Intelligent exercise health assessment method and system based on multi-sensor fusion

Abstract

The invention discloses an intelligent exercise health assessment method and system based on multi-sensor fusion, and relates to the technical field of intelligent health assessment and monitoring, and the method comprises the steps: collecting multi-modal exercise data and physiological signals, and carrying out the preprocessing of the collected data; carrying out data fusion and extracting key motion and health features; constructing an intelligent exercise health evaluation model, and analyzing the health state and the exercise performance of the user; personalized health feedback is generated based on an evaluation result, and real-time early warning is provided; and dynamically updating parameters of the evaluation model, and performing model optimization according to a specific motion scene and health requirements of the user. According to the method, various sensor data can be effectively fused, accurate health scores and real-time early warning can be generated, the health requirements of the user in different exercise scenes are met, the accuracy, adaptability and intelligent level of evaluation are remarkably improved, meanwhile, health improvement suggestions with high pertinence are provided for the user, and the user experience is improved. The method has remarkable technical advantages and application value.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent health assessment and monitoring, and particularly to a method and system for intelligent assessment of sports health based on multi-sensor fusion. Background Art

[0002] With the rapid development of artificial intelligence technology, sensor technology, and big data analysis, sports health assessment has gradually become an important research direction in sports medicine, rehabilitation training, and personal health management. By obtaining users' motion data and physiological signals through sensors and combining advanced analysis algorithms to quantitatively evaluate users' health status and sports performance has become a hot topic in the field of intelligent health. Currently, the widespread application of multi-modal sensors makes it possible to collect multi-dimensional motion and health data, such as acceleration, angular velocity, heart rate, blood oxygen saturation, etc. However, the data generated by these sensors have the characteristics of high dimensionality, diversity, and dynamic changes, which pose higher requirements for data fusion analysis and real-time processing. In addition, the growing demand for personalized health feedback and real-time warning has also promoted the innovative development of sports health assessment technology based on artificial intelligence.

[0003] Although the prior art has made certain progress in sports health assessment, there are still deficiencies in many aspects. First, the prior art mostly focuses on the analysis of single-sensor data, and has weak ability to fuse and process multi-modal data, making it difficult to fully exploit the complementary information between different sensor data. For example, evaluating exercise intensity based only on acceleration data cannot comprehensively reflect the user's health status. Second, the preprocessing methods for the collected data are relatively simple, such as denoising and normalization, and cannot handle the non-linear characteristics in high-dimensional dynamic data, resulting in insufficient accuracy and robustness of the analysis results. In addition, the prior art usually relies on static models and lacks the ability of dynamic update, making it difficult to meet the personalized needs of users in specific sports scenarios, especially the evaluation effect is poor in complex environments or special sports modes. More importantly, the prior art often stays at the basic prompt level in generating personalized health feedback and real-time warning, and cannot provide precise health improvement suggestions and dynamic warning schemes according to the user's real-time status. Therefore, the prior art cannot achieve the effect of intelligent sports health assessment with strong real-time performance, high adaptability, and precise feedback. In view of these deficiencies, there is an urgent need for an intelligent assessment method that can comprehensively fuse multi-modal data, dynamically update the assessment model, and provide personalized feedback. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing motion health assessment methods have insufficient multi-modal data fusion capabilities, limited accuracy in data preprocessing and feature extraction, lack of dynamic update capabilities in the assessment model, and are unable to generate accurate personalized feedback and warnings based on real-time assessment results.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a motion health intelligent assessment method based on multi-sensor fusion, including collecting multi-modal motion data and physiological signals, and preprocessing the collected data;

[0008] Performing data fusion and extracting key motion and health features;

[0009] Constructing a motion health intelligent assessment model to analyze the user's health status and motion performance;

[0010] Generating personalized health feedback based on the assessment results and providing real-time warnings;

[0011] Dynamically updating the parameters of the assessment model and optimizing the model according to the user's specific motion scenarios and health needs.

[0012] As a preferred solution of the motion health intelligent assessment method based on multi-sensor fusion according to the present invention, wherein: the multi-modal motion data includes acceleration data, angular velocity data, displacement trajectory data, and muscle activity data;

[0013] The physiological signals include collecting the user's heart rate data, blood oxygen saturation data, respiratory rate data, skin temperature data, and sweat conductivity data.

[0014] As a preferred solution of the motion health intelligent assessment method based on multi-sensor fusion according to the present invention, wherein: the preprocessing of the collected data includes removing outliers and filling missing values from the collected data, removing noise signals and filling in missing data;

[0015] Performing denoising and smoothing processing on the collected data to remove high-frequency interference and low-frequency drift;

[0016] Performing normalization and standardization processing on different types of data to eliminate the difference in data dimensions;

[0017] Extracting frequency domain or time-frequency domain features to reveal the hidden information in the data;

[0018] Performing timestamp alignment and synchronization processing on the collected data to ensure the consistency of various types of data in the time dimension;

[0019] Reduce the dimension of high-dimensional data, extract the main feature components, and reduce the computational complexity.

