Body-building action evaluation method and system based on pressure-visual detection

Abstract

The invention relates to a fitness action evaluation method and system based on pressure-visual detection, and belongs to the technical field of intelligent motion detection and human body posture analysis. The method comprises the following steps: acquiring a body-building action image, establishing a pressure-visual network, and reconstructing a current body-building posture according to a three-dimensional posture reconstruction method; constructing an action intensity evaluation model according to the fitness posture parameters to obtain pressure balance data for fitness action evaluation; according to the fitness action evaluation, three-dimensional reconstruction is carried out on the fitness action image through a branch perspective feature separation method, timestamp correlation is carried out on the fitness action image and the pressure-visual network, and a fitness intensity evaluation report is output; an intelligent fitness optimization feedback mechanism is established, when it is detected that the fitness intensity is too high compared with the current fitness plan intensity, personalized training feedback and action correction suggestions are generated and stored in a user fitness feedback file, a preset interface is called to dynamically adjust follow-up action chain parameters or replace an action scheme, and closed-loop self-adaptive training is achieved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent motion detection and human posture analysis technology, specifically relating to a fitness movement assessment method and system based on stress-visual detection. Background Technology

[0002] With the rapid development of the fitness industry, the demand for scientific and intelligent training is increasing. Traditional fitness movement assessment mainly relies on coaching experience and simple wearable devices, such as heart rate monitors or fitness trackers, which are insufficient for comprehensive quantitative analysis of muscle group stress distribution, movement stability, and postural deviations. While existing vision-based movement recognition methods can acquire key point location information of the human body, they have limitations in muscle group stress analysis, movement intensity assessment, and dynamic feedback, making it difficult to accurately reflect changes in muscle load and postural balance during movement. Furthermore, single-modal data cannot integrate stress and visual information, resulting in a lack of accuracy and real-time performance in movement intensity assessment. In addition, existing methods lack personalized training optimization mechanisms, making it difficult to provide dynamic adjustments and corrections based on the user's physical condition, training plan, and movement execution. Therefore, there is an urgent need for an intelligent fitness movement assessment method that can simultaneously acquire stress information and visual postural features during fitness movement execution, combined with multimodal data analysis and intensity assessment models, to achieve quantitative assessment of movement intensity, postural stability analysis, and personalized training optimization feedback, thus meeting the needs of scientific training and personalized guidance. Summary of the Invention

[0003] To address the aforementioned problems in the existing technology, this invention provides a fitness movement assessment method based on stress-visual detection. The objective of this invention can be achieved through the following technical solutions: S1: Acquire fitness movement images, establish a pressure-visual network based on the fitness movement images, and analyze the changes in pressure information and standard fitness posture pressure information during the execution of fitness movements using the three-dimensional posture reconstruction method to generate pressure posture information; S2: Construct a motion intensity assessment model based on the pressure posture information, analyze the temporal sequence of the pressure signal and the visual posture image of the pressure-visual network based on the intensity attention mechanism to obtain fitness posture data, and use the fitness posture data as input parameters to evaluate fitness movements through the motion intensity assessment model. S3: The fitness movement assessment uses the branch perspective feature separation method to reconstruct the fitness movement image in three dimensions, and uses the multimodal time feature synchronization mechanism to associate the movement group during the fitness process with the stress-visual network by time stamping, and outputs a fitness intensity assessment report. S4: Establish an intelligent fitness optimization feedback mechanism. Analyze the fitness intensity assessment report using an intensity threshold adaptive method. When a fitness training set is detected as unexecutable, optimize the current fitness training set and display the generated personalized training feedback in real time through the coach's interactive platform. The coach can answer system prompts on the interface of the coach's interactive platform. The answer data is sent back to the feedback learning decision model, stored in the user's fitness feedback file, and the fitness intensity adjustment plan is corrected.

[0004] Specifically, the method for obtaining the fitness movement images is as follows: based on a standard fitness group developed in a fitness plan, the three-dimensional posture of the exerciser is captured through visual recognition to obtain the fitness movement images.

[0005] Specifically, the method for establishing the pressure-visual network is as follows: based on the edge control points of the fitness limbs extracted from the fitness movement images, the movement edge deviation is analyzed by combining the pose solving algorithm with the standard fitness group, and the fitness pressure change of the analyzed movement edge deviation is used to construct the pressure-visual network.

