A Construction Method of a Fitness Training Evaluation and Decision-Making Model Based on VR Virtual Digital Humans

Abstract

The present invention relates to the field of computer technology, and in particular to a method for constructing a fitness training evaluation and decision-making model based on a VR virtual digital human. The method includes the following steps: performing motion capture on the practice actions of an expert through a high-precision inertial camera group network to obtain expert action data, where the expert action data includes joint motion angle data and muscle activity potential data; analyzing the expert action data for joint torque, tendon transmission ratio, and inertial force based on a rigid body dynamics model and musculoskeletal simulation to obtain biomechanical parameters; importing the expert action data and biomechanical parameters into a virtual reality engine, and generating virtual digital human action demonstrations through bone driving and inverse kinematics to obtain virtual action demonstration data. The present invention integrates biomechanical models such as a rigid body dynamics model and a musculoskeletal model, quantitatively calculates biomechanical parameters such as joint torque, tendon transmission ratio, and inertial force, and makes the simulation more realistic.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and particularly to a method for constructing a fitness training evaluation and decision-making model based on a VR virtual digital human. Background Art

[0002] With the rapid development of virtual reality (VR), augmented reality (AR), artificial intelligence (AI), and motion capture technologies, people have gradually started to apply these technologies to the field of fitness and sports training. Traditional fitness training evaluation methods often rely on manual guidance and experience accumulation, lacking objective and accurate evaluation means; this method is not only inefficient but also prone to evaluation errors, affecting the training effect. Against this background, a fitness training evaluation and decision-making model based on a VR virtual digital human has emerged, aiming to use advanced technical means to accurately evaluate and guide fitness training, thereby improving the training effect.

[0003] However, there are still some problems in the current method for constructing a fitness training evaluation and decision-making model based on a VR virtual digital human: there are significant differences among different individuals in terms of body shape, flexibility, cardiopulmonary function, etc., and the difficulty levels of traditional exercise movements are also different for different individuals; how to evaluate the physical characteristics of individuals and provide targeted exercise suggestions according to individual differences is an important problem to be solved. Traditional exercise movements usually involve fine and smooth body movements, and it is necessary to accurately capture and model the movement data of experts; this poses high requirements for the accuracy of motion capture devices and the robustness of motion recognition algorithms. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method for constructing a fitness training evaluation and decision-making model based on a VR virtual digital human to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for constructing a fitness training evaluation and decision-making model based on a VR virtual digital human includes the following steps:

[0006] Step S1: Perform motion capture on the practice movements of an expert through a high-precision inertial camera group network to obtain expert motion data, where the expert motion data includes joint motion angle data and muscle activity potential data;

[0007] Step S2: Analyze the joint torque, tendon transmission ratio, and inertial force of the expert motion data based on the rigid body dynamics model and musculoskeletal simulation to obtain biomechanical parameters;

[0008] Step S3: Import the expert motion data and biomechanical parameters into a virtual reality engine, and generate virtual digital human motion demonstrations through bone driving and inverse kinematics to obtain virtual motion demonstration data;

[0009] Step S4: Use computer vision technology to enhance the feature expression of key joint points in the pre-acquired student action video data, and identify the exercise actions of the student action video data based on the virtual action demonstration data, so as to obtain the student motion pattern data;

[0010] Step S5: Analyze the human biological parameters of the student action video data to obtain the student biological parameters; perform differential metric analysis on the student motion pattern data based on the virtual action demonstration data to obtain action differential metric data; perform action adjustment based on the biomechanical model according to the student biological parameters and the action differential metric data, and generate training evaluation report data; construct a training evaluation decision model based on the decision tree model according to the training evaluation report data.

[0011] The present invention can accurately obtain the motion data of experts, including joint motion angles and muscle activity potential data, which are the basis for subsequent analysis and simulation. Through biomechanical analysis, parameters such as joint torque, tendon transmission ratio, and inertial force can be obtained, which are very important for subsequent action generation and evaluation and can provide specific biomechanical guidance. The virtual action demonstration data presents the simulation results of the expert actions through a virtual reality engine. This visualization method can more intuitively display the action skills of experts and provide specific demonstrations and references for students. Through computer vision technology, the feature expression of key joint points can be extracted from the student's action video, enhancing the understanding and analysis of the student's actions. By comparing with the virtual action demonstration data, the identification of exercise actions and the extraction of the student's motion pattern data can be carried out. By analyzing the human biological parameters of the student action video data, the student's biological parameters such as joint motion range and muscle activity degree can be obtained; based on the differential metric analysis of the virtual action demonstration data and the student motion pattern data, the difference between the student's actions and the expert's actions can be evaluated; according to the student's biological parameters and the action differential metric data, action adjustment based on the biomechanical model can be carried out to provide targeted training suggestions and optimization schemes; finally, based on the training evaluation report data, a training evaluation decision model based on the decision tree model can be constructed to automatically evaluate and make decisions on the student's training performance, providing guidance and feedback for further training. In summary, the effects of the above steps include: providing expert action data and biomechanical parameters as the basis for subsequent analysis and simulation; intuitively displaying the expert action skills through the virtual action demonstration data to provide references and demonstrations for students; using computer vision technology and virtual action demonstration data to analyze and identify the student's actions and extract the student's motion pattern data; analyzing the student's human biological parameters and action differential metric data, performing action adjustment based on the biomechanical model, and optimizing the training plan; constructing a training evaluation decision model based on the decision tree model to automatically evaluate the student's performance and provide guidance and feedback. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Other features, objects, and advantages of the present invention will become more apparent by reading the following detailed description of non - restrictive embodiments with reference to the accompanying drawings:

[0013] Figure 1 It is a schematic flowchart of the steps for the construction method of the fitness training evaluation and decision - making model based on VR virtual digital humans of the present invention;

[0014] Figure 2 is Figure 1 a detailed flowchart of step S1 in

[0015] Figure 3 is Figure 1 a detailed flowchart of step S2 in DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] The following clearly and completely describes the technical method of the present invention with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art within the scope of the present invention without creative work belong to the scope of protection of the present invention.

[0017] In addition, the drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0018] It should be understood that although terms such as "first", "second", etc. may be used here to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed related items.

[0019] To achieve the above - mentioned purpose, please refer to Figures 1 to 3 , the present invention provides a construction method of a fitness training evaluation and decision - making model based on VR virtual digital humans, and the method includes the following steps:

[0020] Step S1: Use a high-precision inertial camera group network to perform motion capture on the expert's practice actions, obtaining expert action data, where the expert action data includes joint motion angle data and muscle activity potential data;

[0021] Step S2: Based on the rigid body dynamics model and musculoskeletal simulation, analyze the expert action data for joint torque, tendon transmission ratio, and inertial force to obtain biomechanical parameters;

[0022] Step S3: Import the expert action data and biomechanical parameters into the virtual reality engine, and generate virtual digital human action demonstrations through bone driving and inverse kinematics to obtain virtual action demonstration data;

[0023] Step S4: Use computer vision technology to enhance the feature expression of key joint points in the pre-acquired trainee action video data, and identify the exercise actions in the trainee action video data according to the virtual action demonstration data, thereby obtaining trainee motion pattern data;

[0024] Step S5: Analyze the trainee's body biological parameters from the trainee action video data to obtain trainee biological parameters; perform differential metric analysis on the trainee motion pattern data according to the virtual action demonstration data to obtain action differential metric data; perform action adjustment based on the biomechanical model according to the trainee biological parameters and action differential metric data, and generate training evaluation report data; construct a training evaluation decision model based on the decision tree model according to the training evaluation report data.

[0025] In the embodiment of the present invention, refer to Figure 1 As described, it is a schematic diagram of the step process of a method for constructing a fitness training evaluation decision model based on a VR virtual digital human. In this example, the method for constructing a fitness training evaluation decision model based on a VR virtual digital human includes the following steps:

[0026] Step S1: Use a high-precision inertial camera group network to perform motion capture on the expert's practice actions, obtaining expert action data, where the expert action data includes joint motion angle data and muscle activity potential data;

[0027] In the embodiment of the present invention, a high-precision inertial camera group network is used for motion capture. This network usually includes multiple inertial sensors, such as accelerometers and gyroscopes, which can accurately measure the position, attitude, and motion of an object; at the same time, muscle activity potential sensors are used to capture the electrical signals of muscles to obtain muscle activity data.