[0020] As a preferred solution of the motion health intelligent evaluation method based on multi-sensor fusion according to the present invention, wherein: the data fusion and extraction of key motion and health features include classifying the preprocessed data according to the acquisition dimension and uniformly representing the data of the same type;

[0021] Analyze the correlation and complementarity of different types of sensor data, and generate a unified feature data set through a specific fusion algorithm;

[0022] Extract the trajectory features of the user for analyzing the motion pattern and dynamic behavior;

[0023] Extract the posture features of the user for evaluating the posture stability and balance ability;

[0024] Extract the muscle activity features of the user for analyzing the muscle activation level and fatigue state;

[0025] Extract the heart rate features of the user for evaluating the cardiac load and exercise intensity;

[0026] Extract the blood oxygen features of the user for analyzing the oxygen utilization efficiency and endurance level;

[0027] Extract the breathing features of the user for evaluating the metabolic state and exercise recovery ability;

[0028] Calculate the correlation coefficient between the features and the evaluation target based on the statistical analysis method, and filter out and remove the features with a correlation coefficient lower than the predetermined threshold;

[0029] Recombine the filtered feature data set to generate an optimized feature input adapted to the intelligent evaluation model.

[0030] As a preferred solution of the motion health intelligent evaluation method based on multi-sensor fusion according to the present invention, wherein: the construction of the motion health intelligent evaluation model includes constructing a feature vector:

[0031] X = [a i , b i , k i , v i , x i , θ, ψ]

[0032] Wherein, a i is the displacement data, describing the spatial position change of the user; b i is the acceleration data, reflecting the exercise intensity and change rate; k i is the dynamic frequency, used to capture the exercise rhythm; v iis the speed data, representing the linear change speed of the user's movement; x i is a health feature; θ is the user's angle data for postural stability analysis; ψ is the postural angle change rate, describing the user's rotation or tilt dynamics;

[0033] Calculate the comprehensive motion performance score S of the user based on the motion data in the feature vector m , and the formula is:

[0034]

[0035] where N is the number of time segments, Δt is the interval of each time segment; β i is a weight parameter, which is optimized and determined through training data;

[0036] Calculate the health status score S based on the health data in the feature vector h , and the formula is:

[0037]

[0038] where γ is the Gaussian weighting parameter for adjusting the contribution of muscle activity to the score; f(y) is the muscle activity distribution function reflecting the user's muscle activation state; y is the muscle activity intensity;

[0039] Calculate the overall score S of the user by combining the comprehensive motion performance score and the health status score, and the formula is:

[0040]

[0041] where w m and w h are the weights of the motion performance score and the health status score respectively, satisfying w m +w h =1; h j is the scene-related health feature; η j is the corresponding scene feature weight; g(h j ) is the mapping function of the scene feature.

[0042] As a preferred solution of the motion health intelligent evaluation method based on multi-sensor fusion described in the present invention, wherein: the generation of personalized health feedback based on the evaluation results includes, according to the comprehensive score S and the health status score S h , the motion performance score S m of the user, combined with historical health data and specific scene features, to divide the user's status level;

[0043] Divide the user's health status and motion performance levels according to the value range of the comprehensive score S;

[0044] When S < T1 When it is determined that the user is in an abnormal state, a health risk warning is generated, the user is notified to stop the current exercise and health check suggestions are provided, the abnormal reason is described in the feedback content, and mitigation measures are recommended;

[0045] When T 1 ≤S≤T 2 When it is determined that the user is in an optimizable state, exercise optimization suggestions are provided;

[0046] When S>T 2 When it is determined that the user is in an excellent state, a status maintenance strategy is provided;

[0047] Among them, T 1 and T 2 are preset scoring thresholds;

[0048] Generate personalized feedback content, which is pushed to the user terminal in the form of text, graphics and voice, including specific health improvement suggestions, exercise plan adjustments or health maintenance strategies;

[0049] Add an abnormal risk reminder to the feedback, explaining the existing risks and their impacts.

[0050] As a preferred solution of the motion health intelligent evaluation method based on multi-sensor fusion described in the present invention, wherein: the parameters for dynamically updating the evaluation model include collecting the user's motion data and health signals in real time, and jointly forming an extended training data set with historical data;

[0051] Dynamically adjust the model parameter set P = [w m , w h , η j , β i through an incremental learning algorithm;

[0052] Use the cross-validation method to evaluate the performance of the updated model to ensure the accuracy and stability of the model;

[0053] Analyze the changing trend of the user's characteristics in a specific exercise scenario;

[0054] Construct a scenario feature set C = [c 1 , c 2 ,..., c n for a specific scenario, and adjust the model input structure through scenario perception;

[0055] Use the objective function to guide the optimization process of the model parameters, and dynamically adjust the model structure and parameter configuration to meet the user's health needs.

