the deceleration segment. The duration of each stage is accumulated respectively, and the ratio is calculated.

[0085] Integrate the explosive force index, the ratio of the duration of the acceleration segment and the duration of the deceleration segment into the motion feature data. The peak value of the acceleration can be selected as the explosive force index, and together with the aforementioned ratio, a two-dimensional feature vector is formed as the final motion feature data.

[0086] Through the above technical solution, the application realizes the fine analysis of the lower limb joint motion. Therefore, the dynamic change law of the lower limb joint in continuous action can be effectively captured, and the evaluation accuracy of the explosive force index and the action coordination is improved.

[0087] The application further proposes that before performing the acquisition of the motion capture data, a test start instruction containing a geographic position identifier is received; the current device position coordinates are acquired based on a positioning device; the Euclidean distance between the current device position coordinates and the preset training area center coordinates is calculated; it is verified that the Euclidean distance is less than or equal to the preset area radius threshold; in response to the verification passing, the data acquisition device matched with the test type is activated and the acquisition parameter configuration instruction is sent.

[0088] The geographic location identifier of the test start instruction can be implemented by using a geofence code or a site number to associate the physical environment parameters of the preset training area. The positioning device can be selected from a GPS module or an indoor UWB positioning system, and the positioning accuracy range is controlled within 0.1-1.5 meters to meet the size requirements of different sites. The center coordinates of the preset training area are generated by site surveying data or standard sports site size calculation, such as the center point of a track or the coordinates of a basketball court circle. The Euclidean distance calculation adopts spatial geometric operation in a three-dimensional coordinate system, specifically including longitude, latitude and altitude data in three dimensions. The preset area radius threshold is dynamically adjusted according to the type of movement, for example, 3 meters for sprint test and 1 meter for weightlifting test. The activation logic of the data acquisition device is based on matching sensor combinations according to the type of test, for example, activating the accelerometer and optical marker point collector for running test, and activating the surface electromyography sensor and pressure sensor for weightlifting test. The acquisition parameter configuration instruction includes sampling rate adjustment, filter cutoff frequency setting and sensor range switching, for example, setting the sampling rate of the optical capture system to 120Hz in an indoor site, and adjusting it to 60Hz in an outdoor site to reduce environmental light interference.

[0089] When the test start instruction is received, the geographic location identifier embedded therein is first parsed to determine the target training area. The three-dimensional coordinate data of the device is obtained in real time by the positioning device, and the spatial distance is calculated with the preset center coordinates. The distance value is used to determine whether the device is within the range of the standard training environment, for example, when the calculated distance value is 2.5 meters and the preset threshold is 3 meters, it is determined that the device position meets the requirements. After verification, the system selects the corresponding data acquisition device combination according to the type of test, for example, selects the lower limb joint marker point capture device for deep squat test, and selects the upper limb electromyography sensor for throwing test. At the same time, the acquisition parameter configuration instruction is sent to each activated device, for example, setting the accelerometer range to ±16g on a plastic track, and adjusting it to ±8g on a grass field to adapt to different ground reaction forces. Through this spatial position constraint mechanism, it is ensured that the data acquisition environment is consistent with the physical conditions preset by the biomechanical model, and the parameter deviation caused by the difference between the sites, such as the ground friction and optical reflectivity, is eliminated, thereby improving the accuracy of the subsequent joint kinematics parameter calculation.

[0090] As a preferred embodiment, before performing the motion capture data acquisition, the system first receives a test start instruction containing a geographic location identifier. The instruction can be sent by an operator through a mobile terminal or automatically triggered by a preset training plan. The instruction contains pre-defined training area information, such as "indoor track and field" or "strength training area".

[0091] Upon receiving the start instruction, the system obtains the current device's position coordinates based on the built-in positioning device. The positioning device can be a GPS module, a Bluetooth beacon, or an indoor positioning system. The obtained coordinate information usually includes longitude, latitude, and altitude.

[0092] The system then calculates the Euclidean distance between the current device's position coordinates and the preset training area's center coordinates. The center coordinates of the preset training area are fixed parameters that are previously entered into the system. The calculation of the Euclidean distance uses the three-dimensional space distance formula.

[0093] The calculated Euclidean distance is compared with the preset area radius threshold. The area radius threshold can be set according to the actual size of different training areas, for example, an indoor training area can be set to 10 meters, and an outdoor training field can be set to 50 meters.