[0028] Step S2: Based on the rigid body dynamics model and musculoskeletal simulation, analyze the expert action data for joint torque, tendon transmission ratio, and inertial force to obtain biomechanical parameters;

[0029] In the embodiments of the present invention, a rigid body dynamics model and musculoskeletal simulation technology are used to analyze expert action data. These models and simulation methods can calculate biomechanical parameters such as joint torque, tendon transmission ratio, and inertial force based on factors such as joint angle, muscle activity, and object mass.

[0030] Step S3: Import the expert action data and biomechanical parameters into a virtual reality engine, and generate a virtual digital human action demonstration through bone driving and inverse kinematics to obtain virtual action demonstration data.

[0031] In the embodiments of the present invention, a virtual reality engine such as Unity or Unreal Engine is used to import the expert action data and biomechanical parameters, and the bone driving and inverse kinematics algorithms are used to generate the action demonstration of the virtual digital human; these engines provide powerful graphics rendering and animation generation functions, capable of presenting a realistic virtual action demonstration.

[0032] Step S4: Use computer vision technology to enhance the feature expression of key joint points in the pre-acquired trainee action video data, and identify the exercise actions in the trainee action video data according to the virtual action demonstration data, so as to obtain the trainee motion pattern data.

[0033] In the embodiments of the present invention, computer vision technology is used to analyze the trainee action video data. A deep learning model such as a convolutional neural network (CNN) can be used to extract and enhance the feature expression of key joint points in the video frames; at the same time, with the help of the virtual action demonstration data, the exercise actions in the trainee's action video can be identified based on the comparison and matching method, so as to obtain the trainee's motion pattern data.

[0034] Step S5: Analyze the human biological parameters of the trainee action video data to obtain the trainee biological parameters; perform differential metric analysis on the trainee motion pattern data according to the virtual action demonstration data to obtain the action differential metric data; perform action adjustment based on the biomechanical model according to the trainee biological parameters and the action differential metric data, and generate training evaluation report data; construct a training evaluation decision model based on the decision tree model according to the training evaluation report data.

[0035] In the embodiments of the present invention, human biological parameter analysis technology is used to process the action video data of trainees, and biological parameters of trainees can be extracted, such as information on joint angles, muscle activities, etc.; meanwhile, using virtual action demonstration data and trainee movement pattern data, differential measurement analysis can be carried out to compare the differences between the actions of trainees and those of experts. Based on the biomechanical model, combining the biological parameters of trainees and the action differential measurement data, action adjustment can be performed, that is, corresponding guidance and training adjustments are made according to the physical characteristics and action differences of trainees; finally, training evaluation report data is generated, including the training results and suggestions of trainees. A decision tree model can be used to analyze and make decisions on the training evaluation report data to construct a training evaluation decision model.

[0036] The present invention can accurately obtain the movement data of experts, including joint movement angles and muscle activity potential data, which are the basis for subsequent analysis and simulation. Through biomechanical analysis, parameters such as joint torque, tendon transmission ratio, and inertial force can be obtained, and these parameters are very important for subsequent action generation and evaluation and can provide specific biomechanical guidance. The virtual action demonstration data presents the simulation results of experts' actions through a virtual reality engine. This visualization method can more intuitively display the action skills of experts and provide specific demonstrations and references for trainees. Through computer vision technology, the feature expressions of key joint points can be extracted from the action videos of trainees, enhancing the understanding and analysis of trainees' actions. By comparing with the virtual action demonstration data, the identification of exercise actions and the extraction of trainee movement pattern data can be carried out. Through human biological parameter analysis of the trainee action video data, the biological parameters of trainees can be obtained, such as joint movement range, muscle activity degree, etc.; based on the differential measurement analysis of the virtual action demonstration data and trainee movement pattern data, the differences between the actions of trainees and those of experts can be evaluated; according to the biological parameters of trainees and the action differential measurement data, action adjustment based on the biomechanical model can be performed to provide targeted training suggestions and optimization plans; finally, based on the training evaluation report data, a training evaluation decision model based on the decision tree model can be constructed to automatically evaluate and make decisions on the training performance of trainees and provide guidance and feedback for further training. In summary, the effects of the above steps include: providing expert action data and biomechanical parameters as the basis for subsequent analysis and simulation; intuitively displaying expert action skills through virtual action demonstration data to provide references and demonstrations for trainees; using computer vision technology and virtual action demonstration data to analyze and identify trainees' actions and extract trainee movement pattern data; analyzing the human biological parameters and action differential measurement data of trainees to perform action adjustment based on the biomechanical model and optimize the training plan; constructing a training evaluation decision model based on the decision tree model to automatically evaluate trainees' performance and provide guidance and feedback.

[0037] Preferably, step S1 includes the following steps:

[0038] Step S11: Use textureless blue - green as the background, arrange multiple high - precision inertial capture cameras for full - view coverage, and obtain a high - precision inertial camera group network;

[0039] Step S12: Paste reflective marker points on the key parts of the expert's body and wear surface electromyogram sensors. Through the expert's complete demonstration of the exercise action process in the inertial camera group network, collect the original electromyogram data and original 3D position data of the marker points;

[0040] Step S13: Eliminate noise and missing data from the original electromyogram data and original 3D position data, and synchronize the time stamps of the position and electromyogram data, so as to obtain muscle activity data and 3D marker point position data;

[0041] Step S14: Use biomechanical modeling software to construct a bone model based on the muscle activity data and 3D marker point position data, so as to obtain expert bone model data;

[0042] Step S15: Bind the 3D marker point position data to the expert bone model data, calculate the joint movement angles, so as to obtain joint movement angle data; Estimate the muscle activity level and action intensity according to the muscle activity data, so as to obtain muscle activity potential data;

[0043] Step S16: Store the joint movement angle data and muscle activity potential data as expert action data.

[0044] As an embodiment of the present invention, referring to Figure 2 shown, for Figure 1 the detailed step - by - step flow diagram of step S1 in

[0045] Step S11: Use textureless blue - green as the background, arrange multiple high - precision inertial capture cameras for full - view coverage, and obtain a high - precision inertial camera group network;

[0046] The embodiment of the present invention uses multiple high - precision inertial capture cameras. These cameras usually have high resolution and fast sampling rate, can accurately capture the position, attitude and movement of objects, arrange these cameras to achieve full - view coverage, and ensure that the background is textureless blue - green for background separation in subsequent processing.

[0047] Step S12: Paste reflective marker points on the key parts of the expert's body and wear surface electromyogram sensors. Through the expert's complete demonstration of the exercise action process in the inertial camera group network, collect the original electromyogram data and original 3D position data of the marker points;

[0048] Embodiment of the present invention: Reflective marker points are pasted on the key parts of the expert's body. These marker points usually have high reflectivity so that the camera can accurately capture their positions. At the same time, the expert also wears surface electromyography sensors to record the electrical signals of muscle activities. The expert conducts a complete demonstration in the inertial camera group network. The camera will capture the positions and postures of the marker points and record the electrical signals of the electromyography sensors.

[0049] Step S13: Noise and missing data elimination are performed on the original electromyography data and the original 3D position data, and timestamp synchronization is performed on the position and electromyography data, so as to obtain muscle activity data and 3D marker point position data.

[0050] The embodiment of the present invention uses signal processing technology to perform noise elimination and filtering operations on the original electromyography data to remove interference and extract effective muscle activity signals; the original 3D position data is processed to remove possible noise and missing data. Through timestamp synchronization, the electromyography data and the position data are aligned to ensure that they are consistent in time.

[0051] Step S14: A bone model is constructed according to the muscle activity data and the 3D marker point position data through biomechanical modeling software, so as to obtain expert bone model data.

[0052] The embodiment of the present invention uses biomechanical modeling software, such as AnyBody or OpenSim, to construct the expert's bone model according to the muscle activity data and the 3D marker point position data. These software provide bone modeling and simulation functions, and can calculate the posture and dynamic characteristics of the bone according to muscle activities and joint positions.

[0053] Step S15: The 3D marker point position data is bound to the expert bone model data, and joint movement angle calculation is performed to obtain joint movement angle data; according to the muscle activity data, the muscle activity level and action intensity are estimated to obtain muscle activity potential data.

[0054] The embodiment of the present invention associates and binds the 3D marker point position data with the expert bone model data to realize the movement of the bone model. By calculating the angle changes of each joint in the bone model, joint movement angle data can be obtained; according to the muscle activity data, the muscle activity level and action intensity can be estimated to obtain muscle activity potential data.

[0055] Step S16: The joint movement angle data and the muscle activity potential data are stored as expert action data.

[0056] In the embodiments of the present invention, the calculated joint motion angle data and muscle activity potential data are saved as expert action data; these data can be stored in an appropriate format, such as a CSV (Comma-Separated Values) file or a database, and the stored data can be used for subsequent analysis, research, or applications.