[0056] In a second aspect, an embodiment of the present invention provides a motion health intelligent evaluation system based on multi-sensor fusion, including:

[0057] Data acquisition and preprocessing module: Collect multi-modal motion data and physiological signals, and preprocess the collected data;

[0058] Data fusion and feature extraction module: Perform data fusion and extract key motion and health features;

[0059] Evaluation model construction and analysis module: Construct an intelligent evaluation model for sports health, and analyze the user's health status and sports performance;

[0060] Personalized feedback and warning module: Generate personalized health feedback based on the evaluation results and provide real-time warnings;

[0061] Model dynamic update and optimization module: Dynamically update the parameters of the evaluation model and optimize the model according to the user's specific sports scenarios and health needs.

[0062] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0063] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0064] Advantages of the present invention: By integrating motion data and physiological signals from multiple sensors (such as acceleration, angular velocity, displacement trajectory, heart rate, blood oxygen saturation, etc.), the complementary information between different data sources is fully exploited, thereby improving the comprehensiveness and accuracy of health assessment.

[0065] Through methods such as outlier rejection, normalization, and time-frequency domain feature extraction, the collected data is deeply processed to ensure the quality and consistency of high-dimensional dynamic data, and provide optimized input features for the subsequent evaluation model.

[0066] Adopt incremental learning and objective function optimization techniques, so that the evaluation model can be dynamically updated according to the user's specific sports scenarios and health needs, and improve the adaptability and real-time performance of the evaluation.

[0067] According to the comprehensive score and the health status evaluation results, generate accurate personalized feedback suggestions, provide targeted improvement measures for different health statuses, and trigger multi-level warnings in combination with real-time monitoring data to ensure user safety.

[0068] It can adapt to different sports environments and user needs, and can be applied both in daily health management and professional sports training, with high practicality and promotion value. Description of the Drawings

[0069] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings, where:

[0070] Figure 1 It is the overall flowchart of a motion health intelligent evaluation method based on multi-sensor fusion provided by the first embodiment of the present invention;

[0071] Figure 2 It is the model construction flowchart of a motion health intelligent evaluation method based on multi-sensor fusion provided by the first embodiment of the present invention;

[0072] Figure 3 It is the module connection diagram of a motion health intelligent evaluation system based on multi-sensor fusion provided by the third embodiment of the present invention. Specific Embodiments

[0073] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0074] Embodiment 1, referring to Figure 1 and Figure 2 This is an embodiment of the present invention, which provides a motion health intelligent evaluation method based on multi-sensor fusion, including:

[0075] S1: Collect multi-modal motion data and physiological signals, and preprocess the collected data.

[0076] The multi-modal motion data includes acceleration data, angular velocity data, displacement trajectory data, and muscle activity data; the physiological signals include collecting the user's heart rate data, blood oxygen saturation data, respiratory rate data, skin temperature data, and sweat conductivity data.

[0077] Furthermore, acceleration data, which is used to describe the acceleration changes of the user in three-dimensional space, is a key parameter for analyzing exercise intensity and dynamic behavior; angular velocity data, which is used to capture the rotational motion characteristics of the user, can evaluate postural stability and balance ability; displacement trajectory data, through the analysis of displacement changes, can reveal the user's motion patterns and motion ranges; muscle activity data, which reflects the activation level and fatigue state of muscles, can supplement physiological information that cannot be covered by traditional exercise data.

[0078] It should be noted that multi-modal data can complement each other. For example, by combining acceleration data and muscle activity data, the exercise intensity can be more accurately reflected; the combination of blood oxygen data and respiratory frequency data can more comprehensively evaluate the user's metabolic level and endurance.

[0079] Outlier removal and missing value filling are performed on the collected data to remove noise signals and fill in missing data; the improvement of data quality directly enhances the accuracy of the evaluation model. For example, in the case of temporary loss of sensor data, the filling technology can restore key information and ensure the coherence of the evaluation results.

[0080] Denoising and smoothing processing are performed on the collected data to remove high-frequency interference and low-frequency drift; the smoothed data after denoising is more suitable for feature extraction and modeling. Especially in real-time evaluation scenarios, smoothing processing can reduce the computational complexity and improve the model response speed.

[0081] Normalization and standardization processing are performed on different types of data to eliminate the differences in data dimensions; frequency domain or time-frequency domain features are extracted to reveal the hidden information in the data.

[0082] Timestamp alignment and synchronization processing are performed on the collected data to ensure the consistency of various data in the time dimension; the aligned data can more accurately reflect the collaborative relationship of multi-modal signals. For example, by combining acceleration and heart rate data during exercise, the user's exercise efficiency can be better evaluated.

[0083] Dimensionality reduction processing is performed on high-dimensional data to extract the main feature components and reduce the computational complexity; dimensionality reduction processing not only reduces the computational overhead of the model but also retains the main information of the data, enabling the intelligent evaluation model to operate efficiently while ensuring the evaluation accuracy.