[0006] Specifically, the method for generating the pressure attitude information is as follows: Based on the 3D pose reconstruction algorithm, the joints of the movement are spatially located during the execution of fitness movements. The 3D coordinates and relative poses of the joints are calculated based on the multi-view visual sequence and the image information of the fitness movements, so as to obtain the spatial pose vector of the movement node during the fitness process. The spatial pose vector of the action node is analyzed based on the pressure-visual network, and the change rate of the force distribution gradient of the visual pose deviation is calibrated by the weighted bias fusion algorithm. The coupling characteristics of the force and pose balance of the action node are analyzed, and the fused dynamic pose response curve is generated. The motion cycle is segmented and its features are normalized using the dynamic response curve of the posture described by the motion cycle detection algorithm. The distribution map of fitness pressure gradient and posture stability is analyzed by the pressure posture mapping algorithm to generate pressure posture information.

[0007] Specifically, the method for constructing the action intensity assessment model is as follows: The pressure signal sequence is uniformly encoded by the pressure-vision network, and the fluctuation of the pressure signal in the time dimension is analyzed based on the temporal convolution mechanism to extract the positional change of pressure at the action node as the fitness movement is performed. A dual-channel feature fusion network based on the intensity attention mechanism is constructed. The influence weight of the pressure signal change amplitude on the overall motion intensity is calculated by the positional change of the pressure of the action node. The intensity contribution region of the visual posture change is analyzed according to the regional gradient weighting method to generate a multi-scale fitness intensity response map. A motion intensity sample library is established based on multi-scale fitness intensity response maps. Fitness plans are optimized through supervised learning strategies. Intensity errors are analyzed in the joint feature space of stress time series and visual posture based on the loss function, and a motion intensity assessment model is constructed.

[0008] Specifically, the multi-scale fitness intensity response map extracts the fitness intensity frame features of each movement unit through the movement cycle segmentation algorithm, and labels the intensity level according to the physiological load level, muscle group activation degree and energy consumption parameters to establish an intensity sample library.

[0009] Specifically, the process of evaluating fitness movements using the aforementioned movement intensity assessment model is as follows: The temporal variation of the input pressure signal and visual posture are analyzed by feature mapping method, the response value of the action load at each moment is extracted, and the strength coupling degree between muscle force and posture change is calculated by temporal tensor decomposition algorithm to generate dynamic intensity distribution matrix. An intensity grading threshold is established to quantify and determine the generated dynamic intensity distribution matrix, outputting an exercise intensity load balance index, and generating a fitness exercise intensity assessment result based on the intensity grading threshold and the user's fitness needs.

[0010] Specifically, the branch perspective feature separation method establishes multiple branch convolutional channels in the pressure-visual fusion network, performs perspective domain mapping based on the pressure signal caused by visual posture changes, projects the three-dimensional fitness posture onto a multi-angle visual plane, separates the main action and auxiliary action, and suppresses interference response based on the feature attention weight allocation mechanism, and independently extracts the fitness action pressure between different branches.

[0011] Specifically, the method for establishing the intelligent fitness optimization feedback mechanism is as follows: Based on the adaptive threshold update algorithm, the movement stability index of the fitness intensity assessment report is extracted, and the individual intensity threshold range is calculated. When the current fitness intensity is detected to exceed the individual threshold range, the decision engine that integrates rule reasoning and neural network is used to analyze the source and risk level of movement deviation, and combined with the user's fitness training plan, a training rhythm adjustment strategy is generated. The training rhythm adjustment strategy is stored in the user's fitness feedback file using a temporal memory network algorithm, and an incremental learning optimization feedback strategy library is established. The user's fitness feedback file is analyzed based on the intelligent fitness optimization feedback mechanism, and optimized personalized fitness movement adjustment guidance is output. The intelligent fitness optimization feedback mechanism is called through the feedback learning decision model to update parameters or replace movements in the preset fitness movement plan database, and the user's fitness plan is automatically adjusted according to the updated fitness movement plan database.

[0012] Specifically, the execution process of the feedback learning decision model is as follows: The algorithm for progressive questioning is used to analyze and evaluate questions issued by the coach's interactive platform in real time, and to generate new question chain data based on the results of the real-time analysis and evaluation. The problem chain data is synchronized to the feedback learning decision model, and the predetermined movement parameters in the fitness movement plan database are evaluated based on the problem chain data and the fitness posture data in combination with the intelligent fitness optimization feedback mechanism to determine the adjustment of subsequent movement planning. The stress-visual network determines the adjustment signal for subsequent action planning. When it detects an increase or decrease in the stress load of adjusting subsequent fitness actions, the progressive problem algorithm inserts, deletes, and replaces actions in the fitness plan for subsequent action chains.