[0057] The present invention uses a network composed of a textureless blue-green background and multiple high-precision inertial cameras, which can provide full-view coverage and reduce background interference, so as to obtain high-quality and relatively accurate expert action data. By pasting reflective marker points on the key parts of the expert's body and wearing surface electromyography sensors, the detailed information of the expert's actions can be recorded; the acquisition of the original electromyography data and the original 3D position data can provide more comprehensive and accurate expert action data. By performing noise elimination and missing data processing on the original electromyography data and the original 3D position data, the quality and accuracy of the data can be improved; the synchronization of timestamps can ensure the consistency and correspondence between the muscle activity data and the 3D marker point position data. Through biomechanical modeling software, a skeletal model of the expert can be constructed based on the muscle activity data and the 3D marker point position data; this provides an accurate skeletal structure for subsequent calculation of joint motion angles and estimation of muscle activity potential data. By binding the 3D marker point position data with the expert skeletal model data, the joint motion angle data can be calculated; this provides an accurate measure of joint motion for analyzing the expert's action skills; at the same time, based on the muscle activity data, the active level and action intensity of the muscles can be estimated to obtain the muscle activity potential data. Saving the joint motion angle data and the muscle activity potential data as expert action data provides the detailed information and characteristics of the expert's actions, and these data will be used in subsequent analysis, research, and applications, such as fields like imitation, evaluation, and optimization of expert actions. Generally speaking, the effects of the above steps include: providing high-quality and relatively accurate expert action data; capturing the detailed information of the expert's actions, including muscle activity data and joint motion angle data; removing noise and missing data in the data to improve the quality and accuracy of the data; establishing a skeletal model of the expert to provide an accurate anatomical structure for subsequent analysis and research; estimating the active level and action intensity of the muscles to provide a deeper understanding of muscle activity; storing the expert action data to provide a basis for subsequent analysis, research, and applications. These effects can provide more accurate, detailed, and comprehensive data support for the research, evaluation, optimization, and application of expert actions, and promote the development and application of related fields.

[0058] Preferably, step S2 includes the following steps:

[0059] Step S21: Construct a rigid body chain model and a muscle model according to the expert action data; map the muscle activity potential data to the activity parameters of the muscle model to obtain a parameterized muscle model;

[0060] Step S22: Use the joint motion angle data as the motion input to drive the rigid body chain model to move, and solve the forces and torques of each joint based on Newton's second law to obtain the torque values of each joint at each time step, thereby obtaining the joint torque data;

[0061] Step S23: Use the joint torque data as the input to perform an inverse solution of the muscle contraction force on the parameterized muscle model, thereby obtaining the muscle contraction force data;

[0062] Step S24: Calculate the torque acting on the bone based on the muscle attachment point position according to the muscle contraction force data, thereby obtaining the bone torque data; Calculate the tendon transmission ratio data based on the bone torque data and the muscle contraction force data, where the tendon transmission ratio = bone torque / muscle contraction force;

[0063] Step S25: Decompose the joint motion angle data into the motion of each rigid body according to the rigid body chain model, and calculate the linear acceleration and angular acceleration based on the rigid body, thereby obtaining the rigid body motion parameter data;

[0064] Step S26: Derive the inertial forces and balance torques of each part of the body based on Newton's laws of motion for the rigid body motion parameter data, thereby obtaining the inertial force data;

[0065] Step S27: Perform biomechanical parameter analysis based on the joint torque data, tendon transmission ratio data, and inertial force data, thereby obtaining the biomechanical parameters.

[0066] As an embodiment of the present invention, refer to Figure 3 shown in Figure 1 is a detailed step flow diagram of step S2 in

[0067] Step S21: Construct a rigid body chain model and a muscle model according to the expert motion data; Map the muscle action potential data to the activity parameters of the muscle model, thereby obtaining the parameterized muscle model;

[0068] In the embodiment of the present invention, a biomechanical modeling software, such as AnyBody or OpenSim, is used to construct a rigid body chain model and a muscle model according to the expert motion data. The rigid body chain model represents the structural relationship of each part of the body (such as bones, joints), while the muscle model represents the position, shape, and mechanical properties of the muscles. By mapping the muscle action potential data to the activity parameters of the muscle model, the contraction and relaxation processes of the muscles can be simulated.

[0069] Step S22: Use the joint motion angle data as the motion input to drive the rigid body chain model to move, and solve the forces and torques of each joint based on Newton's second law to obtain the torque values of each joint at each time step, thereby obtaining the joint torque data;

[0070] In the embodiments of the present invention, a rigid body chain model and joint motion angle data are used to calculate the forces and torques of each joint through Newton's second law (force equals mass times acceleration); the joint motion angle data is used as a motion input to drive the motion of the rigid body chain model, and then by solving the Newton's second law equation, the torque values of each joint at each time step are obtained, thereby obtaining joint torque data.

[0071] Step S23: Using the joint torque data as an input, perform an inverse solution of the muscle contraction force on the parameterized muscle model, thereby obtaining muscle contraction force data;

[0072] In the embodiments of the present invention, a parameterized muscle model and joint torque data are used to inversely solve the joint torque to the muscle contraction force through reverse calculation; this process involves the mechanical properties of muscles and the mechanical relationship of the muscle-skeleton system.

[0073] Step S24: Calculate the torque acting on the bone based on the muscle attachment point position according to the muscle contraction force data, thereby obtaining bone torque data; calculate the tendon transmission ratio data according to the bone torque data and the muscle contraction force data, where the tendon transmission ratio = bone torque / muscle contraction force;

[0074] In the embodiments of the present invention, the muscle contraction force data and the position information of the muscle attachment points in the muscle model are used to calculate the torque acting on the bone; according to the bone torque data and the muscle contraction force data, the tendon transmission ratio can be calculated, and this ratio represents the transmission efficiency of the muscle torque on the bone.

[0075] Step S25: Decompose the joint motion angle data into the motion of each rigid body according to the rigid body chain model, and calculate the linear acceleration and angular acceleration based on the rigid body, thereby obtaining rigid body motion parameter data;

[0076] In the embodiments of the present invention, a rigid body chain model and joint motion angle data are used to decompose the joint motion angle into the motion of each rigid body; according to the motion of the rigid body, the linear acceleration and angular acceleration can be calculated, and these parameters describe the motion state of the rigid body.

[0077] Step S26: Derive the inertial forces and equilibrium torques of each part of the body based on Newton's laws of motion for the rigid body motion parameter data, thereby obtaining inertial force data;

[0078] In the embodiments of the present invention, Newton's laws of motion are used, combined with the rigid body motion parameter data and information such as the mass and inertia matrix of the rigid body, to derive the inertial forces and equilibrium torques of each part of the body; these forces and torques are caused by the motion and inertial characteristics of the rigid body.

[0079] Step S27: Perform biomechanical parameter analysis based on joint torque data, tendon transmission ratio data, and inertial force data to obtain biomechanical parameters.

[0080] In the embodiment of the present invention, joint torque data, tendon transmission ratio data, and inertial force data are used to perform biomechanical parameter analysis; this analysis process involves the balance of moments, the transmission efficiency of moments, the distribution of muscle forces, etc., to obtain the evaluation results of biomechanical parameters.

[0081] In the present invention, by constructing a rigid body chain model and a muscle model, the bone and muscle structures of the human body can be simulated, providing a basis for subsequent analysis and calculation. Mapping the muscle activity potential data to the activity parameters of the muscle model can quantify the activity level of the muscle and convert the electrical signal into the contraction degree of the muscle. By using the joint movement angle data as input, the rigid body chain model can be driven to simulate the movement of the human body. By applying Newton's second law, the forces and moments of each joint can be calculated to provide joint torque data; these data can be used to analyze the load and movement characteristics of the joints. By using the joint torque data as input, the muscle contraction force of the parameterized muscle model can be inversely solved, so that the contraction degree and acting force of the muscle in a specific action can be inferred to obtain muscle contraction force data, which helps to study the mechanical behavior and activity characteristics of the muscle. By calculating the moment acting on the bone according to the muscle contraction force data, the influence and load of the muscle on the bone can be understood; at the same time, by calculating the tendon transmission ratio, the transmission efficiency of muscle force in bone movement can be evaluated, and these data help to understand the interaction relationship between muscle and bone. Through the rigid body chain model, the joint movement angle data can be decomposed into the movement of each rigid body. Based on the calculation of the linear acceleration and angular acceleration of the rigid body, detailed parameter data of the rigid body movement can be provided, which helps to analyze and understand the movement characteristics and coordination of each part of the body. By applying Newton's laws of motion, the inertial forces and balance moments of each part of the body can be deduced, and these data help to understand the inertial influence on the body during movement and the moments required to maintain balance. The analysis of inertial forces can help researchers better understand the dynamic characteristics of human movement. By analyzing the joint torque data, tendon transmission ratio data, and inertial force data, biomechanical parameter analysis can be performed. These parameters include joint moments, muscle forces, mechanical properties of the muscle-skeleton system, etc. These biomechanical parameters are of great significance for studying human movement, evaluating exercise load, optimizing training programs, etc. In summary, by executing the above steps, multiple important parameters regarding human movement can be obtained, including muscle activity level, joint torque, muscle contraction force, bone moment, tendon transmission ratio, rigid body movement parameters, inertial force, etc. These parameter data have important application values for understanding and analyzing human biomechanical characteristics, movement mechanisms, and optimizing exercise training and rehabilitation programs.