[0094] If the verification is passed, that is, the Euclidean distance is less than or equal to the preset area radius threshold, the system activates the data acquisition device matching the current test type. For example, for a running test, an inertial measurement unit and an optical motion capture system can be activated, while for a weightlifting test, a surface electromyography acquisition device can be additionally activated.

[0095] The system further sends a collection parameter configuration instruction to the activated data acquisition device. The configuration instruction can include sampling rate setting, filter parameter adjustment, dynamic range calibration, etc., to adapt to the characteristics of the current training environment.

[0096] After completing the above steps, the system begins to perform the actual motion capture data acquisition process.

[0097] Through the above technical solution, the present application realizes accurate control of the data acquisition environment. Thus, it ensures that the acquisition device is within the preset standard training area, effectively avoiding the problem of inconsistent data caused by environmental differences. The system excludes data distortion that can be caused by non-specified sites or device displacement deviations through a geographic location verification mechanism. This method improves the adaptability of the collected data to the preset biomechanical model, laying a reliable foundation for subsequent joint kinematics parameter mapping and evaluation parameter calculation. Further, by dynamically adjusting the data acquisition parameters according to the position information, the system can adapt to the characteristics of different training environments, improving the quality and consistency of the original data. This environment adaptive mechanism enhances the applicability and data reliability of the system in diversified training scenarios.

[0098] The present application further proposes that the motion capture data acquisition includes: acquiring three-dimensional acceleration and angular velocity data through an inertial measurement unit; acquiring three-dimensional coordinate data of infrared reflective marker points through an optical motion capture system; and generating a two-dimensional motion trajectory based on continuous frame-to-frame motion vector analysis of a video acquisition device.

[0099] The inertial measurement unit can be integrated into a wearable device, and its sampling frequency can be set to a range of 50 Hz to 200 Hz, for example, 100 Hz, for capturing the instantaneous dynamic characteristics of limb movement. An infrared camera array of the optical motion capture system can be arranged around the test area to collect marker point coordinates at intervals of 20 ms to 50 ms, for example, 30 ms, ensuring that the spatial positioning error is less than 1 mm. The frame rate of the video capture device can be configured to be 30 fps to 120 fps, for example, 60 fps, and the motion vector field is calculated by a pixel block matching algorithm between adjacent two frames to generate a two-dimensional plane motion trajectory.

[0100] Specifically, during data acquisition, the three-axis accelerometer and gyroscope of the inertial measurement unit record the linear acceleration and angular velocity of the limb in real time to form time series data of dynamic parameters. At the same time, the infrared reflective marker points arranged at the key joints of the tester are captured by the optical camera, and the three-dimensional space coordinates of each marker point are generated by a triangular positioning algorithm as a reference for kinematic parameters. When the marker points are blocked or the inertial sensor signal is lost, the video capture device generates a two-dimensional plane motion trajectory as a supplementary data source by analyzing the pixel displacement of the human body contour in the continuous video frames. The three data acquisition methods are synchronized in the time dimension through a unified clock source and aligned in the spatial dimension through a preset coordinate system conversion matrix. For example, when the tester performs a jumping action, the inertial data captures the acceleration peak at the moment of take-off, the optical data records the body posture change during the emptying phase, and the video data provides continuous recording of the foot landing trajectory during the landing phase when the sensor is vibrating. Through the spatio-temporal fusion of multi-source data, a multi-dimensional motion capture system covering dynamics, kinematics, and plane trajectory is formed.

[0101] When the athlete performs a jumping action test, the inertial measurement unit is fixed to the waist and proximal extremities of the tester to collect three-dimensional acceleration and angular velocity data at a sampling frequency of 200 Hz. Six infrared cameras of the optical motion capture system are arranged in a ring array around the test area to track the infrared reflective marker points attached to the main joints of the tester at a frequency of 120 Hz, generating a three-dimensional coordinate data set including the hip joint and the knee joint. The video capture device captures the tester's side motion image at a frame rate of 60 frames per second, calculates the motion vector between the continuous five frames of images through an optical flow algorithm, and generates a two-dimensional plane motion trajectory of the foot landing phase. The three-dimensional acceleration data is converted into limb rotation angle through quaternion solution, and is spatially aligned with the three-dimensional coordinates of the knee joint obtained by the optical system. The two-dimensional trajectory generated by the video is mapped to the three-dimensional coordinate system to form a time-stamped multi-modal motion capture data stream.

[0102] By the technical solution, the full-dimension coverage of motion capture data is realized, the inertial measurement unit supplements the high-speed dynamic parameter acquisition capability, the optical system guarantees the joint space positioning accuracy, and the video analysis maintains the continuity of the motion trajectory when the marker point is blocked.