[0013] Specifically, the multimodal time feature synchronization mechanism performs nonlinear time alignment on the extracted peak points of fitness movement stress through a cross-modal correlation matching algorithm, and completes the synchronous mapping of the peak points of fitness movement stress based on a time offset compensation function.

[0014] Specifically, a fitness movement assessment system based on stress-visual detection is characterized by comprising: Intensity data acquisition module: acquires fitness movement images, establishes a pressure-visual network based on the fitness movement images, and analyzes the changes in pressure information and standard fitness posture pressure information during the execution of fitness movements using the three-dimensional posture reconstruction method to generate pressure posture information; Evaluation model construction module: Constructs a motion intensity evaluation model based on the stress posture information, analyzes the stress signal time series and visual posture image of the stress-visual network based on the intensity attention mechanism to obtain fitness posture data, and uses the fitness posture data as input parameters to evaluate fitness movements through the motion intensity evaluation model; 3D Reconstruction Synchronization Module: The fitness movement assessment uses the branch perspective feature separation method to reconstruct the fitness movement image in three dimensions, and uses the multimodal time feature synchronization mechanism to associate the movement group during the fitness process with the stress-visual network by timestamp, and outputs a fitness intensity assessment report. Fitness Optimization Feedback Module: Establishes an intelligent fitness optimization feedback mechanism. It analyzes the fitness intensity assessment report using an intensity threshold adaptive method. When a fitness training set is detected as unenforceable, it optimizes the current training set and displays the generated personalized training feedback in real-time through a coach-side interactive platform. Coaches can answer system prompts on the platform, and the response data is fed back to the feedback learning decision model, stored in the user's fitness feedback file, and used to correct the fitness intensity adjustment plan.

[0015] The beneficial effects of this invention are as follows: This invention fuses pressure information with visual posture features to construct a pressure-visual network, enabling multimodal data analysis during fitness movement execution. This allows for accurate acquisition of the force distribution and dynamic posture changes of each muscle group. Based on 3D posture reconstruction and movement cycle analysis, fitness posture parameters are generated, providing precise quantitative basis for movement intensity assessment. Through intensity attention mechanisms and multi-scale feature fusion, the temporal sequence of pressure signals and visual posture features can be jointly analyzed to generate pressure balance data and a dynamic intensity distribution matrix, achieving quantitative assessment of movement load balance and muscle group force coupling. Furthermore, this invention establishes an intelligent fitness optimization feedback mechanism. Combined with intensity threshold adaptive analysis, it can generate personalized training feedback and movement correction suggestions when the fitness intensity exceeds the user's individual tolerance range, dynamically adjusting the training plan and improving training safety and scientific rigor. Compared with existing technologies, this invention enables real-time monitoring, accurate assessment, and personalized optimization of fitness movements, overcoming the shortcomings of traditional methods in lacking quantitative analysis and dynamic feedback on muscle group force and posture balance. This improves training effectiveness and the scientific rigor of movement execution, providing technical support for intelligent and visualized fitness guidance. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 This is a schematic diagram of the structure of a fitness movement assessment method and system based on pressure-visual detection according to the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the technical flow of a fitness movement assessment method and system based on pressure-visual detection according to the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0020] Please see Figure 1 A fitness movement assessment method based on stress-visual detection: S1: Acquire fitness movement images, establish a pressure-visual network based on the fitness movement images, and analyze the changes in pressure information and standard fitness posture pressure information during the execution of fitness movements using the three-dimensional posture reconstruction method to generate pressure posture information; S2: Construct a motion intensity assessment model based on the pressure posture information, analyze the temporal sequence of the pressure signal and the visual posture image of the pressure-visual network based on the intensity attention mechanism to obtain fitness posture data, and use the fitness posture data as input parameters to evaluate fitness movements through the motion intensity assessment model. S3: The fitness movement assessment uses the branch perspective feature separation method to reconstruct the fitness movement image in three dimensions, and uses the multimodal time feature synchronization mechanism to associate the movement group during the fitness process with the stress-visual network by time stamping, and outputs a fitness intensity assessment report. S4: Establish an intelligent fitness optimization feedback mechanism. Analyze the fitness intensity assessment report using an intensity threshold adaptive method. When a fitness training set is detected as unexecutable, optimize the current fitness training set and display the generated personalized training feedback in real time through the coach's interactive platform. The coach can answer system prompts on the interface of the coach's interactive platform. The answer data is sent back to the feedback learning decision model, stored in the user's fitness feedback file, and the fitness intensity adjustment plan is corrected.