[0082] Preferably, step S27 includes the following steps:

[0083] Step S271: Integrate the joint torque data, tendon transmission ratio data, and inertial force data into an initial biomechanical parameter set;

[0084] In the embodiment of the present invention, according to the data format and storage method, a new data set is created using data processing software (such as MATLAB or Python), which contains the numerical values of joint torque, tendon transmission ratio, and inertial force; the collected data is integrated into the data set to ensure the correct matching of the data with the corresponding joints and actions.

[0085] Step S272: Calculate the muscle contraction work according to the joint torque data to obtain muscle contraction work data; calculate the lost work done by the inertial force according to the inertial force data to obtain inertial lost work data;

[0086] In the embodiment of the present invention, mathematical calculation methods are used to calculate the muscle contraction work. According to the joint torque data, the muscle contraction work can be obtained by integrating the joint torque and the velocity of the joint angle, and numerical integration methods (such as the trapezoidal rule or Simpson's rule) can be used for the integration calculation. Mathematical calculation methods are used to calculate the inertial lost work. According to the inertial force data, the inertial lost work can be obtained by calculating the dot product of the inertial force and the linear velocity of the rigid body.

[0087] Step S273: Calculate the total energy during the movement process according to the muscle contraction work data and the inertial lost work data to obtain energy consumption data, where energy consumption = muscle contraction work + inertial lost work;

[0088] In the embodiment of the present invention, the muscle contraction work data and the inertial lost work data are added together to obtain the total energy consumption data during the movement process. Data addition operations can be performed using mathematical operation software or programming languages (such as MATLAB or Python) to obtain the energy consumption data.

[0089] Step S274: Integrate the joint torque data, tendon transmission ratio data, inertial force data, and energy consumption data into biomechanical parameters.

[0090] In the embodiment of the present invention, a new data set is created to integrate the joint torque data, tendon transmission ratio data, inertial force data, and energy consumption data; these data are integrated into the data set according to the corresponding format to ensure the correspondence and consistency of the data; finally, the integrated biomechanical parameter data can be saved in an appropriate file format using data processing software or programming languages for subsequent analysis and evaluation.

[0091] In the present invention, joint torque data, tendon transmission ratio data, and inertial force data are integrated into an initial biomechanical parameter set, integrating data from different sources. In this way, joint torque, muscle strength, tendon transmission efficiency, and inertial effects during body movement can be comprehensively considered, providing comprehensive biomechanical parameters for subsequent analysis. By calculating the joint torque data, the work done by the muscle during movement, i.e., muscle contraction work, can be derived. This parameter reflects the energy output generated by the muscle during movement. At the same time, by calculating the work done by the inertial force, i.e., inertial loss work, the energy loss generated by the body due to inertia during movement can be understood. By adding the muscle contraction work data and the inertial loss work data, the total energy consumption data during the movement can be obtained. This parameter can be used to evaluate the energy consumed by the human body during a specific movement, thereby better understanding information such as exercise intensity, training load, and energy balance. By integrating joint torque data, tendon transmission ratio data, inertial force data, and energy consumption data into biomechanical parameters, a comprehensive parameter set can be established, covering information such as joint torque, muscle strength, tendon transmission efficiency, inertial effects, and energy consumption. This parameter set can be used to study human movement characteristics, evaluate exercise load and energy consumption, and optimize training and rehabilitation programs. In summary, by performing the above steps, more comprehensive and detailed biomechanical parameters of human movement can be obtained, including muscle contraction work, inertial loss work, and energy consumption. These parameters have important application values for studying the human movement mechanism, evaluating exercise load and energy consumption, and optimizing training and rehabilitation programs. Integrating these parameters can provide a deeper understanding and analysis, providing useful information and guidance for research and practice in the field of sports science.

[0092] Preferably, step S3 includes the following steps:

[0093] Step S31: Import the rigid body chain model and the parameterized muscle model into the virtual reality engine to obtain a virtual human model;

[0094] In the embodiment of the present invention, data of the rigid body chain model and the parameterized muscle model are prepared. The rigid body chain model describes the human skeletal structure, while the parameterized muscle model describes the shape and movement characteristics of the muscle. Using a virtual reality engine (such as Unity, Unreal Engine, etc.) as a development tool, import the data of the rigid body chain model and the parameterized muscle model into the virtual reality engine; adjust the scale and posture of the virtual human model to match the real human body.

[0095] Step S32: Perform data mapping on the virtual human model according to the expert action data and drive the action of the human model to obtain an action-driven human model;

[0096] Embodiments of the present invention collect expert motion data, which describe the skeletal movements of a real human body during specific actions; use data processing software (such as MotionBuilder, Blender, etc.) for data mapping, and apply the expert motion data to the skeletal structure of a virtual human model; drive the virtual human model according to the mapped skeletal animation data to make it simulate the movements in the expert motion data.

[0097] Step S33: Generate a high-fidelity skin mesh model based on the skeleton for the motion-driven human model, so as to obtain a virtual human motion model with a skin mesh.

[0098] Embodiments of the present invention use 3D modeling software (such as Maya, 3ds Max, etc.) to create a mesh model with high-fidelity skin, and the shape and appearance of this model are similar to those of a real human body; apply the skeletal animation to the mesh model with high-fidelity skin to achieve a skin deformation effect that matches the motion-driven human model; ensure the correct matching between the skin mesh and the skeletal animation to guarantee the verisimilitude of the virtual human motion model.

[0099] Step S34: Initially implement human motion according to the virtual human motion model through skeletal animation technology, so as to obtain a virtual action sequence.

[0100] Embodiments of the present invention use the skeletal animation technology in a virtual reality engine to implement the motion of a virtual human according to the skeletal animation data of the motion-driven human model. According to requirements and scenarios, specific action sequences can be defined, including different types of actions such as standing, walking, and running.

[0101] Step S35: Perform detail motion smoothing based on inverse kinematics constraints according to the virtual action sequence and biomechanical parameters, so as to obtain an optimized virtual action sequence.

[0102] Embodiments of the present invention use an inverse kinematics algorithm to calculate the appropriate poses of joints according to the key frames and target positions in the virtual action sequence; apply inverse kinematics constraints to ensure that the joints do not exceed their feasible ranges during motion and have natural motion smoothness; perform detail motion smoothing processing to better simulate the motion details of a real human body.

[0103] Step S36: Perform muscle motion deformation simulation based on biological parameters for the optimized virtual action sequence, so as to obtain virtual action demonstration data with muscle deformation.

[0104] Embodiments of the present invention use a biomechanical parameter dataset, which includes parameters related to muscle strength, contraction characteristics, and muscle deformation; according to the virtual action optimization sequence and biomechanical parameters, simulate the movement and deformation of muscles; through the application of mathematical models and algorithms, calculate and apply muscle deformation on the skin mesh of the virtual human model; according to the movement of the model and muscle deformation, generate virtual action demonstration data with muscle deformation.