[0103] The application further proposes to establish a historical evaluation database that stores athlete identification, test time, and physical action evaluation parameters; compare the currently generated physical action evaluation parameters with the historical data set of the athlete identification; calculate the time series change rate of the joint angle deviation value, and generate an action deformation warning when the change rate exceeds the preset change threshold; perform linear regression modeling on the explosive force index, output the regression coefficient sign change detection result; calculate the recommended recovery period according to the exponential decay curve of the muscle fatigue index; and generate a biomechanics report containing the joint angle deviation value change rate, the regression coefficient sign change detection result, and the recommended recovery period.

[0104] The historical evaluation database realizes data tracing across the time dimension through the association storage mechanism of athlete identification and test time. For example, the athlete identification can use a combination field containing a training number and an identity code, and the test time accuracy can be set to millisecond level to support high-density data analysis. The time series change rate of the joint angle deviation value is calculated by a sliding window algorithm, the window length can be set to 5-10 test periods, and the preset change threshold is dynamically adjusted according to the type of movement, for example, the threshold of weightlifting action is set to 0.8 degrees / week, and the threshold of sprint action is set to 1.2 degrees / week. The linear regression modeling uses the least squares method to fit the trend line of the explosive force index over time, and the regression coefficient sign change detection is realized by comparing the slope direction of adjacent time periods, for example, when the regression coefficient of three consecutive periods changes from positive to negative, an abnormal fluctuation mark is triggered. The exponential decay curve of the muscle fatigue index is generated by a nonlinear fitting algorithm, the decay time constant can be calculated based on the integral value of the electromyographic signal, and the recommended recovery period is set to 2-3 times the time constant, for example, when the time constant is 48 hours, the recommended recovery period is 96-144 hours.

[0105] In the data storage stage, the current test data is associated with the historical data set through the athlete identification, for example, through a hash table to achieve fast retrieval. In the trend analysis stage, the time series change rate of the joint angle deviation value is obtained by calculating the ratio of the deviation difference value of adjacent test periods to the time interval thereof, and when the ratio exceeds a preset threshold, it indicates that the action deformation presents an accelerating trend, triggering a warning signal. In the explosive force analysis stage, a linear regression model is used to fit the historical explosive force indicators, and if the regression coefficient sign changes from positive to negative, it indicates that the explosive force has a sustained decline, and a sign change detection result is output. In the recovery period calculation stage, based on the decay characteristics of the muscle fatigue indicator, the decay rate is determined through curve fitting, and combined with the exercise load intensity, personalized recovery suggestions are generated. Finally, by integrating the joint angle change rate, the regression coefficient sign change, and the recovery period data, a biomechanics report containing quantitative indicators and decision suggestions is generated, for example, the report can include trend charts, abnormal markers, and recovery schedules. Through the above technical solutions, a full-process closed loop from data storage, longitudinal comparison to trend warning is realized, solving the problems of traditional methods that cannot identify chronic action deformation and lack of data-driven recovery strategies.

[0106] Further, a database server is deployed at the training base, a data table structure containing athlete ID, test date, and physical ability evaluation parameters is created, and the data table is set with timestamp index and athlete ID primary key. When a new physical action evaluation parameter is added, historical data retrieval is performed through the database connection interface, and the historical joint angle deviation values of the same ID are extracted to form a time series array. The moving average algorithm is used to calculate the slope change rate of the last three test data, and when the absolute value of the slope exceeds 0.15 rad / week, a warning signal is triggered. The least squares method is used to fit the explosive force indicator data set, and if the regression coefficient changes from positive to negative and lasts for two test periods, it is marked as an abnormal fluctuation event. The half-life of the muscle fatigue indicator is calculated, and the exponential function is used to fit the decay curve, and when the goodness of fit is greater than 0.85, the half-life multiplied by a safety factor of 1.2 is taken as the recommended recovery period. Finally, the warning signal, abnormal fluctuation event, and recovery period value are integrated into a PDF format report, which is automatically pushed to the coach terminal through the email server.

[0107] The present application further proposes a physical training test system, which comprises a data acquisition module, a data verification module and a core processing module. The data acquisition module includes an inertial measurement unit interface, an optical capture parser and a video analysis engine.