[0021] Specifically, the method for obtaining the fitness movement images is as follows: based on a standard fitness group developed in a fitness plan, the three-dimensional posture of the exerciser is captured through visual recognition to obtain the fitness movement images.

[0022] Specifically, the method for establishing the pressure-visual network is as follows: based on the edge control points of the fitness limbs extracted from the fitness movement images, the movement edge deviation is analyzed by combining the pose solving algorithm with the standard fitness group, and the fitness pressure change of the analyzed movement edge deviation is used to construct the pressure-visual network.

[0023] Specifically, the method for generating the pressure attitude information is as follows: Based on the 3D pose reconstruction algorithm, the joints of the movement are spatially located during the execution of fitness movements. The 3D coordinates and relative poses of the joints are calculated based on the multi-view visual sequence and the image information of the fitness movements, so as to obtain the spatial pose vector of the movement node during the fitness process. The spatial pose vector of the action node is analyzed based on the pressure-visual network, and the change rate of the force distribution gradient of the visual pose deviation is calibrated by the weighted bias fusion algorithm. The coupling characteristics of the force and pose balance of the action node are analyzed, and the fused dynamic pose response curve is generated. The motion cycle is segmented and its features are normalized using the dynamic response curve of the posture described by the motion cycle detection algorithm. The distribution map of fitness pressure gradient and posture stability is analyzed by the pressure posture mapping algorithm to generate pressure posture information.

[0024] This embodiment provides a fitness movement assessment method and system based on pressure-visual detection. The system employs a distributed collaborative control architecture consisting of a pressure acquisition unit, a posture visual analysis module, and a movement intensity assessment execution unit. Figure 1 As shown, it can be deployed in gyms, home training spaces, or rehabilitation training environments, and interacts with the central control and management system via industrial Ethernet + OPCUA protocol.

[0025] The specific system technology stack consists of the following: Main control processing platform: ARM Cortex-A72 ×4 (1.8GHz), EtherCAT / Modbus TCP / SPI / CAN; Visual acquisition device: RGB-D depth camera (Intel RealSense D435i / Azure Kinect); Pressure acquisition array: Flexible array thin-film pressure sensor (FPC array / piezoresistive MEMS sensor); Edge Computing Node: NVIDIA Jetson Orin NX / RK3588; Pressure signal analysis module: NumPy + SciPy + TensorRT (edge-accelerated inference); Action intensity evaluation model: dual-channel convolution + temporal attention mechanism (TCN + Attention); Feedback and Adaptive Control Module: Rule-based reasoning + neural network hybrid decision engine; Data Management and Service Layer (Cloud & Database Layer): TimescaleDB + InfluxDB (time-series data storage).

[0026] The system is installed in the gym's comprehensive training area and deployed on the strength training rack, free weight training mat, and treadmill platform. The system mainly consists of four parts: a pressure acquisition unit, a visual posture acquisition module, a motion intensity assessment and feedback module, and a central control and visualization terminal. Each unit achieves high-frequency data synchronization and real-time interaction through an industrial Ethernet (EtherCAT) + MQTT message bus.

[0027] Pressure acquisition unit (pressure signal acquisition and preprocessing): It adopts an ARM Cortex-A72 + real-time RTOS control core and integrates a high-speed data acquisition bus (EtherCAT / Modbus TCP) to acquire muscle surface pressure signals and body contact force data, and perform filtering, normalization and dynamic correction processing to provide real-time pressure data input for action intensity assessment.

[0028] The posture vision analysis module (action posture reconstruction and feature extraction) is built with an embedded Linux + OpenCV / PyTorch deep learning framework. It acquires human 3D posture information through multiple cameras and depth sensors, combines 3D posture reconstruction algorithms and joint spatial positioning models, calculates the 3D coordinates of muscle nodes and the relative pose of joints in real time, and outputs action posture parameters and key posture feature maps.

[0029] Action intensity assessment execution unit (stress-visual fusion and feedback): Based on stress-visual networks and intensity attention mechanisms, a multi-scale action intensity response matrix is ​​constructed; through tensor decomposition, stress posture mapping and dynamic intensity distribution analysis, muscle group load coupling degree and action stability assessment are realized.