[0105] In the present invention, by importing a rigid body chain model and a parameterized muscle model into a virtual reality engine, a virtual human model can be created. This model can be used as a basis for subsequent operations such as action driving, skin mesh generation, and action simulation. The virtual human model can perform various interactions and operations in the virtual reality environment, with more realistic and controllable characteristics. By mapping expert action data to the virtual human model and driving the model with actions, the movement of the virtual human can be achieved. This step enables the virtual human to accurately simulate according to the expert action data and respond to user interactions, achieving a more natural and realistic action performance. By generating a high-fidelity skin mesh for the action-driven human model, a realistic appearance can be added to the virtual human model. This step makes the appearance of the virtual human more real and detailed, presenting the virtual human more realistically in the virtual reality environment. Through bone animation technology, preliminary human movement can be achieved according to the virtual human action model. This step enables the virtual human model to perform basic animation based on the input action data, realizing simple action sequences. By smoothing the detailed movement according to the virtual action sequence and biomechanical parameters, the actions of the virtual human can be made more fluent and natural; this step can take into account the biomechanical constraints and optimize the virtual action sequence, making the movement of the virtual human more realistic and conforming to the biomechanical laws of human movement. By simulating the muscle movement and deformation based on biological parameters for the optimized virtual action sequence, the muscles of the virtual human can show realistic deformation during movement; this step can consider the biomechanical characteristics of human muscles, add muscle deformation to the virtual human, and make the virtual action demonstration more real and vivid. In summary, the effects of the above steps include: creating a virtual human model, achieving accurate action simulation, adding a realistic skin appearance, realizing basic action sequences, optimizing action details, and simulating muscle deformation; these effects make the performance of the virtual human in the virtual reality environment more realistic, natural, and vivid, providing a more immersive interaction experience for users; in addition, these steps are helpful for studying the human movement mechanism, optimizing sports training and rehabilitation programs, and conducting relevant research and applications in the field of virtual reality.

[0106] Preferably, step S4 includes the following steps:

[0107] Step S41: Obtain the action video data of the trainee;

[0108] The embodiment of the present invention collects the sports video data of the trainees, which can be recorded by a camera or a mobile device; the quality and clarity of the video are ensured to facilitate subsequent analysis and processing.

[0109] Step S42: extracting action segments from the student's action video data, and annotating the action segments with action types and key frames, thereby obtaining an action annotation data set;

[0110] The embodiment of the present invention uses video editing software (such as Adobe Premiere Pro, Final Cut Pro, etc.) to edit the student's action video and extract the clips containing specific actions; based on the extracted action clips, each action is labeled, including the marking of the action type and key frames. The labeling can be performed manually or automatically with the help of computer vision algorithms.

[0111] Step S43: performing data enhancement on the action annotation data set, extracting a human skeleton model based on posture estimation, thereby obtaining skeleton action annotation data;

[0112] The embodiment of the present invention performs data enhancement on the action annotation dataset, generates more diverse data samples by scaling, rotating, translating and other operations on the data, and enhances the robustness of the training model; uses a posture estimation algorithm (such as OpenPose, DeepPose, etc.) to process the enhanced data and extracts key node information of the human skeleton model, such as joint positions and angles.

[0113] Step S44: performing skeleton-based motion alignment according to the skeleton motion annotation data and the expert motion data, thereby obtaining aligned student motion data;

[0114] The embodiment of the present invention aligns the skeleton motion annotation data with the expert motion data, and finds the correspondence between the student motion and the expert motion by comparing the positions and motion trajectories of key nodes; using the aligned student motion data, subsequent analysis and comparison can be performed more accurately.

[0115] Step S45: performing key joint feature enhancement expression on the aligned student action data, thereby obtaining student action feature data;

[0116] The embodiment of the present invention extracts feature information of key joint points, such as joint angles, speeds, accelerations, etc., based on the aligned student motion data; feature extraction algorithms (such as principal component analysis, convolutional neural networks, etc.) can be used to enhance the expression of features to reduce redundant information and highlight key features.

[0117] Step S46: Perform identification of the exercise movements based on the virtual action demonstration data and the trainee's movement feature data, so as to obtain the trainee's movement pattern data.

[0118] In the embodiment of the present invention, based on the trainee's movement feature data and the virtual action demonstration data, machine learning algorithms (such as support vector machines, deep neural networks, etc.) are used for identification of the exercise movements; when training the model, the virtual action demonstration data is used as positive samples, and the trainee's movement feature data is used as negative samples. Through model training and classification, the identification of the trainee's movement pattern is achieved.

[0119] In the present invention, by acquiring the trainee's movement video data, the actual performance of the trainee under specific movements can be obtained; these video data provide a basis for analyzing and evaluating the trainee's movements and provide input data for subsequent steps. By intercepting and annotating the trainee's movement video data, the complete video can be divided into movement segments, and the movement type and key frames are annotated for each segment; the effect of this step is to effectively segment and annotate the trainee's movement data, providing an accurate data set for subsequent movement analysis and identification. By performing data augmentation and pose estimation on the action annotation data set, the action information of the human body bone model can be extracted from the video data; the effect of this step is to transform the trainee's movement data into more abstract and processable bone action annotation data, providing a basis for subsequent action alignment and feature extraction. By aligning the trainee's bone movement data with the expert's movement data, the trainee's movements can be compared and matched with the expert's standards; the effect of this step is to eliminate the differences in the trainee's movements, making the trainee's movement data closer to the expert's movement standards and providing a consistent benchmark for subsequent analysis and evaluation. By extracting and enhancing the expression of key joint point features from the aligned trainee's movement data, important feature information can be extracted from the movement data; the effect of this step is to transform the trainee's movement data into more compact and representative feature data, providing convenience for subsequent movement identification and analysis. By comparing and matching the virtual action demonstration data and the trainee's movement feature data, the trainee's movement pattern and performance can be identified; the effect of this step is to classify and identify the trainee's movements, providing personalized feedback and guidance for the trainee to help them improve and optimize their movement skills. To sum up, the effects of the above steps include: acquiring the trainee's movement data, dividing and annotating the movement segments, extracting the bone action annotation data, achieving action alignment, extracting the trainee's movement features, and identifying the trainee's movement pattern; these effects make the trainee's movement data more processable and analyzable, contributing to a deep understanding of the trainee's movement performance, providing personalized feedback and guidance, and improving the teaching and training effects; these steps provide a basis for subsequent movement analysis, evaluation and optimization, and contribute to improving the trainee's movement skills and motor ability.

[0120] Preferably, step S46 includes the following steps:

[0121] Step S461: Decompose the actions in the virtual action demonstration data into semantic action units, thereby obtaining a motion semantic data set;

[0122] In the embodiment of the present invention, the virtual action demonstration data is analyzed and observed, and each action is decomposed into semantic action units, that is, the basic motion elements in the action; according to the semantic information of the action, corresponding labels or codes are assigned to each semantic action unit to construct a motion semantic data set.

[0123] Step S462: Perform convolution feature extraction on the trainee action feature data based on kernels of different sizes, and splice the feature channels at different scales, thereby obtaining multi-scale feature representation data, where the kernels of different sizes include small-size kernels for capturing local spatio-temporal details and large-size kernels for capturing large-scale motion patterns;

[0124] In the embodiment of the present invention, a convolutional neural network (CNN) is used to extract features from the trainee action feature data, convolutional operations are performed using kernels of different sizes, small-size kernels are used to capture local spatio-temporal details, large-size kernels are used to capture large-scale motion patterns, and the feature channels at different scales are spliced to obtain multi-scale feature representation data.

[0125] Step S463: Construct a spatio-temporal attention module based on the multi-scale feature representation data, and calculate the correlation weights of each spatio-temporal position with all other positions through the spatio-temporal attention module, thereby obtaining spatio-temporal dependence feature data, where the correlation weight calculation includes time attention weight calculation based on capturing the temporal dependence of the action and space attention weight calculation based on capturing the correlation between the torso and the limbs;

[0126] In the embodiment of the present invention, a spatio-temporal attention module is constructed, which is used to calculate the correlation weights between spatio-temporal positions; this module is used to process the multi-scale feature representation data and calculate the correlation weights between each spatio-temporal position and other positions; the calculation of the correlation weights includes time attention weight calculation based on temporal dependence and space attention weight calculation based on the correlation between the torso and the limbs.

[0127] Step S464: Annotate the expert action data according to the motion semantic data set, and train a preset LSTM model based on the annotated data set, thereby obtaining an action recognition model;

[0128] In the embodiment of the present invention, the motion semantic data set is used to annotate the expert action data, and the semantic labels of each action are corresponded to the expert action data; a long short-term memory network (LSTM) model is constructed, and the model is trained using the annotated data set so that it can learn to recognize the patterns of different actions.

[0129] Step S465: Input the spatio-temporal dependence feature data into the action recognition model and perform recognition of the exercise actions to obtain the movement pattern data of the trainee.

[0130] In the embodiment of the present invention, the data processed by the spatio-temporal dependence feature is input into the trained action recognition model, and the action recognition model is used to recognize and classify the movement pattern of the trainee to obtain the movement pattern data of the trainee.

