Group physical training control system based on motion capture and ai load matching

Abstract

This invention relates to the field of intelligent motion analysis and group collaborative control, and particularly to a group physical training control system based on motion capture and AI load matching. The group physical training control system based on motion capture and AI load matching includes: a 3D motion acquisition and manifold construction module, used to acquire 3D skeletal joint data of each participant, construct a standard motion manifold space, and map the skeletal sequences of each participant to the standard motion manifold space to obtain posture deviations and group spatial topological relationships; and a motion cycle division and joint feature extraction module, used to divide the movements of each participant into cycles and extract the motion trajectory and velocity change features of each joint within the cycle. This invention, by constructing indirect quantization based on joint energy distribution and group phase coupling, achieves calculable characterization and real-time risk assessment of group motion disturbances, improving collaborative stability and the accuracy of control.

Description

Technical Field

[0001] This invention relates to the field of intelligent motion analysis and group collaborative control, and in particular to a group physical training control system based on motion capture and AI load matching. Background Technology

[0002] In group movement training and collaborative drills, the consistency and rhythmic synchronization of multiple participants are crucial factors affecting overall performance and safety. Existing technologies typically assess the degree of movement standardization by collecting three-dimensional skeletal joint data of trainees and comparing it with standard movements. However, in actual training, the impact of individuals on group rhythm is not entirely manifested as explicit postural deviations and trajectory errors. Instead, it often appears as temporal misalignment and uneven energy distribution during multi-joint coordination. This is characterized by its concealment and cumulative nature, making it difficult to identify through direct observation. However, it gradually amplifies during continuous movement execution, ultimately disrupting the overall rhythmic stability. Therefore, analytical methods relying solely on spatial structure or instantaneous states are insufficient to accurately reflect potential rhythmic disturbances during movement execution. In group movement analysis, most existing technologies rely on joint position deviations or posture similarity indices for evaluation, and depend on uniform beat guidance or synchronization strategies based on average phase for rhythm control. They primarily process directly measurable data, lacking the ability to quantify the degree of individual influence on group rhythm within the movement cycle. However, the degree of influence on group rhythm is a latent variable that is difficult to obtain directly and cannot be measured by sensors or single feature parameters. Due to the lack of a mechanism to establish a mapping relationship between the distribution characteristics of energy over time during joint movement and latent variables, existing methods struggle to identify key individuals and their degree of influence in the early stages of rhythm instability. Furthermore, the control process typically employs uniform or empirical strategies, lacking adaptive intervention capabilities based on individual differences, thus limiting the precision and real-time performance of group movement rhythm control. Summary of the Invention

[0003] To overcome the shortcomings of group rhythm disturbances being difficult to quantify directly, implicit variables being difficult to perceive, and the lack of adaptive intervention mechanisms, this invention provides a group physical training control system based on motion capture and AI load matching.

[0004] The technical implementation scheme of the present invention is as follows: a group physical training control system based on motion capture and AI load matching, comprising: The 3D motion acquisition and manifold construction module is used to acquire 3D skeletal joint data of each trainee, construct a standard motion manifold space, and map the skeletal sequence of each trainee to the standard motion manifold space to obtain posture deviation and group spatial topology. The motion cycle division and joint feature extraction module is used to divide the motion of each trainee into cycles and extract the motion trajectory and speed change features of each joint within the cycle. The joint energy distribution and concentration calculation module is used to construct the motion energy distribution based on the velocity change characteristics of each joint, determine the energy peak position, calculate the energy proportion within a preset time window, and obtain the energy concentration of each joint. The energy leakage index calculation module is used to calculate the average energy concentration and dispersion based on the energy concentration of each joint, and to calculate the energy leakage index based on the average energy concentration and dispersion. The group relationship and phase feature calculation module is used to determine the relative position of the trainees based on the group spatial topology and to calculate the movement progress to obtain the movement phase difference; The risk assessment and intervention decision-making module is used to construct risk potential energy and determine intervention conditions based on the contribution of group rhythm disturbance, the relative position and the phase difference of the action. The auditory phase perturbation execution module is used to apply phase-leading audio beat pulses to adjust the rhythm of the movement when intervention conditions are met.

[0005] Preferably, the acquisition of three-dimensional skeletal joint data for each trainee includes: in a group training scenario, acquiring the three-dimensional skeletal joint position data of each trainee through a three-dimensional vision acquisition device, and constructing a set of joint temporal data for each trainee, represented as: ;in, For the first Each student at any moment The set of joint spatial locations For the first The first student's Each joint at any time The three-dimensional spatial position vector, Number the joints. , The total number of joints; the joint timing data set is segmented according to the continuity of movement to determine the interval of each movement cycle, represented as: ;in, For the first The first student's Each action cycle interval The start time of the cycle, To correspond to the duration of the action cycle, within the action cycle interval, the spatial position vectors of each joint are aligned using a unified time reference.

[0006] Preferably, the step of dividing each student's movements into periods and extracting the motion trajectory and velocity change features of each joint within the period includes: analyzing the joint spatial position vectors of each student. Time series analysis was performed to obtain the continuous motion trajectory of each joint within the motion cycle interval; based on the joint spatial position vector, the velocity change characteristics of each joint between adjacent sampling times were calculated using a discrete difference method. ;in, For the first The first student's Each joint at any time The velocity vector, The time interval between adjacent sampling moments is defined as follows: The velocity vector is smoothed using a sliding window process to obtain a smoothed velocity change sequence, as shown in the formula: ;in, The smoothed velocity variation characteristics, This is the length of the sliding window.

[0007] Preferably, the step of constructing the motion energy distribution based on the velocity change characteristics of each joint, determining the energy peak position, and calculating the energy proportion within a preset time window to obtain the energy concentration of each joint includes: based on the smoothed velocity change characteristics of each joint. Construct the motion energy distribution function of each joint in the time dimension, the expression of which is: ;in, For the first The first student's Each joint at any time kinetic energy density, The vector magnitude is represented by: Within the motion cycle interval, extreme value detection is performed on the motion energy distribution function to determine the peak time of motion energy at each joint, expressed as: ;in, For the first The first student's The peak energy time of each joint within the current motion cycle; a preset time window is constructed centered on the peak energy time: ;in, For the first Energy concentration time window for each joint The window width is used to characterize the time range of energy concentration analysis; based on the energy concentration time window and the motion energy distribution function, the energy proportion of each joint within the energy concentration time window is calculated to obtain the energy concentration of each joint.

[0008] Preferably, the step of calculating the energy proportion of each joint within the energy concentration time window based on the energy concentration time window and the motion energy distribution function to obtain the energy concentration degree of each joint includes: calculating the energy concentration degree of each joint within the energy concentration time window. The formulas for calculating the window energy and the total energy over the entire action cycle are as follows: ;in, For the first The first student's The cumulative energy value of each joint within the energy concentration time window. For the first The first student's The cumulative energy value of each joint throughout the entire motion cycle; based on the ratio between the window energy and the total energy, the energy concentration of each joint is calculated, expressed as: ;in, For the first The first student's The energy concentration of a joint during the current motion cycle is used to characterize the degree of concentration of the joint's motion energy in the time dimension.