[0030] Data visualization and multimodal synchronization service: Using the Web HMI front-end framework (Vue 3 + ECharts), it displays motion intensity curves, pressure balance distribution, and dynamic posture response; it transmits pressure and visual feature peaks through MQTT / Redis Stream to achieve multimodal time feature synchronization and motion cycle alignment, ensuring accurate correspondence between pressure signals and posture features within the motion cycle.

[0031] Anomaly detection and dynamic feedback scheduling service: When the posture deviation or pressure distribution abnormality exceeds the set threshold during the execution of the movement, the system triggers a feedback signal and updates the movement correction strategy through incremental learning algorithm to achieve personalized fitness movement optimization with closed-loop control.

[0032] Coach Control Terminal Module: An interactive control unit based on a tablet computing device (such as an iPad app), serving as the human-computer interaction interface for the fitness system. It displays training suggestions, movement evaluation results, and intensity optimization plans generated by the AI ​​analysis engine in real time through a graphical interface. This module integrates semantic parsing and input acquisition components to receive the coach's operation instructions, text feedback, and confirmation input. After local caching and encryption, the collected data is transmitted back to the system data management center in real time via a wireless communication protocol. Upon receiving feedback, the system dynamically adjusts the AI ​​suggestions, achieving closed-loop interactive control between the coach and AI evaluation ends, thereby supporting real-time correction and personalized optimization of training decisions.

[0033] Specifically, the method for constructing the action intensity assessment model is as follows: The pressure signal sequence is uniformly encoded by the pressure-vision network, and the fluctuation of the pressure signal in the time dimension is analyzed based on the temporal convolution mechanism to extract the positional change of pressure at the action node as the fitness movement is performed. A dual-channel feature fusion network based on the intensity attention mechanism is constructed. The influence weight of the pressure signal change amplitude on the overall motion intensity is calculated by the positional change of the pressure of the action node. The intensity contribution region of the visual posture change is analyzed according to the regional gradient weighting method to generate a multi-scale fitness intensity response map. A motion intensity sample library is established based on multi-scale fitness intensity response maps. Fitness plans are optimized through supervised learning strategies. Intensity errors are analyzed in the joint feature space of stress time series and visual posture based on the loss function, and a motion intensity assessment model is constructed.

[0034] Specifically, the multi-scale fitness intensity response map extracts the fitness intensity frame features of each movement unit through the movement cycle segmentation algorithm, and labels the intensity level according to the physiological load level, muscle group activation degree and energy consumption parameters to establish an intensity sample library.

[0035] Specifically, the process of evaluating fitness movements using the aforementioned movement intensity assessment model is as follows: The temporal variation of the input pressure signal and visual posture are analyzed by feature mapping method, the response value of the action load at each moment is extracted, and the strength coupling degree between muscle force and posture change is calculated by temporal tensor decomposition algorithm to generate dynamic intensity distribution matrix. An intensity grading threshold is established to quantify and determine the generated dynamic intensity distribution matrix, outputting an exercise intensity load balance index, and generating a fitness exercise intensity assessment result based on the intensity grading threshold and the user's fitness needs.

[0036] Specifically, the branch perspective feature separation method establishes multiple branch convolutional channels in the pressure-visual fusion network, performs perspective domain mapping based on the pressure signal caused by visual posture changes, projects the three-dimensional fitness posture onto a multi-angle visual plane, separates the main action and auxiliary action, and suppresses interference response based on the feature attention weight allocation mechanism, and independently extracts the fitness action pressure between different branches.

[0037] In this embodiment, as Figure 2 As shown, the overall technical flow of the system is as follows: Pressure data acquisition and attitude detection flow → Multimodal feature fusion module → Real-time intensity and balance control bus (based on EtherCAT / real-time bus) → (Attitude parameters and pressure time-series features are imported into InfluxDB for offline model training) → (Feature vectors are cached in embedded SQLite / local Flash | Action intensity assessment model self-learning) → Control command API → Feedback generation layer → Action chain update interface (calls the course plan database API to modify subsequent action parameters or replace actions).

[0038] In the real-time fusion module, the Δ attitude integral synchronization mechanism and the Watermark delay compensation strategy are enabled to ensure that the maximum out-of-order time of visual and stress data does not exceed 60 ms, so as to achieve time alignment between stress peak and attitude keyframe.

[0039] The fused temporal tensor is defined as: , Where α and β are modal weights, and V^(t) and P^(t) are normalized features.

[0040] To suppress asynchronous noise, the system employs sliding window smoothing and dynamic weighting: , Where ω(τ)=e−λ(t−τ) is the time-series decay coefficient.