[0131] In the present invention, by decomposing the virtual action demonstration data into semantic action units, complex actions can be split into smaller semantic units, and the basic components of the actions can be extracted; the effect of this step is to obtain a movement semantic data set, where each semantic action unit represents a specific action segment, providing a basis for subsequent feature extraction and action recognition. By performing multi-scale feature extraction on the trainee's action feature data, the details and overall patterns of the actions can be captured at different scales; feature extraction is performed using convolutional kernels of different sizes, and the feature channels at different scales are spliced together to obtain a more comprehensive and rich feature representation; the effect of this step is to improve the expressive ability of the features, which helps to better capture the spatio-temporal features of the trainee's actions. By constructing a spatio-temporal attention module, spatio-temporal dependence modeling and capture can be performed on the multi-scale feature representation data. The spatio-temporal attention module calculates the correlation weights between each spatio-temporal position and other positions, and weights the important spatio-temporal positions and associated features. The effect of this step is to strengthen the temporal dependence and spatial correlation of the trainee's action features, thereby improving the accuracy and robustness of action recognition. By annotating the movement semantic data set and training the LSTM model using the annotated data set, a model for action recognition can be established; the effect of this step is to enable the model to learn and recognize the feature representations of different semantic action units, providing a reliable model basis for subsequent action recognition. By inputting the data processed by the spatio-temporal dependence feature into the action recognition model, the recognition and classification of the trainee's actions can be realized; the effect of this step is to match the trainee's actions with the known exercise actions and obtain the movement pattern data of the trainee, providing a basis for further analysis and evaluation; through this step, accurate recognition and classification of the trainee's actions can be achieved, so as to provide personalized and targeted guidance and feedback for the trainee's movement training.

[0132] Preferably, step S5 includes the following steps:

[0133] Step S51: Perform human key point detection and bone model construction on the trainee's action according to the trainee's action video data to obtain the biological parameters of the trainee.

[0134] The embodiments of the present invention use computer vision technologies, such as pose estimation or human keypoint detection methods, to process the video data of the trainee's movements; by analyzing the human poses and joint positions in the video, biological parameters of the trainee are extracted, such as joint angles, body postures, etc.; according to the extracted joint positions and pose information, a skeletal model of the trainee is constructed for subsequent biomechanical analysis and modeling. OpenPose: a commonly used human pose estimation library for detecting human keypoints; Kinect or camera: for collecting video data of the trainee's movements; computer vision algorithms and deep learning models: for performing human keypoint detection and pose estimation.

[0135] Step S52: Perform a virtual-real action difference metric analysis on the trainee's motion pattern data according to the virtual action demonstration data to obtain action difference metric data;

[0136] The embodiments of the present invention use the virtual action demonstration data as a reference to compare and analyze the trainee's motion pattern data and measure the difference between the trainee and the virtual action; according to the difference metric indicators (such as Euclidean distance, angle difference, etc.), the action difference metric data of the trainee is calculated to reflect the degree of difference between the trainee and the virtual action. Motion capture system: for obtaining virtual action demonstration data and trainee motion pattern data; Motion difference metric algorithm: Select an appropriate metric method for action difference metric according to specific requirements and data characteristics.

[0137] Step S53: Construct a personalized rigid body-muscle model representation based on the trainee's biological parameters and biomechanical parameters, thereby obtaining a personalized biomechanical model;

[0138] The embodiments of the present invention utilize the trainee's biological parameters and motion data to construct a rigid body model, including the skeletal structure and joint connection relationships; according to the trainee's biomechanical parameters, such as muscle length, strength, etc., combined with the rigid body model, a personalized rigid body-muscle model representation is constructed for simulating the trainee's motion process. Rigid body model construction software: such as 3D modeling software like Blender, Maya, etc.; Biomechanical parameter database: for obtaining the trainee's biomechanical parameters; Biomechanical simulation tool: such as OpenSim, etc., for constructing and simulating rigid body-muscle models.

[0139] Step S54: Perform constraint optimization according to the personalized biomechanical model and the action difference metric data to obtain an action correction plan data;

[0140] In the embodiments of the present invention, a personalized biomechanical model is combined with action difference measurement data to establish an optimization problem; through a constrained optimization method, an action correction scheme is found to make the trainee's motion pattern closer to the virtual action; during the optimization process, various constraint conditions can be set, such as joint angle ranges, muscle strength limits, etc., to ensure that the generated action correction scheme conforms to physiological constraints and individual characteristics. Optimization algorithms: such as nonlinear programming, evolutionary algorithms, etc.; optimization libraries or frameworks: such as SciPy, MATLAB, etc.; constraint condition setting: according to biomechanical knowledge and actual requirements, appropriate constraint conditions are set.

[0141] Step S55: According to the action correction scheme data, calculate the action differences before and after correction, evaluate the improvement degree, so as to obtain the training evaluation report data;

[0142] In the embodiments of the present invention, the trainee's motion pattern is corrected according to the action correction scheme, the corrected motion data is compared with the virtual action, the action difference measurement data before and after correction is calculated, and the correction effect is evaluated; according to the evaluation results, training evaluation report data is generated, including information such as the improvement degree and evaluation indicators. Motion capture system: used to obtain the corrected trainee motion data; motion difference measurement algorithm: the same method as in step S52, used to calculate the action difference measurement data; data analysis and report generation tools: such as Python, MATLAB, etc., used to analyze data and generate evaluation reports.

[0143] Step S56: Construct a training evaluation decision model based on the decision tree model according to the training evaluation report data.

[0144] In the embodiments of the present invention, according to the training evaluation report data, data analysis and feature extraction are performed; based on the decision tree model or other machine learning algorithms, a training evaluation decision model is constructed for automated evaluation and decision-making; the model is trained and verified and optimized to ensure the accuracy and reliability of the model. Data analysis and feature extraction tools: such as data analysis libraries of Python (such as NumPy, Pandas), feature selection algorithms, etc.; machine learning libraries and algorithms: such as scikit-learn, XGBoost, etc., used to construct the decision model; model verification and optimization methods: such as cross-validation, grid search, etc.

[0145] Through human key point detection and skeletal model construction on the action video data of trainees, the biological parameters of trainees can be obtained, including information such as joint positions and posture angles. The effect of this step is to provide the biological characteristics and posture information of trainees during exercise, laying a foundation for subsequent personalized biomechanical model construction and action analysis. By comparing and performing metric analysis on the motion pattern data of trainees and virtual action demonstration data, the differences between the actions of trainees and standard actions can be evaluated. The effect of this step is to obtain action difference metric data, which can be used to evaluate the motion skill level and action execution accuracy of trainees. By combining the biological parameters and biomechanical parameters of trainees, a personalized rigid body-muscle model can be constructed to simulate the biomechanical characteristics and action execution ability of trainees. The effect of this step is to obtain a personalized biomechanical model, which can more accurately describe the motion characteristics and action execution ability of trainees. By combining the personalized biomechanical model with the action difference metric data, constraint optimization can be performed to find the optimal action correction plan. The effect of this step is to obtain action correction plan data, which can guide trainees to improve and optimize their actions, improving the accuracy and effect of actions. By comparing the action differences before and after correction, the effect and improvement degree of action correction can be evaluated. The effect of this step is to obtain training evaluation report data, which can provide a quantitative evaluation of trainees' action improvement, helping to understand the effectiveness of the correction plan and the guidance of improvement strategies. By constructing a decision tree model based on the training evaluation report data, the evaluation results can be associated with corresponding decisions. The effect of this step is to obtain a training evaluation decision model, which can make corresponding decisions according to the training evaluation results of trainees, such as formulating personalized training plans and adjusting action correction strategies, to achieve more effective training and improve the sports performance of trainees.

[0146] Preferably, step S52 includes the following steps:

[0147] Step S521: Align the motion pattern data of trainees with the virtual action demonstration data, and calculate the similarity of the skeletal motion trajectories using the dynamic time warping algorithm, so as to obtain skeletal trajectory comparison data;

[0148] In the embodiments of the present invention, the trainee's motion pattern data and virtual action demonstration data are time-aligned to ensure that the motion sequence lengths and frequencies of both are consistent. The similarity of the skeletal motion trajectories is calculated using the Dynamic Time Warping (DTW) algorithm; the DTW algorithm can find the optimal matching path between two time series while considering time delay and deformation. According to the similarity calculated by the DTW algorithm, skeletal trajectory comparison data is obtained, reflecting the degree of trajectory difference between the trainee's motion pattern and the virtual action. Motion capture system: used to obtain the trainee's motion pattern data and virtual action demonstration data; Time alignment method: techniques such as interpolation or time warping can be used to ensure that the data lengths and frequencies are consistent; Dynamic Time Warping algorithm: common algorithms include the DTW algorithm based on distance metric and the DTW algorithm based on dynamic programming; Distance metric method: Euclidean distance, Manhattan distance, etc. can be used to calculate the similarity between skeletal motion trajectories.