[0009] Preferably, the step of calculating the average energy concentration and dispersion based on the energy concentration of each joint, and calculating the energy leakage index based on the average energy concentration and dispersion, includes: calculating the energy concentration of each joint... Calculate the average energy concentration of each trainee during the current action cycle: ;in, For the first The average energy concentration of each trainee during the current movement cycle. The total number of joints; based on the deviation between the energy concentration of each joint and the average energy concentration, the dispersion of energy concentration is calculated, expressed as: ;in, For the first The dispersion of energy concentration at each joint of each trainee; based on the average energy concentration and the dispersion, an energy leakage index is constructed to indirectly quantify the contribution of the group rhythm disturbance, the formula is: ;in, For the first The energy leakage index of each trainee during the current action cycle. The proportional coefficient used to adjust the weighting of dispersion is used to comprehensively characterize the dispersion of joint motion energy in the time dimension and the degree of inconsistency between joints; the contribution of group rhythm perturbation is defined. The contribution of group rhythm disturbance is a latent variable used to characterize the degree of influence of an individual learner on the overall rhythm stability. There is a mapping relationship between the energy leakage index and the degree of dispersion of joint motion energy over time and the coordination consistency between joints. As the contribution of the group rhythm disturbance The only computable indirect representation of , namely: .

[0010] Preferably, determining the relative positions of trainees based on the group's spatial topology and calculating the movement progress to obtain the movement phase difference includes: extracting the spatial topology of each trainee in the group by using the embedding positions of each trainee's skeletal sequence in the standard movement manifold space; and determining the relative positional relationship between any two trainees based on the adjacency relationship of each trainee in the manifold space, expressed as: ;in, For the first The student and the first Each student at any moment spatial distance For the first The position representation of each trainee in the standard motion manifold space; the time mapping of the joint timing data set is normalized, and the motion progress function is constructed, with the expression: ;in, For the first The percentage of movement completed by each trainee within the current movement cycle; by comparing the differences in movement progress among different trainees, the movement phase difference is calculated using the following formula: ;in, For the first The student and the first Each student at any moment The phase difference of the movements is used to reflect the sequential relationship between trainees in the rhythm of movement execution.

[0011] Preferably, the step of constructing risk potential energy and determining intervention conditions based on energy leakage index, relative position, and action phase difference includes: for each participant within the same action cycle, comprehensively considering the contribution of the group rhythm disturbance. Spatial distance between trainees and motion phase difference A risk potential function reflecting the group's collaborative risk state is constructed. By coupling the individual abnormality degree represented by the group rhythm disturbance contribution and the group interaction relationship, the risk potential value of any student is obtained. ;in, For the first Each student at any moment The risk potential value, To prevent tiny positive numbers with a denominator of zero; Used to extract the first The degree to which a student's movements are ahead of other students is only when the first student... When a participant is ahead of others in terms of progress, they participate in risk accumulation; the risk potential energy of all participants at the same moment is normalized to obtain a standardized risk index, expressed as: ;in, For the normalized risk potential energy, This represents the current set of trainees participating in the training. Based on this, by comparing the normalized risk potential with the preset risk threshold and combining the changing trend within multiple consecutive action cycles, the satisfaction of the intervention conditions is determined. When the normalized risk potential continues to rise and exceeds the preset risk threshold, an intervention decision is triggered.

[0012] Preferably, the Used to extract the first The degree to which a student's movements are ahead of other students is only when the first student... When a participant is ahead of other participants in the progress of a movement, they participate in risk accumulation, including: the phase difference of the movement. Perform direction-selective mapping to construct a directed phase action function, the expression of which is: ;in, For the first The student on the first Effective leading phase effect of each student For a unit step function, when When, take 1, when When the time is 0; a phase effect normalization process is introduced, which is expressed as: ;in, This is the normalized directed phase action. To prevent tiny positive numbers with a denominator of zero, the rhythmic influence relationship of each trainee in the group is reconstructed by weighting based on the normalized directed phase action.

[0013] Preferably, the step of applying a phase-leading audio beat pulse to adjust the movement rhythm when the intervention conditions are met includes: after determining that the intervention conditions are met, using the normalized risk potential... As the basis for adjustment, the phase shift of the audio beat pulse is adaptively calculated, and a phase perturbation function is constructed, the expression of which is: ;in, To be applied to the first The phase offset of the audio beat for each student. The phase adjustment amplitude coefficient is greater than zero. This is the phase difference suppression coefficient. For the first The deviation of each student from the group reference phase; based on the phase offset, the standard beat signal is phase-shifted to generate a personalized audio beat pulse signal, expressed as: ;in, The output audio beat signal, The frequency is the beat angular frequency. During the application of the audio beat pulse, a gradual adjustment strategy is introduced. By setting a phase change rate constraint, the phase shift change between adjacent moments is ensured to satisfy the following: ;in The time interval between adjacent control cycles. The upper limit of the phase change rate; monitoring the phase difference of the action within multiple consecutive action cycles. With the energy leakage index The trend of change is such that when the phase difference of the action gradually decreases and the contribution of the group rhythm disturbance drops below the preset stable threshold, the phase offset is gradually reduced until the standard beat signal is restored.

[0014] Beneficial Effects: This invention constructs a unified technical framework by coupling three-dimensional motion manifold modeling and joint temporal energy distribution analysis with group coordination relationships, achieving a fusion expression of motion spatial structure features and temporal rhythm features. By extracting the concentration and dispersion characteristics of joint motion energy in the time dimension, the contribution of group rhythm disturbances, which cannot be directly observed, is transformed into a calculable indirect representation, enabling a quantitative characterization of the individual's impact on overall rhythm stability, thereby improving the ability to perceive hidden rhythm instability problems. By modeling the consistency of multi-joint energy distribution, early identification of coordination mismatch during motion execution is achieved, reducing the overall risk caused by the accumulation of rhythm deviations. The contribution of group rhythm disturbances is coupled with the group spatial topology and motion phase difference to construct a risk potential function, enabling a unified analysis of individual anomalies and group interaction relationships, and implementing phase-advanced audio beat perturbation control, making the control process targeted and adaptive. The introduction of progressive phase change constraints and multi-cycle trend judgment mechanisms reduces the interference of beat mutations on motion stability and achieves smooth exit from control, improving the consistency and rhythm stability of group motions while enhancing the real-time performance and precision of the control process. Attached Figure Description

[0015] Figure 1 This is a structural diagram of the group physical training control system based on motion capture and AI load matching of the present invention. Figure 2 This is a flowchart of the method for quantifying and assessing the group rhythm disturbance of the present invention. Detailed Implementation