[0041] To address the relationship between muscle stress and body center of gravity shift during training, a nonlinear intensity compensation function, Piecewise Adaptive-Intensity Compensation Function (PAICF), is constructed: , Where ΔI(t) = |I target -I real (t)∣,I crit a is a sensitive turning point for action intensity. i ,b i It is dynamically distributed from individual characteristic profiles to distinguish between the two strategies of "high-intensity rapid response" and "low-intensity stable response".

[0042] The real-time balance index B(t) is defined by a weighted coupling of attitude offset angle and pressure distribution: , Where θ(t) is the trunk-lower limb posture angle vector, V ar γ1 and γ2 represent the variance of the pressure distribution, and γ1 and γ2 are the weights.

[0043] In the vision-stress-feedback three-branch network, a collaborative evaluation matrix is ​​established: , Among them, W v (t), W p (t), W f (t) represents the weights of the visual, stress, and feedback branches, respectively, and p ij (t) represents the intermodal coupling strength, and the real-time update rule is: , Where η is the learning rate, y^(t) is the predicted intensity, and y(t) is the actual measured intensity.

[0044] By introducing a multimodal synchronization mechanism, a piecewise compensation function, and a three-branch cooperative matrix model, high-precision dynamic intensity assessment and personalized adaptive feedback control of fitness movements are achieved. It can run stably in a gym environment with an assessment latency of less than 100 ms and an error control within ±3%.

[0045] Specifically, the method for establishing the intelligent fitness optimization feedback mechanism is as follows: Based on the adaptive threshold update algorithm, the movement stability index of the fitness intensity assessment report is extracted, and the individual intensity threshold range is calculated. When the current fitness intensity is detected to exceed the individual threshold range, the decision engine that integrates rule reasoning and neural network is used to analyze the source and risk level of movement deviation, and combined with the user's fitness training plan, a training rhythm adjustment strategy is generated. The training rhythm adjustment strategy is stored in the user's fitness feedback file using a temporal memory network algorithm, and an incremental learning optimization feedback strategy library is established. The user's fitness feedback file is analyzed based on the intelligent fitness optimization feedback mechanism, and optimized personalized fitness movement adjustment guidance is output. The intelligent fitness optimization feedback mechanism is called through the feedback learning decision model to update parameters or replace movements in the preset fitness movement plan database, and the user's fitness plan is automatically adjusted according to the updated fitness movement plan database.

[0046] Specifically, the execution process of the feedback learning decision model is as follows: The algorithm for progressive questioning is used to analyze and evaluate questions issued by the coach's interactive platform in real time, and to generate new question chain data based on the results of the real-time analysis and evaluation. The problem chain data is synchronized to the feedback learning decision model, and the predetermined movement parameters in the fitness movement plan database are evaluated based on the problem chain data and the fitness posture data in combination with the intelligent fitness optimization feedback mechanism to determine the adjustment of subsequent movement planning. The stress-visual network determines the adjustment signal for subsequent action planning. When it detects an increase or decrease in the stress load of adjusting subsequent fitness actions, the progressive problem algorithm inserts, deletes, and replaces actions in the fitness plan for subsequent action chains.

[0047] Specifically, the multimodal time feature synchronization mechanism performs nonlinear time alignment on the extracted peak points of fitness movement stress through a cross-modal correlation matching algorithm, and completes the synchronous mapping of the peak points of fitness movement stress based on a time offset compensation function.