[0084] It should be noted that the high-dimensional data in the present invention includes data from multiple sensors, such as acceleration data, angular velocity data, displacement trajectory data, muscle activity data, heart rate data, blood oxygen data, etc. These data usually have multiple dimensions in the time series, such as frequency domain features, spatio-temporal correlation features, etc.

[0085] The specific dimensionality reduction method is generally principal component analysis (PCA): Calculate the covariance matrix of the input data, solve the eigenvectors and eigenvalues, and retain the principal components with the highest contribution rate;

[0086] Linear discriminant analysis (LDA): Used to optimize data dimensionality reduction in classification tasks and maximize the between-class differences;

[0087] Autoencoder: Use neural networks to extract low-dimensional representations and optimize the dimensionality reduction process through reconstruction errors.

[0088] S2: Perform data fusion and extract key motion and health features.

[0089] Classify the preprocessed data according to the acquisition dimensions and represent the same type of data uniformly; Solve the problem of inconsistent data formats of different sensors and lay a foundation for subsequent fusion and analysis.

[0090] Analyze the correlation and complementarity of different types of sensor data, and generate a unified set of feature data through specific fusion algorithms; By calculating the correlation coefficients between specific sensor data, identify the complementary characteristics of multimodal data, such as the synergistic effect of heart rate and blood oxygen saturation in evaluating endurance.

[0091] Extract the trajectory features of the user, based on displacement, velocity, and acceleration, for analyzing motion patterns and dynamic behaviors.

[0092] Extract the pose features of the user, including body angle changes and rotational angular velocity, for evaluating pose stability and balance ability.

[0093] Extract the muscle activity features of the user for analyzing muscle activation levels and fatigue states.

[0094] Extract the heart rate features of the user for evaluating cardiac load and exercise intensity.

[0095] Extract the blood oxygen features of the user for analyzing oxygen utilization efficiency and endurance levels.

[0096] Extract the breathing features of the user for evaluating metabolic status and exercise recovery ability.

[0097] Calculate the correlation coefficients between the features and the evaluation objectives based on statistical analysis methods, and filter out and remove the features with correlation coefficients lower than the predetermined threshold.

[0098] Recombine the filtered set of feature data to generate an optimized feature input adapted to the intelligent evaluation model.

[0099] It should be noted that after feature screening, the remaining highly correlated features may still have problems such as dimensional differences or uneven distributions. Through feature recombination, not only is the matching degree between features and evaluation objectives enhanced, but also the computational complexity of the high-dimensional feature set is further reduced. These features can be further optimized to generate feature inputs suitable for intelligent evaluation models. Ensuring that the optimized feature inputs can be adapted to intelligent evaluation models in diverse scenarios improves the generality and robustness of the models.

[0100] Feature screening and recombination techniques not only improve the evaluation accuracy of the model but also ensure its ability to operate efficiently in application scenarios with high real-time requirements (such as motion monitoring and health warnings).

[0101] Furthermore, specific fusion algorithms generally include data-level fusion, feature-level fusion, and decision-level fusion;

[0102] Data-level fusion: Align and interpolate multi-sensor data at the same timestamp to ensure data synchronization.

[0103] Feature-level fusion: Calculate the mutual information of different sensor data to evaluate the correlation, and perform feature weighted fusion based on highly correlated features; Use the attention mechanism to adaptively adjust the weights of each sensor data to enhance the contribution of key features.

[0104] Decision-level fusion: Combine the decision results of multiple models using voting or weighted average methods to improve the robustness of the evaluation.

[0105] S3: Build a sports health intelligent evaluation model to analyze the user's health status and sports performance.

[0106] Building a sports health intelligent evaluation model includes building a feature vector:

[0107] X = [a i , b i , k i , v i , x i , θ, ψ]

[0108] where a i is displacement data, describing the user's spatial position change; b i is acceleration data, reflecting the exercise intensity and rate of change; k i is the dynamic frequency, used to capture the exercise rhythm; v i is speed data, indicating the linear change speed of the user's movement; x iis a health feature; θ is the angular data of the user for postural stability analysis; ψ is the rate of change of the attitude angle, describing the rotation or tilt dynamics of the user.

[0109] Calculate the comprehensive exercise performance score S of the user based on the motion data in the feature vector m , and the formula is:

[0110]

[0111] where N is the number of time segments, Δt is the interval of each time segment; β i is a weight parameter, which is optimized and determined through training data. Combining the dynamic changes of acceleration, speed and frequency, the volatility of exercise performance in the time dimension is captured. The weight parameter β i is dynamically optimized through training data to ensure that the influence weights of different features on the score are adaptive. w i is the time step adjustment weight, which is used to adjust the contribution of different time slices to the final score and can be optimized according to exercise intensity, time window or machine learning methods.