[0149] Step S522: Calculate the differences in joint angles, torques, and tendon transmission ratio biological parameters based on the trainee's biological parameters and biomechanical parameters, so as to obtain biological parameter difference data;

[0150] In the embodiments of the present invention, the biological parameters and biomechanical parameters of the trainee are used to analyze the values of biological parameters such as joint angles, torques, and tendon transmission ratios; the biological parameters of the trainee are compared with the corresponding parameters of the virtual action demonstration data, and the difference values are calculated to reflect the degree of biological parameter difference between the trainee and the virtual action. Biomechanical parameter database: used to obtain the trainee's biomechanical parameters; Numerical calculation tools: such as the numerical calculation libraries of MATLAB and Python, used to calculate parameters such as joint angles, torques, and tendon transmission ratios; Parameter comparison and difference calculation method: according to the nature of the specific parameters, appropriate comparison methods such as absolute difference and relative difference are selected.

[0151] Step S523: Combine the skeletal trajectory comparison data and the biological parameter difference data into action difference metric data.

[0152] In the embodiments of the present invention, the skeletal trajectory comparison data and the biological parameter difference data are combined to form comprehensive action difference metric data; different data can be weighted or normalized according to specific requirements to obtain a unified metric index. Data processing tools: such as Python and MATLAB, used to process and combine data; Data weighting and normalization methods: according to the importance or range difference of the data, methods such as weighted average and maximum-minimum normalization can be used to ensure the comparability and consistency of the data.

[0153] The present invention can quantitatively compare the similarity between the actual movements of trainees and the standard movements by aligning the trainee movement pattern data with the virtual action demonstration data and calculating the similarity of the skeletal movement trajectories using the dynamic time warping algorithm. The effect of this step is to obtain skeletal trajectory comparison data, which can evaluate whether the trainee's action execution is consistent with the standard action and provide a basis for subsequent action difference measurement. By comparing the differences between the trainee's biological parameters and biomechanical parameters and the standard parameters, including aspects such as joint angles, torques, and tendon transmission ratios, the degree of difference between the trainee's biomechanical characteristics and the standard model can be evaluated. The effect of this step is to obtain biological parameter difference data, which can quantitatively understand the difference between the trainee's biomechanical characteristics and the standard model. By combining the skeletal trajectory comparison data and the biological parameter difference data, comprehensive action difference measurement data can be obtained. The effect of this step is to comprehensively consider the movement trajectory similarity and biological parameter differences of the trainee and provide comprehensive action difference measurement information. These data can be used to evaluate the trainee's action performance, guide action correction and improvement, and provide a basis for personalized training plans and decisions.

[0154] Preferably, step S56 includes the following steps:

[0155] Step S561: Format and clean the training evaluation report data, and perform quantization index and text extraction to obtain text feature data and quantization index data;

[0156] In the embodiment of the present invention, the training evaluation report data is formatted and cleaned to remove unnecessary punctuation marks, blank characters, etc., and to ensure the consistency and standardization of the data; according to the content of the evaluation report, the quantization index data therein is extracted, which involves using regular expressions or specific text processing methods to identify and extract relevant information such as numbers, percentages, time, etc.; similarly, the text feature data is extracted from the evaluation report. This includes extracting keywords, phrases, descriptive texts, etc.; appropriate preprocessing is performed on the extracted quantization index data and text feature data, such as removing stop words, performing stemming, etc.

[0157] Step S562: Normalize the quantization index data to obtain normalized quantization vector data; perform word segmentation and word vector encoding on the text feature data to obtain distributed text feature data;

[0158] In the embodiments of the present invention, the extracted quantitative index data is normalized and mapped to a unified numerical range. Common normalization methods include min-max normalization, Z-score normalization, etc.; the text feature data is tokenized to split the text into a sequence of words or phrases; the tokenized text feature data is encoded into word vectors to convert each word or phrase into a vector representation. Common word vector encoding methods include the bag-of-words model, TF-IDF, Word2Vec, GloVe, etc.

[0159] Step S563: Perform feature information fusion of different modalities based on the normalized quantization vector data and the distributed text feature data, and construct a composite feature group to obtain composite feature data.

[0160] In the embodiments of the present invention, the normalized quantization vector data and the distributed text feature data are subjected to feature fusion, and different methods can be used to combine the two types of feature information. Common feature fusion methods include concatenation, averaging, weighted averaging, etc.; according to actual requirements, a suitable feature fusion strategy can be selected to organically combine the feature information of different modalities to form composite feature data.

[0161] Step S564: Use the composite feature data as the training set of the decision tree model and perform parameter tuning to obtain a preliminary decision tree model.

[0162] In the embodiments of the present invention, the constructed composite feature data is used as input and divided into a training set and a validation set; a preliminary decision tree model is constructed using a decision tree algorithm (such as CART, ID3, C4.5, etc.). The decision tree model class in the scikit-learn library in Python can be used for modeling; the parameters of the decision tree model are tuned, and appropriate parameter settings can be selected through methods such as cross-validation. Grid search, random search, etc. can be used for parameter search and selection.

[0163] Step S565: Evaluate the generalization performance of the preliminary decision tree model on the reserved test set, and optimize the structure of the preliminary decision tree model based on the evaluation results to obtain the construction of the training evaluation decision model.

[0164] In the embodiments of the present invention, the reserved test set is used to evaluate the performance of the preliminary decision tree model, and the generalization ability of the model on new data is evaluated. Common evaluation metrics include accuracy, precision, recall, F1 value, etc. According to the evaluation results, the structure of the preliminary decision tree model is optimized. Pruning operations can be considered to reduce the risk of overfitting and improve the generalization performance of the model; through iterative optimization of steps S564 and S565 until a satisfactory training evaluation decision model is obtained.

[0165] Through formatting, cleaning, and extracting the training evaluation report data, the present invention can obtain structured data, including quantitative metrics and text features. The effect of this step is to transform the original evaluation report data into a processable data form, preparing for subsequent feature extraction and model construction. By normalizing the quantitative metric data, metric data with different ranges and units can be transformed into a unified scale, eliminating the influence of dimensionality and facilitating subsequent feature fusion and model training. At the same time, by performing word segmentation and word vector encoding on the text feature data, text information can be transformed into a numerical feature representation, enabling text features to be used together with other numerical features. By fusing the normalized quantitative vector data and distributed text feature data, feature information of different modalities can be combined to provide a more comprehensive and diverse feature representation. The effect of this step is to obtain composite feature data, which contains features of different dimensions and types and can better reflect the training evaluation of trainees. By using the composite feature data as the training set of the decision tree model, the decision tree algorithm can be used to learn the relationships between features and establish a preliminary decision tree model. Through parameter tuning, the performance and generalization ability of the model can be optimized, and the fitting degree of the model to the training evaluation data can be improved. By evaluating the preliminary decision tree model on the reserved test set, the generalization ability and prediction performance of the model can be evaluated. According to the evaluation results, the structure of the decision tree model can be optimized, including operations such as pruning and parameter adjustment, to improve the accuracy and stability of the model. The effect of this step is to obtain a training evaluation decision model, which can make corresponding decisions and recommendations based on the trainee's feature data, helping to develop personalized training plans and improvement strategies.

[0166] Therefore, in every aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be encompassed within the present invention.