[0108] The inertial measurement unit interface is configured to receive three-dimensional motion data, which can integrate SPI or I2C communication protocol converters to convert raw byte streams output by devices from different manufacturers into a standardized quaternion format. The optical capture parser has a built-in homogeneous coordinate transformation algorithm that maps infrared marker point coordinates from the optical system coordinate system to the biomechanical model coordinate system through rotation matrix and translation vector calculation. The video analysis engine uses the optical flow method to calculate inter-frame motion vectors, and eliminates lens jitter interference through feature point matching and motion compensation technology. The data verification module is configured to perform multi-level verification logic, the primary verification stage detects whether the total number of data frames reaches the preset threshold, the intermediate verification stage checks the continuity of the timestamp, and the advanced verification stage identifies the distribution pattern of invalid data points. The core processing module is configured to input multi-source motion data into the biomechanical model after data validity is confirmed, to generate physical action evaluation parameters including joint angle deviation values and instantaneous speed vectors.

[0109] The inertial measurement unit interface converts raw binary streams into standard data formats through protocol conversion mechanisms when receiving wearable sensor data. For example, when the sensor transmits data at a baud rate of 1200bps, the buffer queue built-in the interface is configured to reorganize data packets by 16 bytes in length. The optical capture parser converts raw three-dimensional coordinates to the biomechanical model coordinate system using a spatial transformation algorithm when processing infrared marker point coordinates. This conversion process is achieved through a 4x4 homogeneous transformation matrix, and the matrix parameters are dynamically adjusted according to the relative position relationship between the optical system and the biomechanical model. The video analysis engine uses the Lucas-Kanade algorithm to track feature point displacement when calculating inter-frame motion vectors, generating two-dimensional motion trajectory data with a video stream input at 30fps. The data verification module performs progressive verification logic, terminating the processing flow when the total number of data frames does not reach 100 frames, marking abnormal data segments when the timestamp interval exceeds 33ms, and triggering data reacquisition mechanisms when the invalid data point ratio exceeds 5%. The core processing module calculates joint kinematics parameters in a unified space-time coordinate system by fusing inertial data, optical coordinates, and video trajectories, eliminating timing misalignment problems caused by sampling rate differences in multi-source data, and finally generating evaluation indicators including three-dimensional joint angles and linear velocities.

[0110] Preferably, the data acquisition module receives three-dimensional motion data transmitted by the wearable sensor through the inertial measurement unit interface, wherein the data is standardized and packaged in JSON format protocol, and the original signals of the accelerometer and the gyroscope are converted into physical quantity data in the International System of Units. The optical capture parser performs homogeneous coordinate transformation on the marker point coordinates captured by the infrared camera, maps the original coordinate system to the center coordinate system of the pelvis biomechanical model through a rotation and translation matrix, and eliminates the joint positioning deviation caused by the difference in camera viewing angle. The video analysis engine calculates the motion vector of the pixel points in the consecutive video frames through the Horn-Schunck optical flow algorithm, and generates two-dimensional plane motion trajectory data. The data verification module performs timestamp synchronization processing on the received three modal data, and adjusts the sampling frequency difference to 200Hz uniform frequency by using the linear interpolation method. The core processing module generates physical action evaluation parameters such as knee flexion angle and hip rotation speed by fusing the multi-source data through the inverse kinematics algorithm.

[0111] Through the above technical solutions, the application effectively eliminates the data format difference and coordinate system mismatch between the multi-source heterogeneous devices, and realizes the accurate alignment of the inertial measurement unit, the optical marker system and the video stream data in the space-time dimension.

[0112] The above only describes the embodiments of the application and is not used to limit the protection scope of the application. For those skilled in the art, the application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A physical training test method, characterized by, The method comprises: synchronously acquiring motion capture data through a multi-modal data acquisition device, the motion capture data comprising optical action data and surface electromyography signals; performing data preprocessing on the original motion capture data, the data preprocessing comprising coordinate system normalization processing and timestamp calibration processing; performing multi-level data validity verification processing on the preprocessed motion capture data to obtain a verification result; in response to the verification result representing data validity, when the motion capture data is optical action data, mapping the motion capture data to joint kinematics parameters based on a biomechanical model, and generating physical action evaluation parameters according to the joint kinematics parameters; wherein the multi-level data validity verification processing comprises: primary verification: verifying that the total number of data frames of the motion capture data reaches a preset frame number threshold; intermediate verification: verifying that the time interval between adjacent data frames is less than or equal to a preset time threshold; advanced verification: confirming that the proportion of invalid data points in the motion capture data is lower than a preset proportion threshold, and the length of the continuous valid data segment is greater than a preset continuous frame threshold; the multi-level data validity verification processing adopts progressive verification logic, and the subsequent verification is terminated when the current level verification fails; the generation of physical action evaluation parameters comprises: performing frame processing on the motion capture data to obtain an action frame sequence; for each action frame in the action frame sequence, performing: verifying that the human motion region bounding box coverage rate is greater than a preset coverage rate threshold; performing bone joint positioning processing to obtain a set of bone joint position coordinates; generating motion trajectory data based on the set of bone joint position coordinates; performing speed analysis processing on the motion trajectory data to obtain an instantaneous speed vector; performing angle deviation calculation processing on the motion trajectory data to obtain a joint angle deviation value; generating a dynamics index according to the instantaneous speed vector and the joint angle deviation value; screening action frames with a joint angle deviation value less than a preset angle threshold as key action frames; performing feature extraction processing on the key action frames to obtain motion feature data; generating the physical action evaluation parameters according to the motion feature data.