[0112] Calculate the health status score S based on the health data in the feature vector h , and the formula is:

[0113]

[0114] where γ is the Gaussian weighting parameter, which is used to adjust the contribution of muscle activity to the score; y is the muscle activity intensity; f(y) is the muscle activity distribution function, which reflects the muscle activation state of the user and describes the distribution of muscle activation intensity in space or time. It can be fitted with a Gaussian distribution function, and the formula is:

[0115]

[0116] where μ is the central value of muscle activation and σ is the distribution width of muscle activity.

[0117] Furthermore, the muscle activity characteristics can also be divided into different parts (such as upper limbs, lower limbs), and the influence of muscle activity on the overall health status can be analyzed by region.

[0118] Calculate the overall score S of the user by integrating the comprehensive exercise performance score and the health status score. The formula is:

[0119]

[0120] where w m and w h are the weights of the exercise performance score and the health status score respectively, satisfying w m +w h =1; hj is the scenario-related health feature; η j is the corresponding scenario feature weight. For example, the exercise performance under high temperature may be more interfered; g(h j ) is the mapping function of the scenario feature, which can be implemented by a multi-layer neural network to more accurately quantify the impact of complex scenarios on the user's state. Its expression is:

[0121] g(h j ) = σ(W h h j + b h )

[0122] where, W h is the mapping weight matrix, b h is the bias term, and σ(·) is the activation function (such as ReLU).

[0123] It should be noted that the core of the feature vector design is to achieve the fusion and unified representation of multi-modal data, solving the problem of one-sided evaluation caused by a single data source in traditional methods. By adding angular data, muscle activity features and scenario features, the model can maintain the accuracy and robustness of the evaluation in complex scenarios.

[0124] Furthermore, the exercise performance score and the health status score are calculated separately, which not only maintains the independence of each dimension, but also can be integrated through the comprehensive score, improving the flexibility of the model. The integral scoring method captures the dynamic changes of the time series, overcoming the limitations of static scoring.

[0125] The weight parameters are dynamically adjusted by machine learning algorithms (such as Bayesian optimization or gradient descent), making the model highly adaptable to different user needs and scenarios, and also dynamically adjusting the proportion of different weights according to user preferences and usage scenarios.

[0126] Adding the scenario feature to the scoring formula solves the problem of insufficient environmental adaptability in traditional methods. For example, in scenarios with high temperature or high humidity, the scenario feature correction can significantly improve the accuracy of the scoring result.

[0127] S4: Generate personalized health feedback based on the evaluation results and provide real-time warnings.

[0128] According to the user's comprehensive score S and the health status score S h , the exercise performance score S m , combined with historical health data and specific scenario features, divide the user's state level;

[0129] According to the value range of the comprehensive score S, divide the user's health status and exercise performance levels;

[0130] When S < T1 When it is determined that the user is in an abnormal state, a health risk warning is generated, and the health status feedback mechanism is activated; when T 1 ≤S≤T 2 When it is determined that the user is in an optimizable state, exercise optimization suggestions are provided; when S>T 2 When it is determined that the user is in an excellent state, a status maintenance strategy is provided; where T 1 and T 2 are preset scoring thresholds.

[0131] Generate personalized feedback content, and the feedback content is pushed to the user terminal in the form of text, graphics, and voice, including specific health improvement suggestions, exercise plan adjustments, or health maintenance strategies;

[0132] Add an abnormal risk reminder to the feedback, explaining the existing risks and their impacts.

[0133] Furthermore, when the user's health status is evaluated as abnormal, the health feedback mechanism is activated, and the user's health status evaluation result is compared with the historical health record, and combined with the downward trend of the health status score S h to dynamically evaluate the health risk. Through recursive analysis of time series data, identify the key factors leading to the score decline. For example, if the heart rate variability (HRV) is abnormal and accompanied by a sharp drop in blood oxygen saturation (SpO2), the system will judge that there may be problems with the cardiopulmonary system.

[0134] According to the degree of abnormality, generate feedback content at different levels. The system divides abnormal situations into three levels:

[0135] Mild abnormality: For example, a slight heart rate fluctuation, and the feedback suggestions include short-term rest or reducing the exercise intensity;

[0136] Moderate abnormality: For example, a significant decrease in HRV but no other concurrent risks, and the feedback suggestions may include adjusting the exercise plan or strengthening heart rate monitoring after exercise;

[0137] Severe abnormality: For example, heart rate fluctuations accompanied by a sharp drop in blood oxygen level, immediately trigger a real-time warning and recommend stopping exercise, and at the same time recommend that the user seek medical examination.

[0138] By analyzing the user's personalized parameters (such as age, gender, health history) and real-time health data, combined with natural language processing (NLP) technology to optimize the expression of the feedback content to make it easier for users to understand. For example, for the feedback of mild abnormality, use the natural language description of "Your heart rate fluctuation may be due to fatigue in a short period of time. Please take appropriate rest", and for severe abnormality, issue a warning in the form of a warning message.