[0016] The present invention will be further described below with reference to specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0017] Example 1: A group physical training control system based on motion capture and AI load matching, such as Figure 1 As shown, it includes: The 3D motion acquisition and manifold construction module is used to acquire 3D skeletal joint data of each trainee, construct a standard motion manifold space, and map the skeletal sequence of each trainee to the standard motion manifold space to obtain posture deviation and group spatial topology. The motion cycle division and joint feature extraction module is used to divide the motion of each trainee into cycles and extract the motion trajectory and speed change features of each joint within the cycle. The joint energy distribution and concentration calculation module is used to construct the motion energy distribution based on the velocity change characteristics of each joint, determine the energy peak position, calculate the energy proportion within a preset time window, and obtain the energy concentration of each joint. The energy leakage index calculation module is used to calculate the average energy concentration and dispersion based on the energy concentration of each joint, and to calculate the energy leakage index based on the average energy concentration and dispersion. The group relationship and phase feature calculation module is used to determine the relative position of the trainees based on the group spatial topology and to calculate the movement progress to obtain the movement phase difference; The risk assessment and intervention decision-making module is used to construct risk potential energy and determine intervention conditions based on the contribution of group rhythm disturbance, the relative position and the phase difference of the action. The auditory phase perturbation execution module is used to apply phase-leading audio beat pulses to adjust the rhythm of the movement when intervention conditions are met.

[0018] In group training scenarios, 3D skeletal joint position data of each trainee are acquired using 3D vision acquisition devices, and a set of joint temporal data for each trainee is constructed, represented as follows: ;in, For the first Each student at any moment The set of joint spatial locations For the first The first student's Each joint at any time The three-dimensional spatial position vector, Number the joints. , The total number of joints; the joint timing data set is segmented according to the continuity of movement to determine the interval of each movement cycle, represented as: ;in, For the first The first student's Each action cycle interval The start time of the cycle, To correspond to the duration of the action cycle, within the action cycle interval, the spatial position vectors of each joint are aligned using a unified time reference.

[0019] It should be explained that the 3D vision acquisition device is a multi-view depth sensing device deployed in a group training scenario, using a multi-view RGB-D camera array to output structured 3D skeletal joint position data of each trainee in real time; the data is transmitted in a time-series stream, with each frame containing a trainee identifier and a timestamp, as well as the 3D spatial coordinates of each joint node in a preset human skeletal model; the preset human skeletal model is a set of hierarchical joint structures defined according to human biomechanical standards, sourced from publicly available industry standards for human posture estimation skeletal models; the preset human skeletal model provides a unified numbering index rule and spatial connection relationship definition for each joint, enabling comparable anatomical correspondences for joint position data collected from different trainees at different times; for the first... Each student, the students in continuous time The joint position data are organized in chronological order to construct the joint temporal data set. ; where subscript For student identifiers; Includes the three coordinate components of the joint in a Cartesian coordinate system; subscript The joints are numbered according to the joint indexing rules in the human skeletal model; the step of segmenting according to the continuity of movement involves detecting zero-crossing points on the velocity curves of key joints, using the moment when the velocity direction of the hip joint reverses in the main direction of movement as the period segmentation boundary, and determining each movement cycle interval after statistical filtering; the movement cycle interval is represented as... superscript The period number indicates the number of the period. The first student's A complete action cycle; This is the start time of the cycle; The period duration is calculated from the time difference between the end time and the start time; the unified time reference alignment process determines an equally spaced reference time sequence within the action period interval, and uses cubic spline interpolation to resample the original spatial position vector sequence of each joint onto the reference time sequence, so that each joint has the same number of data points with the same timestamp within the period.

[0020] Joint spatial position vectors for each trainee Time series analysis was performed to obtain the continuous motion trajectory of each joint within the motion cycle interval; based on the joint spatial position vector, the velocity change characteristics of each joint between adjacent sampling times were calculated using a discrete difference method. ;in, For the first The first student's Each joint at any time The velocity vector, The time interval between adjacent sampling moments is defined as follows: The velocity vector is smoothed using a sliding window process to obtain a smoothed velocity change sequence, as shown in the formula: ;in, The smoothed velocity variation characteristics, This is the length of the sliding window.

[0021] It should be explained that, within the aforementioned action cycle range Within this framework, the acquisition of the continuous motion trajectory is based on the joint spatial position vector sequence after alignment processing using the unified time reference. Since the alignment process gives each joint data point the same timestamp sequence, the time series analysis is performed directly on discrete data points at equal time intervals. By connecting the spatial coordinates of adjacent sampling points in time sequence, a continuous motion trajectory representation of each joint in three-dimensional space is formed. In the velocity change characteristic formula, the... The time interval between adjacent sampling moments is determined by the nominal sampling frequency of the 3D vision acquisition device. Equal to the reciprocal of the sampling frequency; the velocity vector The orientation represents the joint at time. The instantaneous direction of motion, and the modulus length characterizes the joint at time [time]. The instantaneous motion rate; the three-dimensional vision acquisition device has inherent measurement noise in the image sensing and skeleton estimation algorithm regression stages. The velocity change characteristics calculated directly based on the original position vector will be superimposed with high-frequency noise components, affecting the accuracy of energy distribution analysis. Therefore, the velocity vector is smoothed using a sliding window; in the smoothing formula, This represents the variable after smoothing. This is the index of the sampling point offset within the sliding window. Corresponding to the current moment sampling points, The corresponding backward tracing One historical sampling point. The sliding window length. The acquisition method is as follows: based on the noise power spectral density characteristics of the three-dimensional vision acquisition device, the cutoff frequency boundary between the action signal and the noise signal is determined by frequency domain analysis, and the corresponding time domain window width is calculated according to the cutoff frequency, and optimization is performed with the retention of action detail features as a constraint. It retains the core information of motion trends while suppressing high-frequency random noise.

[0022] Based on the smoothed velocity change characteristics of each joint Construct the motion energy distribution function of each joint in the time dimension, the expression of which is: ;in, For the first The first student's Each joint at any time kinetic energy density, The vector magnitude is represented by: Within the motion cycle interval, extreme value detection is performed on the motion energy distribution function to determine the peak time of motion energy at each joint, expressed as: ;in, For the first The first student's The peak energy time of each joint within the current motion cycle; a preset time window is constructed centered on the peak energy time: ;in, For the first Energy concentration time window for each joint The window width is used to characterize the time range of energy concentration analysis; based on the energy concentration time window and the motion energy distribution function, the energy proportion of each joint within the energy concentration time window is calculated to obtain the energy concentration of each joint.