2. The method of claim 1, wherein, The speed analysis processing comprises: performing time differentiation processing on the motion trajectory data to generate an original speed sequence; performing filtering processing on the original speed sequence to generate a smoothed speed sequence; performing acceleration calculation processing on the smoothed speed sequence to generate an acceleration sequence; extracting the maximum acceleration value of the acceleration sequence as an explosive force index; determining the smoothed speed sequence as the instantaneous speed vector.

3. The method of claim 1, wherein, The feature extraction processing comprises: performing bone point detection processing on the key action frames to obtain bone point coordinates; performing time sequence alignment processing on the bone point coordinates to generate an aligned bone sequence; performing electromyography signal integration processing on the aligned bone sequence to generate muscle activation time sequence data; determining the muscle activation time sequence data as the motion feature data.

4. The method of claim 1, wherein, when the motion capture data is a surface electromyography signal: the data validity verification processing comprises calculating the ratio of signal power to noise power; generating physical action evaluation parameters from motion capture data comprises: Baseline correction processing and power frequency filtering processing are performed on the surface electromyography signal to obtain a clean electromyography signal; Feature extraction processing is performed on the clean electromyography signal to generate an electromyography feature sequence; Fatigue calculation processing is performed on the electromyography feature sequence to obtain a muscle fatigue index; The muscle fatigue index is determined as the physical action evaluation parameter.

5. The method of claim 2, wherein, The feature extraction processing further includes: Lower limb joint detection processing is performed on the key action frame to obtain a lower limb bone joint position coordinate set, which includes hip joint coordinates, knee joint coordinates, and ankle joint coordinates; Cross-frame trajectory alignment processing is performed on the lower limb bone joint position coordinate set to generate a lower limb motion trajectory; Velocity analysis processing is performed on the lower limb motion trajectory: The displacement change amount per unit time is calculated to generate a lower limb instantaneous velocity sequence; Smoothing processing is performed on the lower limb instantaneous velocity sequence; The ratio of the acceleration segment duration to the deceleration segment duration is extracted; The explosive strength index and the ratio of the acceleration segment duration to the deceleration segment duration are integrated as motion feature data.

6. The method of claim 1, wherein, Before performing the obtaining motion capture data, further includes: Receiving a test start instruction containing a geographic location identifier; Obtaining the current device position coordinates based on a positioning device; Calculating the Euclidean distance between the current device position coordinates and the pre-set training area center coordinates; Verifying that the Euclidean distance is less than or equal to the pre-set area radius threshold; In response to the verification passing, activating the data acquisition device matching the test type and sending the acquisition parameter configuration instruction.

7. The method of claim 6, wherein, The obtaining motion capture data includes: Acquiring three-dimensional acceleration and angular velocity data through an inertial measurement unit; The obtaining motion capture data includes: Acquiring three-dimensional coordinate data of infrared reflective marker points through an optical motion capture system; The obtaining motion capture data includes: Generating a two-dimensional motion trajectory based on continuous inter-frame motion vector analysis of a video acquisition device.

8. The method of claim 7, wherein, Further includes: Establishing a historical evaluation database storing athlete identifiers, test times, and physical action evaluation parameters; Comparing the currently generated physical action evaluation parameters with the historical data set of the same athlete identifier to: Calculate the time series change rate of the joint angle deviation value, and generate an action deformation warning when the change rate exceeds a pre-set change threshold; Perform linear regression modeling on the explosive strength index to output a regression coefficient sign change detection result; Calculate a recommended recovery period based on the exponential decay curve of the muscle fatigue index; Generating a biomechanics report containing the joint angle deviation value change rate, the regression coefficient sign change detection result, and the recommended recovery period.

Images