[0048] Specifically, a fitness movement assessment system based on stress-visual detection is characterized by comprising: Intensity data acquisition module: acquires fitness movement images, establishes a pressure-visual network based on the fitness movement images, and analyzes the changes in pressure information and standard fitness posture pressure information during the execution of fitness movements using the three-dimensional posture reconstruction method to generate pressure posture information; Evaluation model construction module: Constructs a motion intensity evaluation model based on the stress posture information, analyzes the stress signal time series and visual posture image of the stress-visual network based on the intensity attention mechanism to obtain fitness posture data, and uses the fitness posture data as input parameters to evaluate fitness movements through the motion intensity evaluation model; 3D Reconstruction Synchronization Module: The fitness movement assessment uses the branch perspective feature separation method to reconstruct the fitness movement image in three dimensions, and uses the multimodal time feature synchronization mechanism to associate the movement group during the fitness process with the stress-visual network by timestamp, and outputs a fitness intensity assessment report. Fitness Optimization Feedback Module: Establishes an intelligent fitness optimization feedback mechanism. It analyzes the fitness intensity assessment report using an intensity threshold adaptive method. When a fitness training set is detected as unenforceable, it optimizes the current training set and displays the generated personalized training feedback in real-time through a coach-side interactive platform. Coaches can answer system prompts on the platform, and the response data is fed back to the feedback learning decision model, stored in the user's fitness feedback file, and used to correct the fitness intensity adjustment plan.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A fitness movement assessment method based on stress-visual detection, characterized in that, include: S1: Acquire fitness movement images, establish a pressure-visual network based on the fitness movement images, and analyze the changes in pressure information and standard fitness posture pressure information during the execution of fitness movements using the three-dimensional posture reconstruction method to generate pressure posture information; S2: Construct a motion intensity assessment model based on the pressure posture information, analyze the temporal sequence of the pressure signal and the visual posture image of the pressure-visual network based on the intensity attention mechanism to obtain fitness posture data, and use the fitness posture data as input parameters to evaluate fitness movements through the motion intensity assessment model. S3: The fitness movement assessment uses the branch perspective feature separation method to reconstruct the fitness movement image in three dimensions, and uses the multimodal time feature synchronization mechanism to associate the movement group during the fitness process with the stress-visual network by time stamping, and outputs a fitness intensity assessment report. S4: Establish an intelligent fitness optimization feedback mechanism. Analyze the fitness intensity assessment report using an intensity threshold adaptive method. When a fitness training set is detected as unexecutable, optimize the current fitness training set and display the generated personalized training feedback in real time through the coach's interactive platform. The coach can answer system prompts on the interface of the coach's interactive platform. The answer data is sent back to the feedback learning decision model, stored in the user's fitness feedback file, and the fitness intensity adjustment plan is corrected.

2. The method according to claim 1, characterized in that, The method for obtaining the fitness movement images is as follows: based on a standard fitness group developed in a fitness plan, the three-dimensional posture of the exerciser is captured through visual recognition to obtain the fitness movement images.

3. The method according to claim 1, characterized in that, The method for establishing the pressure-visual network is as follows: based on the edge control points of the fitness limbs extracted from the fitness movement images, the movement edge deviation is analyzed by combining the pose solving algorithm with the standard fitness group, and the fitness pressure change of the analyzed movement edge deviation is used to construct the pressure-visual network.

4. The method according to claim 1, characterized in that, The method for generating the pressure attitude information is as follows: Based on the 3D pose reconstruction algorithm, the joints of the movement are spatially located during the execution of fitness movements. The 3D coordinates and relative poses of the joints are calculated based on the multi-view visual sequence and the image information of the fitness movements, so as to obtain the spatial pose vector of the movement node during the fitness process. The spatial pose vector of the action node is analyzed based on the pressure-visual network, and the change rate of the force distribution gradient of the visual pose deviation is calibrated by the weighted bias fusion algorithm. The coupling characteristics of the force and pose balance of the action node are analyzed, and the fused dynamic pose response curve is generated. The motion cycle is segmented and its features are normalized using the dynamic response curve of the posture described by the motion cycle detection algorithm. The distribution map of fitness pressure gradient and posture stability is analyzed by the pressure posture mapping algorithm to generate pressure posture information.

5. The method according to claim 2, characterized in that, The method for constructing the motion intensity assessment model is as follows: The pressure signal sequence is uniformly encoded by the pressure-vision network, and the fluctuation of the pressure signal in the time dimension is analyzed based on the temporal convolution mechanism to extract the positional change of pressure at the action node as the fitness movement is performed. A dual-channel feature fusion network based on the intensity attention mechanism is constructed. The influence weight of the pressure signal change amplitude on the overall motion intensity is calculated by the positional change of the pressure of the action node. The intensity contribution region of the visual posture change is analyzed according to the regional gradient weighting method to generate a multi-scale fitness intensity response map. A motion intensity sample library is established based on multi-scale fitness intensity response maps. Fitness plans are optimized through supervised learning strategies. Intensity errors are analyzed in the joint feature space of stress time series and visual posture based on the loss function, and a motion intensity assessment model is constructed.

6. The method according to claim 5, characterized in that, The multi-scale fitness intensity response map extracts the fitness intensity frames of each movement unit through the movement cycle segmentation algorithm, and labels the intensity level according to the physiological load level, muscle group activation degree and energy consumption parameters to establish an intensity sample library.