[0023] It should be explained that, within the aforementioned action cycle range Within, the kinetic energy distribution function Based on the smoothed velocity change characteristics Construction; the motion energy density, in a physical sense, characterizes the joint at time... The instantaneous intensity of motion is calculated using the square of the velocity vector magnitude; the sign of the vector magnitude is... This involves calculating the Euclidean norm of a three-dimensional velocity vector by taking the square root of the sum of the squares of its three coordinate components. In discrete data implementation, the magnitude calculation directly performs numerical operations on the three components of the velocity vector. Within the motion cycle interval, extreme value detection is performed on the motion energy distribution function, and the values ​​at each sampling time within the interval are... The values ​​form a discrete sequence. By comparing the energy density values ​​of each sampling point with those of its neighbors, local maxima are identified. When the energy density value of a sampling point is simultaneously greater than that of its immediate neighbors, that sampling point is marked as a candidate energy peak point. For cases with multiple candidate energy peak points, the absolute values ​​of the energy densities of each candidate peak point are further compared, and the candidate point with the highest energy density is selected as the main energy peak moment of the joint within the action cycle. When the energy distribution exhibits a multi-peak structure due to the action characteristics, the main peak with the highest energy density is used as the central reference point for energy concentration analysis. The time corresponding to the energy peak is represented as... The preset time window The window is constructed symmetrically, expanding outwards from the energy peak moment; the half-width of the window... To define the time range for energy concentration analysis, based on the statistical characteristics of the energy distribution curves of each joint in the standard motion sample database, the half-width at half-maximum (WHM) parameter of the energy distribution curve is calculated. The WHM parameter is defined as the time span between the two moments corresponding to when the energy density value drops to half of its peak value; the window half-width... The value of is related to the half-height full width parameter, with the goal of ensuring that the energy concentration time window covers the core region of the energy peak. Specifically, statistical analysis is performed on the energy distribution curves of each joint in the same type of movement in the standard movement sample to obtain the distribution characteristics of the half-height full width parameter, and the corresponding value that can cover the half-height full width range of the preset proportion is determined as the half-width of the window. The energy concentration time window The interval length is The left boundary is the peak time minus half the window width, and the right boundary is the peak time plus half the window width; the process of calculating the energy ratio based on the energy concentration time window and the motion energy distribution function is used as the ratio of the energy accumulation value within the window to the energy accumulation value within the entire motion cycle; the ratio reflects the degree of concentration of motion energy of each joint in the time dimension.

[0024] Calculate the time window of energy concentration for each joint The formulas for calculating the window energy and the total energy over the entire action cycle are as follows: ;in, For the first The first student's The cumulative energy value of each joint within the energy concentration time window. For the first The first student's The cumulative energy value of each joint throughout the entire motion cycle; based on the ratio between the window energy and the total energy, the energy concentration of each joint is calculated, expressed as: ;in, For the first The first student's The energy concentration of a joint during the current motion cycle is used to characterize the degree of concentration of the joint's motion energy in the time dimension.

[0025] It should be explained that, within the aforementioned action cycle range Inside, the window energy and the total energy The calculation is performed in the discrete data domain using numerical integration; due to the motion energy distribution function Defined on an aligned, equally spaced discrete-time series, the integral sign is... The process is converted into a weighted summation of energy density values ​​at discrete sampling points. The numerical integration method employs the trapezoidal rule, where the energy contribution between any two adjacent sampling points within the integration interval is approximated by multiplying the average energy density values ​​at those two points by the sampling time interval. The energy contributions of all small intervals within the interval are then summed to obtain the integration result. The discrete calculation formula for the trapezoidal rule is expressed as: ,in and These represent the start and end times of the integration interval, respectively. This represents the total number of sampling points within the interval. For the first Each sampling point corresponds to a time; for the calculation of the window energy, the The value is taken as the left boundary of the energy concentration time window. The The value is taken as the right boundary of the energy concentration time window. Regarding the calculation of the total energy, the The value is the start time of the action cycle interval. The The value is the end time of the action cycle interval. The sampling time interval Determined by the nominal sampling frequency of the 3D vision acquisition device; based on the duration of the action cycle. Based on this, the number of sampling points within a single action cycle is calculated. When the number of sampling points results in the relative error between the discrete numerical integration result and the theoretical value of the continuous integration being less than a preset error tolerance, the current sampling frequency is determined to meet the calculation accuracy requirements. The preset error tolerance is based on energy concentration. The calculation formula establishes the sampling time interval. Discretization error to The error propagation model of the calculation results, with The preset error tolerance is derived by reverse derivation from the allowable deviation range; the allowable deviation range is based on the population energy leakage index. The sensitivity requirement for differentiating individual movement quality scores is set by analyzing the differences between standard movement samples and abnormal movement samples. The statistical difference magnitude determines the upper limit of the allowable deviation; the sampling time interval The determination method is the same as the method for determining the current sampling frequency. With action cycle duration When the ratio is less than a preset ratio threshold, it is determined that... Sufficiently small to meet the discrete integration accuracy requirements; using the preset error tolerance as a constraint, numerical integration accuracy is tested on simulation data of a standard sinusoidal motion trajectory, establishing a mapping relationship curve between the ratio of sampling interval to motion cycle and the relative integration error, and selecting the ratio that makes the relative integration error first enter the preset error tolerance range as the preset ratio threshold; for the window energy The calculation, the integration interval is the energy concentration time window. Because the window boundary time may not precisely coincide with the discrete sampling time, in actual calculations, the sampling interval falling within the window boundary time is used as a reference. The smallest subset of discrete sampling points covering the window is selected as the actual integration range, and the trapezoidal rule is used for accumulation calculation. When the window boundary exceeds the action cycle interval... At that time, using the boundary of the action cycle interval as the cutoff limit, only the energy accumulation within the intersection range of the energy concentration time window and the action cycle interval is calculated; for the total energy The calculation, the integration interval is the entire action cycle interval. The start and end times of the interval correspond to the start and end times of the aligned discrete time series, and the energy density values ​​of all sampling points within the interval are accumulated over the entire interval using the trapezoidal rule; the energy concentration... The energy concentration is calculated from the ratio of the window energy to the total energy, with a value between 0 and 1. The energy concentration measure quantifies the degree to which joint motion energy gathers towards the peak moment in the time dimension. The closer the energy concentration value is to 1, the more concentrated the joint motion energy is within a narrow time window near the peak moment, and the movement exhibits characteristics of strong explosive power and crisp force exertion. The closer the value is to 0, the more evenly the joint motion energy is distributed within the movement cycle, and the movement has problems such as dragging, hesitation, or insufficient control precision.

[0026] Based on the energy concentration of each joint Calculate the average energy concentration of each trainee during the current action cycle: ;in, For the first The average energy concentration of each trainee during the current movement cycle. The total number of joints; based on the deviation between the energy concentration of each joint and the average energy concentration, the dispersion of energy concentration is calculated, expressed as: ;in, For the first The dispersion of energy concentration at each joint of each trainee; based on the average energy concentration and the dispersion, an energy leakage index is constructed to indirectly quantify the contribution of the group rhythm disturbance, the formula is: ;in, For the first The energy leakage index of each trainee during the current action cycle. The proportional coefficient used to adjust the weighting of dispersion is used to comprehensively characterize the dispersion of joint motion energy in the time dimension and the degree of inconsistency between joints; the contribution of group rhythm perturbation is defined. The contribution of group rhythm disturbance is a latent variable used to characterize the degree of influence of an individual learner on the overall rhythm stability. There is a mapping relationship between the energy leakage index and the degree of dispersion of joint motion energy over time and the coordination consistency between joints. As the contribution of the group rhythm disturbance The only computable indirect representation of , namely: .