[0167] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A construction method of a fitness training evaluation and decision-making model based on VR virtual digital humans, characterized in that, Including the following steps: Step S1: Perform motion capture on the expert's practice actions through a high-precision inertial camera group network to obtain expert action data, where the expert action data includes joint motion angle data and muscle activity potential data; Step S2: Based on the rigid body dynamics model and musculoskeletal simulation, analyze the expert action data for joint torque, tendon transmission ratio, and inertial force to obtain biomechanical parameters. Specifically, Step S2 is as follows: Step S21: Construct a rigid body chain model and a muscle model according to the expert action data; map the muscle activity potential data to the activity parameters of the muscle model to obtain a parameterized muscle model; Step S22: Use the joint motion angle data as the motion input to drive the rigid body chain model to move, and solve the forces and torques of each joint based on Newton's second law to obtain the torque values of each joint at each time step, thereby obtaining joint torque data; Step S23: Use the joint torque data as the input to perform inverse solution of muscle contraction force on the parameterized muscle model to obtain muscle contraction force data; Step S24: Calculate the torque acting on the bone based on the muscle attachment point position according to the muscle contraction force data to obtain bone torque data; Calculate the tendon transmission ratio data based on the bone torque data and the muscle contraction force data, where the tendon transmission ratio = bone torque / muscle contraction force; Step S25: Decompose the joint motion angle data into the motion of each rigid body according to the rigid body chain model, and calculate the linear acceleration and angular acceleration based on the rigid body to obtain rigid body motion parameter data; Step S26: Derive the inertial force and balance torque of each part of the body based on Newton's laws of motion for the rigid body motion parameter data to obtain inertial force data; Step S27: Perform biomechanical parameter analysis based on the joint torque data, tendon transmission ratio data, and inertial force data to obtain biomechanical parameters; Step S3: Import the expert action data and biomechanical parameters into the virtual reality engine, and generate virtual digital human action demonstrations through bone driving and inverse kinematics to obtain virtual action demonstration data; Step S4: Use computer vision technology to enhance the feature expression of key joint points in the pre-acquired student action video data, and identify the exercise actions in the student action video data according to the virtual action demonstration data to obtain student motion pattern data; Perform human biomechanical parameter analysis on the student action video data to obtain student biomechanical parameters; perform difference measurement analysis on the student motion pattern data according to the virtual action demonstration data to obtain action difference measurement data; perform action adjustment based on the biomechanical model according to the student biomechanical parameters and action difference measurement data, and generate training evaluation report data; construct a training evaluation decision model based on the decision tree model according to the training evaluation report data.

2. The construction method of the fitness training evaluation and decision-making model based on VR virtual digital humans according to claim 1, characterized in that Step S1 includes the following steps: Step S11: Use untextured blue-green as the background, arrange multiple high-precision inertial capture cameras for full-angle coverage to obtain a high-precision inertial camera group network; Step S12: Paste reflective marker points on the key parts of the expert's body and wear surface electromyography sensors. Through the expert's complete demonstration of the exercise action process in the inertial camera group network, the original electromyography data and the original 3D position data of the marker points are collected; Step S13: Eliminate noise and missing data from the original electromyography data and the original 3D position data, and synchronize the timestamps of the position and electromyography data, so as to obtain muscle activity data and 3D marker point position data; Step S14: Construct a bone model based on the muscle activity data and the 3D marker point position data through biomechanical modeling software, so as to obtain the expert bone model data; Step S15: Bind the 3D marker point position data to the expert bone model data and calculate the joint movement angles, so as to obtain the joint movement angle data; Estimate the muscle activity level and action intensity according to the muscle activity data, so as to obtain the muscle activity potential data; Step S16: Store the joint movement angle data and the muscle activity potential data as expert action data.

3. The construction method of the fitness training evaluation decision model based on the VR virtual digital human according to claim 2, characterized in that, Step S27 includes the following steps: Step S271: Integrate the joint torque data, tendon transmission ratio data and inertial force data into an initial set of biomechanical parameters; Step S272: Calculate the muscle contraction work according to the joint torque data to obtain the muscle contraction work data; Calculate the lost work done by the inertial force according to the inertial force data to obtain the inertial lost work data; Step S273: Calculate the total energy during the action process according to the muscle contraction work data and the inertial lost work data to obtain the energy consumption data, where energy consumption = muscle contraction work + inertial lost work; Step S274: Integrate the joint torque data, tendon transmission ratio data, inertial force data and energy consumption data into biomechanical parameters.

4. The construction method of the fitness training evaluation and decision-making model based on the VR virtual digital human according to claim 3, characterized in that, Step S3 includes the following steps: Step S31: Import the rigid body chain model and the parameterized muscle model into the virtual reality engine to obtain a virtual human model; Step S32: Map the data of the virtual human model according to the expert action data and drive the action of the human model to obtain an action-driven human model; Step S33: Generate a high-fidelity skin mesh model based on the bones for the action-driven human model to obtain a virtual human action model with a skin mesh; Step S34: Initially realize human movement according to the virtual human action model through bone animation technology to obtain a virtual action sequence; Step S35: Smooth the detailed movement based on inverse kinematics constraints according to the virtual action sequence and the biomechanical parameters to obtain an optimized virtual action sequence; Step S36: Simulate the muscle movement deformation of the optimized virtual action sequence based on the biological parameters to obtain the virtual action demonstration data with muscle deformation.

5. The construction method of the fitness training evaluation and decision-making model based on the VR virtual digital human according to claim 4, characterized in that Step S4 includes the following steps: Step S41: Obtain the student action video data; Step S42: Intercept the action segments of the student action video data, annotate the action types and key frames of the action segments to obtain an action annotation data set; Step S43: Perform data augmentation on the action annotation dataset and extract based on the pose estimation human skeleton model to obtain skeleton action annotation data; Step S44: Perform skeleton-based action alignment according to the skeleton action annotation data and the expert action data to obtain aligned trainee action data; Step S45: Strengthen the expression of key joint point features for the aligned trainee action data to obtain trainee action feature data; Step S46: Perform qigong action recognition according to the virtual action demonstration data and the trainee action feature data to obtain trainee motion pattern data.

6. The construction method of the fitness training evaluation and decision-making model based on VR virtual digital humans according to claim 5, characterized in that, Step S46 includes the following steps: Step S461: Decompose the actions in the virtual action demonstration data into semantic action units to obtain a motion semantics dataset; Step S462: Perform convolution feature extraction based on kernels of different sizes on the trainee action feature data, and splice the feature channels at different scales to obtain multi-scale feature representation data, where the kernels of different sizes include small-sized kernels to capture local spatio-temporal details and large-sized kernels to capture large-scale motion patterns; Step S463: Construct a spatio-temporal attention module based on the multi-scale feature representation data, and calculate the correlation weights of its spatio-temporal positions with all other positions through the spatio-temporal attention module to obtain spatio-temporal dependence feature data, where the correlation weight calculation includes time attention weight calculation based on capturing the temporal dependence of actions and spatial attention weight calculation based on capturing the correlation between the torso and limbs; Step S464: Annotate the expert action data according to the motion semantics dataset and train a preset LSTM model based on the annotation dataset to obtain an action recognition model; Step S465: Input the spatio-temporal dependence feature data into the action recognition model and perform qigong action recognition to obtain trainee motion pattern data.

7. The construction method of the fitness training evaluation and decision-making model based on the VR virtual digital human according to claim 6, characterized in that, Step S5 includes the following steps: Step S51: Detect human key points and construct a skeleton model for the trainee's actions according to the trainee action video data to obtain trainee biological parameters; Step S52: Perform virtual-real action difference metric analysis on the trainee motion pattern data according to the virtual action demonstration data to obtain action difference metric data; Step S53: Construct a personalized rigid body-muscle model representation according to the trainee biological parameters and biomechanical parameters to obtain a personalized biomechanical model; Step S54: Perform constraint optimization according to the personalized biomechanical model and the action difference metric data to obtain an action correction scheme data; Step S55: Evaluate the improvement degree of the action difference before and after correction according to the action correction scheme data to obtain training evaluation report data; Step S56: Construct a training evaluation decision model based on the decision tree model according to the training evaluation report data.

8. The construction method of the fitness training evaluation and decision-making model based on the VR virtual digital human according to claim 7, characterized in that, Step S52 includes the following steps: Step S521: Align the trainee motion pattern data with the virtual action demonstration data and calculate the similarity of the skeleton motion trajectories using the dynamic time warping algorithm to obtain skeleton trajectory comparison data; Step S522: Calculate the differences in joint angle, torque, and tendon transmission ratio bioparameters based on the trainee's bioparameters and biomechanical parameters, so as to obtain bioparameter difference data; Step S523: Combine the bone trajectory comparison data and the bioparameter difference data into action difference metric data.

9. The construction method of the fitness training evaluation decision-making model based on VR virtual digital humans according to claim 8, characterized in that, Step S56 includes the following steps: Step S561: Format and clean the training evaluation report data, and perform quantification index and text extraction to obtain text feature data and quantification index data; Step S562: Perform normalization processing on the quantification index data to obtain normalized quantification vector data; perform word segmentation and word vector encoding on the text feature data to obtain distributed text feature data; Step S563: Perform feature information fusion of different modalities based on the normalized quantification vector data and the distributed text feature data, and construct a composite feature group to obtain composite feature data; Step S564: Use the composite feature data as the training set of the decision tree model and perform parameter tuning to obtain a preliminary decision tree model; Step S565: Evaluate the generalization performance of the preliminary decision tree model on the reserved test set, and optimize the structure of the preliminary decision tree model based on the evaluation results to obtain the construction of the training evaluation decision model.

Images