[0139] The system not only provides instant feedback but also generates long-term health improvement suggestions according to the type of anomaly, such as recommending specific forms of exercise (e.g., aerobic exercise) or healthy diet plans. The content of the suggestions is optimized through a cyclic learning algorithm to make subsequent feedback more in line with the actual needs of users.

[0140] Furthermore, the system monitors the user's exercise performance data in real time. For example, it detects abnormal exercise rhythms through the standard deviation of acceleration data and combines the dynamic frequency k i change rate to judge the user's exercise state. For each outlier, the system sets a dynamic threshold L v (t) to adjust the warning trigger standard in real time, ensuring the sensitivity and accuracy of the warning.

[0141] After detecting an anomaly, the system determines the source of the anomaly through a hierarchical analysis method. For example, if the change trend of displacement data is normal but the acceleration data shows a sudden change, it may be caused by the user's posture imbalance; if both are abnormal, it may be due to fatigue or excessive exercise.

[0142] The system classifies the warnings into three categories according to the degree of anomaly:

[0143] Level 1 warning: The amplitude of the user's movement is slightly abnormal but does not reach the danger threshold, and it is recommended to adjust the movement posture;

[0144] Level 2 warning: The anomaly is obvious. For example, the movement amplitude is too large and the frequency is abnormal, and it is recommended to reduce the intensity;

[0145] Level 3 warning: It may involve serious situations such as falling or extreme fatigue. The system immediately triggers an emergency notice and recommends stopping the exercise.

[0146] Through the feedback loop, the system records the user's response behavior to the warning. For example, when the user ignores a level 2 warning, the system will adjust the sensitivity and enforcement level of subsequent warnings, and at the same time analyze the user's exercise pattern to further optimize the feedback content. This mechanism combines real-time dynamic adjustment and hierarchical warning strategies, especially achieving dynamic optimization through the feedback loop on the basis of hierarchical processing, making the warning more accurate and adaptable.

[0147] It should also be noted that the user's exercise and health data are normal, but the environmental data is abnormal (such as high temperature, high humidity, or poor air quality). The system incorporates real-time environment into the comprehensive assessment scope and generates environmental perception feedback and warnings through intersection processing.

[0148] The system collects environmental temperature, humidity, and air quality index (AQI) and calculates the comprehensive environmental score E s :

[0149]

[0150] Among them, T is the temperature, and T w is the temperature weight, H is the humidity, R is the user's real-time exercise intensity, and the system determines the environmental suitability according to the E s value.

[0151] The system performs intersection analysis on environmental data and the user's exercise data. For example:

[0152] If the user's exercise intensity is low and the environmental score E s is high, it prompts "Your current environment is relatively suitable. Please continue to maintain it.";

[0153] If the user's exercise intensity is high and E s is low, it generates feedback of "It is recommended to reduce the exercise intensity or replenish water."

[0154] When E s < T e (the environmental score is lower than the threshold), the system triggers an early warning mechanism and generates specific suggestions by combining the user's health data and environmental data. For example:

[0155] In high temperature, it prompts "It is recommended to stop exercising and replenish water in time to prevent heatstroke.";

[0156] In the case of poor air quality, it prompts "The air quality is not good. It is recommended to reduce outdoor activities."

[0157] By circularly feeding back, the system records the user's environmental adaptability. For example, users who are long-term exposed to high-temperature environments may be less sensitive to temperature changes, and the system will adjust the temperature weight T w in the environmental score formula to generate feedback and early warnings that are more in line with the user's actual needs. This mechanism improves the system's comprehensive perception ability through intersection processing of environmental data and exercise health data, especially generating accurate feedback in real-time environmental anomalies.

[0158] S5: Dynamically update the parameters of the evaluation model and optimize the model according to the user's specific exercise scenarios and health needs.

[0159] Real-time collect the user's exercise data and health signals, and jointly form an extended training data set with historical data; data augmentation techniques (such as the SMOTE method) can be used to generate synthetic data when the samples are insufficient to ensure the diversity of training data.

[0160] It should be noted that the real-time collected data ensures that the input of the model is the latest. Combining historical data to expand the training data set improves the model's dynamic perception ability of the user's exercise state and health signals. The introduction of data augmentation techniques solves the problems of insufficient or unevenly distributed training data samples, and enhances the robustness and generalization ability of the model.

[0161] Dynamically adjust the model parameter set P = [w m , w h , η j , β i through the incremental learning algorithm.

[0162] It should be noted that the incremental learning algorithm avoids the problem of retraining due to data updates by gradually optimizing the model parameters, significantly reducing the computational cost of the model; this method improves the response speed of the model to real-time data changes, enabling it to adapt to the dynamic changes in the user's motion patterns and health status.

[0163] Use the cross-validation method to evaluate the performance of the updated model to ensure the accuracy and stability of the model; the cross-validation method provides a multi-dimensional evaluation of the model performance (such as accuracy, stability, and generalization ability), ensuring the effect after model optimization. It avoids the overfitting problem that may be caused by single data verification and improves the reliability of the model in practical applications.