[0027] It should be explained that, within the aforementioned action cycle range Within, the average energy concentration The calculation is based on the energy concentration of each joint in the trainee's body. Based on the statistical average; the summation symbol Indicates the numbering of joints From 1 to The energy concentration of all joints is summed up, and the coefficient is... The average factor for the accumulated value; the average energy concentration reflects the overall level of the trainee's whole-body movement quality within the current movement cycle. A higher average energy concentration indicates that the trainee's joints exhibit good explosive power concentration characteristics; a lower average energy concentration indicates that the trainee's overall movement suffers from energy dispersion and sluggish force exertion; the degree of dispersion... The calculation uses variance statistics, and in the formula, the... The term represents the first The deviation between the energy concentration of each joint and the average energy concentration of the whole body is calculated by squaring the deviations, summing them, and averaging them to obtain a measure of the dispersion of the energy concentration of each joint around the mean. This dispersion quantifies the degree of coordination between the student's joints. A smaller dispersion indicates that the energy concentration values ​​of the student's joints are close to each other, and that the joints have good coordination in terms of timing and explosiveness. A larger dispersion indicates that there is a lack of coordination, with some joints moving crisply while others move sluggishly, resulting in poor consistency between joints. The energy leakage index... It consists of two weighted sums; the first term For the overall energy dispersion term, since The value is between 0 and 1. The first term converts the average energy concentration into a measure of the overall energy dispersion. When the value is close to 1, the first term is close to 0, indicating a low overall energy leakage rate; when When the value is close to 0, the first term is close to 1, indicating a high overall energy leakage rate; the first term quantifies the degree to which energy is not concentrated at its peak; the second term... The second term, which is a penalty for inconsistency between joints, incorporates the dispersion of energy concentration at each joint into the calculation framework of the energy leakage index. The greater the dispersion, the greater the contribution of the second term, indicating additional energy leakage caused by poor coordination between joints. The combined logic of the two terms is that a trainee's energy leakage may stem from both sluggish overall body movements and misalignment of the timing of force exertion between joints. The energy leakage index... Simultaneously capturing motion quality information in both dimensions; the scaling factor The proportionality coefficient is used to adjust the relative contribution weight of the inter-joint inconsistency penalty term in the energy leakage index. The determination method involves collecting sample data from a standard movement sample database, wherein the sample data includes movement quality rating levels marked by professional coaches; and calculating the average energy concentration for each sample. and degree of dispersion To maximize the energy leakage index The correlation coefficient between the action quality rating and the target value is used as the optimization objective. A one-dimensional search method is employed to optimize the coefficient within a preset value range. Optimization will maximize the correlation coefficient. The value is determined to be the aforementioned proportionality coefficient; based on the energy leakage index. The two components and The difference in numerical magnitude in the standard motion sample, so that... The lower limit of the interval can be set based on the principle of adjusting the two items to have similar orders of magnitude of contribution. The upper limit of the interval is set based on the principle that further increasing the value will no longer significantly improve the correlation coefficient, thus determining the preset value interval; the contribution of the group rhythm disturbance... , is a latent variable used to characterize the degree of influence of an individual trainee on the overall rhythm stability; the contribution of the group rhythm disturbance. The technical principle underlying the mapping relationship between the degree of energy dispersion in the temporal dimension of joint movement and the coordination consistency between joints is that the micro-instability of a trainee's own hesitation, adjustment, and correction in movement affects the movement rhythm of surrounding trainees through visual coupling and rhythm perception. The externally observable manifestations of these micro-instabilities are the dispersion of energy in the temporal dimension and the inconsistency in the timing of force exertion between joints; the energy leakage index... The microscopic instability was quantified from two dimensions: overall energy dispersion and inter-joint inconsistency, and the energy leakage index was... As the contribution of the group rhythm disturbance The only computable indirect representation of .

[0028] By analyzing the embedding positions of each trainee's skeletal sequence in the standard action manifold space, the spatial topological relationships of each trainee within the group are extracted. Based on the adjacency relationships of each trainee in the manifold space, the relative positional relationships between any two trainees are determined, expressed as follows: ;in, For the first The student and the first Each student at any moment spatial distance For the first The position representation of each trainee in the standard motion manifold space; the time mapping of the joint timing data set is normalized, and the motion progress function is constructed, with the expression: ;in, For the first The percentage of movement completed by each trainee within the current movement cycle; by comparing the differences in movement progress among different trainees, the movement phase difference is calculated using the following formula: ;in, For the first The student and the first Each student at any moment The phase difference of the movements is used to reflect the sequential relationship between trainees in the rhythm of movement execution.

[0029] It should be explained that the standard movement manifold space is a low-dimensional manifold embedding space. By collecting skeletal sequence data from standard movement demonstrators as training samples for manifold learning, an isometric mapping algorithm is used to perform nonlinear dimensionality reduction on the high-dimensional skeletal pose data, extracting low-dimensional manifold coordinates that reflect the inherent geometric structure of the movement, thus forming the standard movement manifold space. After each trainee's skeletal sequence is projected onto the standard movement manifold space through a manifold mapping function, each trainee at time... The posture is represented as a low-dimensional coordinate point in the standard action manifold space, i.e., the position representation. The location representation The dimension of the coordinates is lower than that of the original skeletal joint position vector, and in the standard motion manifold space, the relative distance between coordinate points reflects the semantic similarity of the corresponding poses; the extraction process of the group space topology is based on the position representation of each trainee in the standard motion manifold space; the adjacency relationship is determined by using the position representation of each trainee. For each node, the manifold distance between that node and other student nodes is calculated, and an adjacency graph is constructed using either the k-nearest neighbor method or a distance threshold-based method. When using the k-nearest neighbor method, for each student node, a preset number of other student nodes with the closest manifold distance to that node are selected to establish adjacency edges. When using the distance threshold method, an adjacency edge is established between two student nodes when the manifold distance between them is less than a preset adjacency distance threshold. The preset number and the preset adjacency distance threshold are determined based on statistical analysis of the spatial distribution density of students in a standard group training scenario, aiming to maintain the connectivity of the adjacency graph and avoid redundant edges. The relative positional relationships... The calculation uses Euclidean distance in the manifold space; in the formula, the subscripts... and They represent the first The first student and the first The identifier of each student, the The symbol represents the Euclidean norm of the difference between two positional representation vectors in the manifold space; the spatial distance Quantified the students With students At the semantic level of proximity in motion, a smaller distance indicates that the current postures of the two trainees are closer in the motion manifold, and their motion forms are more similar; a larger distance indicates that the differences in motion forms between the two trainees are more significant; the motion progress function... In the formula, For the current moment, This represents the start time of the current action cycle interval. The duration of the cycle corresponding to the action cycle; the action progress The value is between 0 and 1, when hour , indicates the start of the action cycle; when hour This indicates the completion of the action cycle; the normalization mapping process enables a unified comparison of action progress among trainees with different starting times and different cycle durations; the action phase difference It is calculated from the difference in the progress of the two trainees' movements. For the first Percentage of each student's actions completed. For the first The percentage of each student's actions completed; when When, it indicates the first The student is ahead of the first in terms of progress. One student; when When, it indicates the first One student is lagging behind the first in terms of progress. One student; when When the two trainees' movements are synchronized, it indicates that their movements are in sync. The phase difference of the movements serves as an input parameter for subsequent risk assessment, used to quantify the sequential relationship between trainees in terms of the rhythm of their movements. The trainee in the lead will have a rhythmic guiding effect on the trainee who is lagging behind.