7. The method according to claim 4, characterized in that, The process of evaluating fitness movements using the aforementioned movement intensity assessment model is as follows: The temporal variation of the input pressure signal and visual posture are analyzed by feature mapping method, the response value of the action load at each moment is extracted, and the strength coupling degree between muscle force and posture change is calculated by temporal tensor decomposition algorithm to generate dynamic intensity distribution matrix. An intensity grading threshold is established to quantify and determine the generated dynamic intensity distribution matrix, outputting an exercise intensity load balance index, and generating a fitness exercise intensity assessment result based on the intensity grading threshold and the user's fitness needs.

8. The method according to claim 2, characterized in that, The branch perspective feature separation method establishes multiple branch convolutional channels in the pressure-visual fusion network, performs perspective domain mapping based on the pressure signal caused by visual posture changes, projects the three-dimensional fitness posture onto a multi-angle visual plane, separates the main action and auxiliary action, and suppresses interference response based on the feature attention weight allocation mechanism, and independently extracts the fitness action pressure between different branches.

9. The method according to claim 4, characterized in that, The method for establishing the intelligent fitness optimization feedback mechanism is as follows: Based on the adaptive threshold update algorithm, the movement stability index of the fitness intensity assessment report is extracted, and the individual intensity threshold range is calculated. When the current fitness intensity is detected to exceed the individual threshold range, the decision engine that integrates rule reasoning and neural network is used to analyze the source and risk level of movement deviation, and combined with the user's fitness training plan, a training rhythm adjustment strategy is generated. The training rhythm adjustment strategy is stored in the user's fitness feedback file using a temporal memory network algorithm, and an incremental learning optimization feedback strategy library is established. The user's fitness feedback file is analyzed based on the intelligent fitness optimization feedback mechanism, and optimized personalized fitness movement adjustment guidance is output. The intelligent fitness optimization feedback mechanism is called through the feedback learning decision model to update parameters or replace movements in the preset fitness movement plan database, and the user's fitness plan is automatically adjusted according to the updated fitness movement plan database.

10. The method according to claim 2, characterized in that, The execution process of the feedback learning decision model is as follows: The algorithm for progressive questioning is used to analyze and evaluate questions issued by the coach's interactive platform in real time, and to generate new question chain data based on the results of the real-time analysis and evaluation. The problem chain data is synchronized to the feedback learning decision model, and the predetermined movement parameters in the fitness movement plan database are evaluated based on the problem chain data and the fitness posture data in combination with the intelligent fitness optimization feedback mechanism to determine the adjustment of subsequent movement planning. The stress-visual network determines the adjustment signal for subsequent action planning. When it detects an increase or decrease in the stress load of adjusting subsequent fitness actions, the progressive problem algorithm inserts, deletes, and replaces actions in the fitness plan for subsequent action chains.

11. The method according to claim 7, characterized in that, The multimodal time feature synchronization mechanism performs nonlinear time alignment on the extracted peak stress points of fitness movements using a cross-modal correlation matching algorithm, and completes the synchronous mapping of the peak stress points of fitness movements based on a time offset compensation function.

12. A fitness movement assessment system based on stress-visual detection, characterized in that, include: Intensity data acquisition module: acquires fitness movement images, establishes a pressure-visual network based on the fitness movement images, and analyzes the changes in pressure information and standard fitness posture pressure information during the execution of fitness movements using the three-dimensional posture reconstruction method to generate pressure posture information; Evaluation model construction module: Constructs a motion intensity evaluation model based on the stress posture information, analyzes the stress signal time series and visual posture image of the stress-visual network based on the intensity attention mechanism to obtain fitness posture data, and uses the fitness posture data as input parameters to evaluate fitness movements through the motion intensity evaluation model; 3D Reconstruction Synchronization Module: The fitness movement assessment uses the branch perspective feature separation method to reconstruct the fitness movement image in three dimensions, and uses the multimodal time feature synchronization mechanism to associate the movement group during the fitness process with the stress-visual network by timestamp, and outputs a fitness intensity assessment report. Fitness Optimization Feedback Module: Establishes an intelligent fitness optimization feedback mechanism. It analyzes the fitness intensity assessment report using an intensity threshold adaptive method. When a fitness training set is detected as unenforceable, it optimizes the current training set and displays the generated personalized training feedback in real-time through a coach-side interactive platform. Coaches can answer system prompts on the platform, and the response data is fed back to the feedback learning decision model, stored in the user's fitness feedback file, and used to correct the fitness intensity adjustment plan.

Images