[0164] Analyze the trend of feature changes of users in specific motion scenarios; the analysis of feature change trends can capture the user's behavior patterns at different time periods, such as the gradual increase in exercise intensity or the fluctuation of health signals. Combining with the scene feature set, the model input can be dynamically adjusted to enhance the model's context awareness ability.

[0165] Construct a scene feature set C = [c 1 , c 2 ,..., c n for a specific scenario and adjust the model input structure through scene awareness.

[0166] Use the objective function to guide the optimization process of the model parameters, dynamically adjust the model structure and parameter configuration to meet the user's health needs; the objective function combines the comprehensive scoring error term and the health target adaptability term to achieve multi-objective balance. Optimize the model performance by dynamically adjusting the weight coefficients.

[0167] Furthermore, the scene features include environmental features (temperature, humidity, air quality, wind speed, air pressure, weather conditions, light) and individual features (exercise mode, exercise intensity, temperature tolerance, fatigue, health status, exercise habits).

[0168] Even further, the scene awareness adjusts the calculation method of the model: based on a rule-based decision tree, define the environmental safety threshold and automatically adjust the score; based on Bayesian inference, calculate the exercise risk probability of different scenarios to optimize the evaluation results; based on neural network self-learning adjustment, use historical data to train the neural network to achieve personalized adjustment.

[0169] The finally optimized environmental scoring formula:

[0170]

[0171] Among them, f(c i ) is for personalized correction by combining different scenario features, optimizing the exercise suitability by combining environmental factors and individual characteristics, and its calculation method is as follows:

[0172] f(c i ) = w i ·g(c i )

[0173] Among them, c i is the scenario feature, including but not limited to environmental factors such as temperature, humidity, air quality, wind speed, etc., and individual characteristics such as the user's temperature adaptability, fatigue degree, exercise state, etc. w i is the adjustment coefficient of this feature, which changes dynamically according to the importance of the feature and the user's individual adaptability. g(c i ) is the adjustment function, which is used to perform a non-linear transformation on the feature c i so that its influence on the score conforms to the actual situation.

[0174] In actual calculation, g(c i ) can adopt different mathematical models, for example:

[0175] For some features (such as temperature, humidity), a quadratic function or a linear relationship can be used for adjustment to reflect the influence of the feature deviating from the optimal range on the score;

[0176] For environmental factors such as air quality, a smoothing function (such as the Sigmoid function) can be used to avoid drastic fluctuations in the score caused by mutations;

[0177] For user characteristics such as individual adaptability, it can be adaptively adjusted based on historical data to make the score more in line with the individual needs of the user.

[0178] Example 2 is the second example of the present invention. The difference from the previous example is:

[0179] If the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the current technical solution can be embodied in the form of a software product. The current computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0180] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or used in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0181] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part (electronic device) having one or more wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0182] Embodiment 3, refer to Figure 3 , which is an embodiment of the present invention, provides a motion health intelligent evaluation system based on multi-sensor fusion, including a data acquisition and preprocessing module, a data fusion and feature extraction module, an evaluation model construction and analysis module, a personalized feedback and warning module, and a model dynamic update and optimization module.

[0183] Data Acquisition and Preprocessing Module: Collect multi-modal motion data and physiological signals, and preprocess the collected data.

[0184] Data Fusion and Feature Extraction Module: Perform data fusion and extract key motion and health features.

[0185] Evaluation Model Construction and Analysis Module: Construct an intelligent motion health evaluation model to analyze the user's health status and motion performance.

[0186] Personalized Feedback and Warning Module: Generate personalized health feedback based on the evaluation results and provide real-time warnings.

[0187] Model Dynamic Update and Optimization Module: Dynamically update the parameters of the evaluation model and optimize the model according to the user's specific motion scenarios and health needs.

[0188] Example 4, an embodiment of the present invention, provides a method for intelligent evaluation of motion health based on multi-sensor fusion. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / contrast experiments.

[0189] In order to verify the effectiveness of the core technology of the present invention - evaluation through comprehensive motion performance scoring, the experiment conducted systematic tests on 6 test subjects (users A to F respectively). The experimental objective was to collect basic data, calculate the comprehensive motion performance score through the evaluation model, and compare it with the traditional single-index method.

[0190] The test subjects wore multi-sensor devices, including triaxial acceleration sensors, electromyography sensors, and heart rate monitors, while recording environmental data such as air quality (AQI), temperature, and humidity.

[0191] Data acquisition was divided into two parts: static (rest state) and dynamic (moderate-intensity exercise), and the sampling frequency was set to 10 times per second. The following preprocessing was performed on the collected data:

[0192] Remove outliers and fill in missing values.

[0193] Use a low-pass filtering method to denoise.

[0194] Normalize various types of data to eliminate dimensional differences.

[0195] By fusing motion features (acceleration, steps, electromyographic activity, etc.) and health features (heart rate, blood oxygen saturation, etc.), calculate the comprehensive motion performance score S m and the health status score S h . The model dynamically adjusted the weights of each score to adapt to the characteristics of different test subjects.