[0030] For each participant within the same movement cycle, the contribution of the group rhythm disturbance is considered comprehensively. Spatial distance between trainees and motion phase difference A risk potential function reflecting the group's collaborative risk state is constructed. By coupling the individual abnormality degree represented by the group rhythm disturbance contribution and the group interaction relationship, the risk potential value of any student is obtained. ;in, For the first Each student at any moment The risk potential value, To prevent tiny positive numbers with a denominator of zero; Used to extract the first The degree to which a student's movements are ahead of other students is only when the first student... When a participant is ahead of others in terms of progress, they participate in risk accumulation; the risk potential energy of all participants at the same moment is normalized to obtain a standardized risk index, expressed as: ;in, For the normalized risk potential energy, This represents the current set of trainees participating in the training. Based on this, by comparing the normalized risk potential with the preset risk threshold and combining the changing trend within multiple consecutive action cycles, the satisfaction of the intervention conditions is determined. When the normalized risk potential continues to rise and exceeds the preset risk threshold, an intervention decision is triggered.

[0031] It should be explained that the aforementioned risk potential function The construction process involves quantifying the rhythm-following pressure faced by an individual trainee in a group training scenario into a calculable potential energy value; the risk potential energy function is determined by its contribution to the rhythm disturbance of the group. The individual's inherent instability is coupled with the group's interaction relationship in a coupled modeling process, enabling a numerical description of the group's collaborative risk state; in the formula, This indicates that, except for the first in the group The risk potential is accumulated by iterating through all students except the current student, and the risk potential is determined by all other students' actions on the first student. The rhythmic pressure components generated by each participant are superimposed; the contribution of the group rhythmic disturbance is... As a multiplicative factor placed inside the summation term, it means that the first term is the first multiplicative factor. The instability of a trainee's own movements determines their sensitivity to external rhythmic influences. The more severe the energy leakage, the greater the perceived risk potential energy under the same external rhythmic pressure. This allows risk assessment to adapt to individual differences, giving higher risk attention to trainees with poor movement quality. The item is used to extract the first... The degree to which each student's movements are ahead of the other students. Due to When the first The student's progress is ahead of the first. When there are individual students, , The function outputs the phase difference value; when the first The progress of the first student is lagging behind that of the second student. When there are individual students, , The function outputs 0, the This item does not participate in risk accumulation; the technical basis for selective directional handling includes the fact that in group physical training, trainees who move too quickly are more likely to disrupt the synchronization of group movements, therefore, by retaining the first... The lead of each participant relative to others accumulates risk for actions that might cause rhythm imbalance within the group, while the rhythm of lagging participants does not provide positive rhythmic impetus to the leading participant. By retaining only the phase information of the leading participant, the risk potential function can focus on the sources of actual rhythmic pressure exerted on the participants; the spatial distance... This means that the closer the spatial distance between trainees, the stronger the visual coupling and rhythm perception, and the higher the transmission efficiency of rhythm influence. Therefore, the risk potential energy component is inversely proportional to the distance. When two trainees are close in the manifold space, the spatial distance term in the denominator decreases, and the risk potential energy component increases, reflecting the strong rhythmic pressure at close range. To prevent The tiny positive number introduced to zero is set according to the lower limit of the precision of computer floating-point arithmetic, for example, the smallest positive floating-point value that can be represented; in the normalization process of the risk potential energy, the denominator This represents the set of trainees currently participating in the training. The maximum value of the risk potential energy of all trainees at the same moment; after normalization, the standardized risk index. The values ​​are mapped to the closed interval [0, 1] to provide a unified benchmark for risk levels at different times and for different group sizes. Based on statistical analysis of historical group training data, the temporal distribution of standardized risk indicators for each trainee under normal training scenarios is obtained. The feature values ​​distributed in the steady-state training phase are extracted, and the values ​​of the feature values ​​after being adjusted upwards by a preset safety margin are used as the preset risk threshold. The feature values ​​are the mean of the standardized risk indicators in the normal training phase, and the preset safety margin is set according to the training safety level requirements, with the principle of achieving a balance between sensitivity and false alarm rate. A sliding observation window covering the most recent preset number of action cycles is maintained to record the standardized risk indicators at the end of each cycle. Value; for the window The sequence is subjected to linear regression, and the slope is calculated as an indicator of trend; when the slope is greater than zero and the current period's... When the value exceeds the preset risk threshold, the intervention conditions are determined and an intervention decision is triggered. The preset number is determined by statistical analysis of the duration of precursors to risk events in historical data, based on the principle of being able to filter out random fluctuations while responding promptly to continuous deterioration trends.

[0032] Regarding the action phase difference Perform direction-selective mapping to construct a directed phase action function, the expression of which is: ;in, For the first The student on the first Effective leading phase effect of each student For a unit step function, when When, take 1, when When the time is 0; a phase effect normalization process is introduced, which is expressed as: ;in, This is the normalized directed phase action. To prevent tiny positive numbers with a denominator of zero, the rhythmic influence relationship of each trainee in the group is reconstructed by weighting based on the normalized directed phase action.

[0033] It should be explained that the directional phase action function For the The functionalized expression of the term is used to characterize the first term. The student relative to the first The effective leading phase action of each student; the unit step function In the discrete implementation, this is accomplished through conditional branching. The function returns 1 if the input variable is greater than zero, otherwise it returns 0; only the first condition is retained. The positive leading phase information of the first student relative to other students, when the first... When a student's progress lags behind other students, the corresponding function value is 0, suppressing the interference of the lagging student on risk assessment; in the phase action normalization process, the denominator Indicates the first The sum of the absolute values ​​of the directed phase actions of all other students on each student; after normalization. The numerical value represents the first The student relative to the first The relative proportion of each student's leading phase effect in the total leading phase effect; the To prevent tiny positive numbers with a denominator of zero, a weighted reconstruction is performed based on the normalized directed phase action, with the aim of ensuring that risk accumulation takes into account the relative contribution ratios of each leading participant.

[0034] After determining that the intervention conditions are met, the normalized risk potential is used. As the basis for adjustment, the phase shift of the audio beat pulse is adaptively calculated, and a phase perturbation function is constructed, the expression of which is: ;in, To be applied to the first The phase offset of the audio beat for each student. The phase adjustment amplitude coefficient is greater than zero. This is the phase difference suppression coefficient. For the first The deviation of each student from the group reference phase; based on the phase offset, the standard beat signal is phase-shifted to generate a personalized audio beat pulse signal, expressed as: ;in, The output audio beat signal, The frequency is the beat angular frequency. During the application of the audio beat pulse, a gradual adjustment strategy is introduced. By setting a phase change rate constraint, the phase shift change between adjacent moments is ensured to satisfy the following: ;in The time interval between adjacent control cycles. The upper limit of the phase change rate; monitoring the phase difference of the action within multiple consecutive action cycles. With the energy leakage index The trend of change is such that when the phase difference of the action gradually decreases and the contribution of the group rhythm disturbance drops below the preset stable threshold, the phase offset is gradually reduced until the standard beat signal is restored.

[0035] It should be explained that the phase offset amount This represents the amount of phase shift applied to the standard beat signal at the current moment, used to provide beat prompts to trainees in advance. The larger the value, the more advanced the audio beat is relative to the standard beat, thus providing a stronger rhythmic guidance to the learner; The term makes the phase adjustment amplitude proportional to the normalized risk potential energy; the higher the risk, the greater the adjustment amplitude. The time indicates the output of a standard beat signal, without rhythmic intervention; the exponent term This term describes the characteristic that a learner's sensitivity to changes in beat decreases as phase deviation increases; the exponential term... The calculation logic is as follows: when the first When the deviation of an individual student from the group reference phase is small, When the value is relatively small, the exponent term is close to 1, thus retaining a larger phase adjustment range, allowing trainees to achieve effective rhythm correction even during slight deviations; when the... When the deviation of an individual student from the group reference phase continues to increase, As the phase shift increases, the exponential term gradually decreases to suppress the phase adjustment amplitude, avoiding an excessively large phase shift at once. The phase adjustment strategy employs a progressive correction mechanism because when trainees have already deviated significantly from the group rhythm, continuing to linearly increase the phase adjustment according to the magnitude of the deviation can easily lead to abrupt changes in the beat cues, causing additional cognitive load and secondary fluctuations in the movement rhythm. By introducing an exponential suppression term that decreases with increasing phase deviation, minor deviations are corrected quickly first, while larger deviations are restored to synchronization through a gradual, cycle-by-cycle guidance approach, ensuring the stability and adaptability of the rhythm adjustment process. The group reference phase is the median of the movement progress of all trainees; the phase adjustment amplitude coefficient... Take a positive value, the phase adjustment amplitude coefficient The determination method is as follows: based on the beat adaptability test in historical training data, the maximum phase shift that does not cause discomfort to trainees is used as a benchmark, and the phase difference suppression coefficient is calculated by back-calculating the coefficient in combination with the typical range of risk potential energy; The method for determining the personalized audio beat pulse signal is to analyze the response delay data of trainees to beat changes under different phase deviations and fit an exponential decay curve; The beat angular frequency is generated using a standard sine function. It is derived from the standard rhythm frequency of the training movements; the formula contains This represents the time interval between adjacent control cycles, and its value is consistent with the action sampling frequency and audio control refresh cycle; in the progressive adjustment strategy, the upper limit of the phase change rate... The preset stability threshold is set based on the perceptible threshold of human ear for beat changes, ensuring that changes in phase shift between adjacent beats do not cause abruptness. The mean of the energy leakage index during the trainee's historical normal training phases is statistically analyzed, and the value after adjusting for a preset recovery margin is used as the preset stability threshold. Based on the statistical distribution of the energy leakage index during the trainee's historical normal training phases, the standard deviation of the distribution is used as a reference benchmark for the recovery margin, ensuring that the recovery threshold is slightly higher than the upper limit of the normal fluctuation range. This ensures that the movement quality has stabilized and returned to normal levels before intervention is withdrawn. When the phase difference of the movement is gradually reduced and the energy leakage index drops below the preset stability threshold over multiple consecutive movement cycles, the phase shift is gradually reduced according to a preset decreasing ratio until the standard beat signal is restored.

[0036] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A group physical training control system based on motion capture and AI load matching, characterized in that, include: The 3D motion acquisition and manifold construction module is used to acquire 3D skeletal joint data of each trainee, construct a standard motion manifold space, and map the skeletal sequence of each trainee to the standard motion manifold space to obtain posture deviation and group spatial topology. The motion cycle division and joint feature extraction module is used to divide the motion of each trainee into cycles and extract the motion trajectory and speed change features of each joint within the cycle. The joint energy distribution and concentration calculation module is used to construct the motion energy distribution based on the velocity change characteristics of each joint, determine the energy peak position, calculate the energy proportion within a preset time window, and obtain the energy concentration of each joint. The energy leakage index calculation module is used to calculate the average energy concentration and dispersion based on the energy concentration of each joint, and to calculate the energy leakage index based on the average energy concentration and dispersion. The group relationship and phase feature calculation module is used to determine the relative position of the trainees based on the group spatial topology and to calculate the movement progress to obtain the movement phase difference; The risk assessment and intervention decision-making module is used to construct risk potential energy and determine intervention conditions based on the contribution of group rhythm disturbance, the relative position and the phase difference of the action. The auditory phase perturbation execution module is used to apply phase-leading audio beat pulses to adjust the rhythm of the movement when intervention conditions are met.

2. The group physical training control system based on motion capture and AI load matching according to claim 1, characterized in that, The acquisition of 3D skeletal joint data for each participant includes: acquiring 3D skeletal joint position data for each participant in a group training scenario using a 3D vision acquisition device, and constructing a temporal data set of joints for each participant, represented as follows: ;in, For the first Each student at any moment The set of joint spatial locations For the first The first student's Each joint at any time The three-dimensional spatial position vector, Number the joints. , The total number of joints; the joint timing data set is segmented according to the continuity of movement to determine the interval of each movement cycle, represented as: ;in, For the first The first student's Each action cycle interval The start time of the cycle, To correspond to the duration of the action cycle, within the action cycle interval, the spatial position vectors of each joint are aligned using a unified time reference.

3. The group physical training control system based on motion capture and AI load matching according to claim 1, characterized in that, The process of dividing each student's movements into periods and extracting the motion trajectory and velocity change characteristics of each joint within a period includes: the spatial position vectors of each student's joints. Time series analysis was performed to obtain the continuous motion trajectory of each joint within the motion cycle interval; based on the joint spatial position vector, the velocity change characteristics of each joint between adjacent sampling times were calculated using a discrete difference method. ;in, For the first The first student's Each joint at any time The velocity vector, The time interval between adjacent sampling moments is defined as follows: The velocity vector is smoothed using a sliding window process to obtain a smoothed velocity change sequence, as shown in the formula: ;in, The smoothed velocity variation characteristics, This is the length of the sliding window.

4. The group physical training control system based on motion capture and AI load matching according to claim 1, characterized in that, The process involves constructing a motion energy distribution based on the velocity change characteristics of each joint, determining the energy peak location, calculating the energy percentage within a preset time window, and obtaining the energy concentration of each joint. This includes: based on the smoothed velocity change characteristics of each joint. Construct the motion energy distribution function of each joint in the time dimension, the expression of which is: ;in, For the first The first student's Each joint at any time kinetic energy density, The vector magnitude is represented by: Within the motion cycle interval, extreme value detection is performed on the motion energy distribution function to determine the peak time of motion energy at each joint, expressed as: ;in, For the first The first student's The peak energy time of each joint within the current motion cycle; a preset time window is constructed centered on the peak energy time: ;in, For the first Energy concentration time window for each joint The window width is used to characterize the time range of energy concentration analysis; based on the energy concentration time window and the motion energy distribution function, the energy proportion of each joint within the energy concentration time window is calculated to obtain the energy concentration of each joint.

5. The group physical training control system based on motion capture and AI load matching according to claim 4, characterized in that, The step of calculating the energy proportion of each joint within the energy concentration time window based on the energy concentration time window and the motion energy distribution function to obtain the energy concentration degree of each joint includes: calculating the energy proportion of each joint within the energy concentration time window. The formulas for calculating the window energy and the total energy over the entire action cycle are as follows: ;in, For the first The first student's The cumulative energy value of each joint within the energy concentration time window. For the first The first student's The cumulative energy value of each joint throughout the entire motion cycle; based on the ratio between the window energy and the total energy, the energy concentration of each joint is calculated, expressed as: ;in, For the first The first student's The energy concentration of a joint within the current motion cycle is used to characterize the degree of concentration of the joint's motion energy in the time dimension.

6. The group physical training control system based on motion capture and AI load matching according to claim 1, characterized in that, The calculation of average energy concentration and dispersion based on the energy concentration of each joint, and the calculation of the energy leakage index based on the average energy concentration and dispersion, includes: energy concentration based on each joint. Calculate the average energy concentration of each trainee during the current action cycle: ;in, For the first The average energy concentration of each trainee during the current movement cycle. The total number of joints; based on the deviation between the energy concentration of each joint and the average energy concentration, the dispersion of energy concentration is calculated, expressed as: ;in, For the first The dispersion of energy concentration at each joint of each trainee; based on the average energy concentration and the dispersion, an energy leakage index is constructed to indirectly quantify the contribution of the group rhythm disturbance, the formula is: ;in, For the first The energy leakage index of each trainee during the current action cycle. The proportional coefficient used to adjust the weighting of dispersion is used to comprehensively characterize the dispersion of joint motion energy in the time dimension and the degree of inconsistency between joints; the contribution of group rhythm perturbation is defined. The contribution of group rhythm disturbance is a latent variable used to characterize the degree of influence of an individual learner on the overall rhythm stability. There is a mapping relationship between the energy leakage index and the degree of dispersion of joint motion energy over time and the coordination consistency between joints. As the contribution of the group rhythm disturbance The only computable indirect representation of , namely: .

7. The group physical training control system based on motion capture and AI load matching according to claim 1, characterized in that, The step of determining the relative positions of trainees based on the group's spatial topology and calculating the movement progress to obtain the movement phase difference includes: extracting the spatial topology of each trainee in the group by using the embedding position of each trainee's skeletal sequence in the standard movement manifold space; and determining the relative positional relationship between any two trainees based on the adjacency relationship of each trainee in the manifold space, expressed as: ;in, For the first The student and the first Each student at any moment spatial distance For the first The position representation of each trainee in the standard motion manifold space; the time mapping of the joint timing data set is normalized, and the motion progress function is constructed, with the expression: ;in, For the first The percentage of movement completed by each trainee within the current movement cycle; by comparing the differences in movement progress among different trainees, the movement phase difference is calculated using the following formula: ;in, For the first The student and the first Each student at any moment The phase difference of the movements is used to reflect the sequential relationship between trainees in the rhythm of movement execution.

8. The group physical training control system based on motion capture and AI load matching according to claim 1, characterized in that, The process of constructing risk potential energy and determining intervention conditions based on the energy leakage index, relative position, and action phase difference includes: for each participant within the same action cycle, comprehensively considering the contribution of the group rhythm disturbance. Spatial distance between trainees and motion phase difference A risk potential function reflecting the collaborative risk state of the group is constructed. By coupling the individual abnormality degree represented by the contribution of the group rhythm disturbance with the group interaction relationship, the risk potential value of any student is obtained: ;in, For the first Each student at any moment The risk potential value, To prevent tiny positive numbers with a denominator of zero; Used to extract the first The degree to which a student's movements are ahead of other students is only when the first student... When a participant is ahead of others in terms of progress, they participate in risk accumulation; the risk potential energy of all participants at the same moment is normalized to obtain a standardized risk index, expressed as: ;in, For the normalized risk potential energy, This represents the current set of trainees participating in the training. Based on this, by comparing the normalized risk potential with the preset risk threshold and combining the changing trend within multiple consecutive action cycles, the satisfaction of the intervention conditions is determined. When the normalized risk potential continues to rise and exceeds the preset risk threshold, an intervention decision is triggered.

9. The group physical training control system based on motion capture and AI load matching according to claim 8, characterized in that, The Used to extract the first The degree to which a student's movements are ahead of other students is only when the first student... When a participant is ahead of other participants in the progress of a movement, they participate in risk accumulation, including: the phase difference of the movement. Perform direction-selective mapping to construct a directed phase action function, the expression of which is: ;in, For the first The student on the first Effective leading phase effect of each student For a unit step function, when When, take 1, when When the time is 0; a phase effect normalization process is introduced, which is expressed as: ;in, This is the normalized directed phase action. To prevent tiny positive numbers with a denominator of zero, the rhythmic influence relationship of each trainee in the group is reconstructed by weighting based on the normalized directed phase action.

10. The group physical training control system based on motion capture and AI load matching according to claim 1, characterized in that, The method for applying a phase-leading audio beat pulse to adjust the movement rhythm when intervention conditions are met includes: after determining that intervention conditions are met, using the normalized risk potential... As the basis for adjustment, the phase shift of the audio beat pulse is adaptively calculated, and a phase perturbation function is constructed, the expression of which is: ;in, To be applied to the first The phase offset of the audio beat for each student. The phase adjustment amplitude coefficient is greater than zero. This is the phase difference suppression coefficient. For the first The deviation of each student from the group reference phase; based on the phase offset, the standard beat signal is phase-shifted to generate a personalized audio beat pulse signal, expressed as: ;in, The output audio beat signal, The frequency is the beat angular frequency. During the application of the audio beat pulse, a gradual adjustment strategy is introduced. By setting a phase change rate constraint, the phase shift change between adjacent moments is ensured to satisfy the following: ;in The time interval between adjacent control cycles. The upper limit of the phase change rate; monitoring the phase difference of the action within multiple consecutive action cycles. With the energy leakage index The trend of change is such that when the phase difference of the action gradually decreases and the contribution of the group rhythm disturbance drops below the preset stable threshold, the phase offset is gradually reduced until the standard beat signal is restored.

Images