[0196] According to S m and Sh Generate user status levels (abnormal, optimized, excellent) based on the scoring range, and provide specific improvement suggestions. For example, provide breathing adjustment suggestions for users with high exercise intensity but low health scores; suggest increasing the activity frequency for users with low exercise intensity.

[0197] Specific experimental data content can be referred to in Table 1.

[0198] Table 1 Experimental Data Table

[0199]

[0200] It can be seen from the experimental data that user E has the highest exercise intensity (acceleration 10.2m / s 2 , steps 1180), and at the same time the comprehensive exercise performance score S m =9.0, indicating that their exercise efficiency is relatively high; while user C, although with a relatively low exercise intensity (8.9m / s 2 ), but through a relatively high muscle activity level (0.90mV) and health score S h =9.0, obtained a relatively high comprehensive score.

[0201] The health status score S of user A h =9.2, significantly higher than other users, indicating that their exercise intensity is moderate (acceleration 9.8m / s 2 ), but the health status is excellent.

[0202] The value range of the comprehensive score S is between 8.5 and 8.95. The score distribution of different users reflects the accurate trade-off of the model between exercise characteristics and health characteristics.

[0203] Different from traditional methods that only rely on single heart rate or step data, the present invention can comprehensively capture changes in exercise intensity and health status through the fusion and feature extraction of multi-modal data. The model of the present invention can adapt to different scenarios and individual characteristics by adjusting the weights w m and w h in real time, improving the accuracy and personalization of the score.

[0204] The present invention overcomes the problem of insufficient generalization ability of existing single-index methods in complex scenarios. The model score can dynamically adapt to environmental factors (such as temperature, humidity). It can generate scientific and accurate exercise and health feedback suggestions for users, improving the user experience and practical application value compared with traditional methods.

[0205] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered by the scope of the claims of the present invention.

Claims

1. A sports health intelligent assessment method based on multi-sensor fusion, characterized in that: include: Collect multimodal motion data and physiological signals, and pre-process the collected data; Perform data fusion and extract key sports and health features; Build an intelligent sports health assessment model to analyze the user's health status and sports performance; Generate personalized health feedback based on assessment results and provide real-time warnings; Dynamically update the parameters of the evaluation model and optimize the model according to the user's specific sports scenarios and health needs.

2. The sports health intelligent assessment method based on multi-sensor fusion as claimed in claim 1, characterized in that: The multimodal motion data includes acceleration data, angular velocity data, displacement trajectory data and muscle activity data; The physiological signals include collecting the user's heart rate data, blood oxygen saturation data, breathing rate data, skin temperature data and sweat conductivity data.

3. The sports health intelligent assessment method based on multi-sensor fusion as claimed in claim 2, characterized in that: The preprocessing of the collected data includes removing outliers and filling missing values ​​in the collected data, removing noise signals and filling missing data; De-noise and smooth the collected data to remove high-frequency interference and low-frequency drift; Normalize and standardize different types of data to eliminate differences in data dimensions; Extract frequency domain or time-frequency domain features to reveal implicit information in the data; Perform time stamp alignment and synchronization on the collected data to ensure the consistency of various data in the time dimension; Perform dimensionality reduction on high-dimensional data, extract the main feature components, and reduce the computational complexity.

4. The sports health intelligent assessment method based on multi-sensor fusion as claimed in claim 3, characterized in that: The data fusion and extraction of key sports and health features include classifying the pre-processed data according to the collection dimension and uniformly representing the same type of data; Analyze the correlation and complementarity of different types of sensor data and generate a unified feature data set through a specific fusion algorithm; Extract user trajectory features to analyze movement patterns and dynamic behaviors; Extract user posture features to evaluate posture stability and balance ability; Extract the user's muscle activity characteristics to analyze muscle activation level and fatigue status; Extract the user's heart rate characteristics to assess heart load and exercise intensity; Extract the user's blood oxygen characteristics to analyze oxygen utilization efficiency and endurance level; Extract the user's breathing characteristics to assess metabolic status and exercise recovery ability; Calculate the correlation coefficient between the feature and the evaluation target based on the statistical analysis method, and filter out the features with correlation coefficients below the predetermined threshold; The feature data set after reorganization and screening is generated to generate optimized feature inputs suitable for the intelligent evaluation model.

5. The sports health intelligent assessment method based on multi-sensor fusion as claimed in claim 4, characterized in that: The construction of the sports health intelligent assessment model includes constructing a feature vector: X=[a i ,b i ,k i ,v i ,x i ,θ,ψ] Among them, a i is the displacement data, describing the change of the user's spatial position; b i is acceleration data, reflecting the intensity and rate of change of movement; k i is the dynamic frequency, used to capture the rhythm of movement; v i is the speed data, indicating the linear change speed of the user's movement; x i is a health feature; θ is the user's angle data, which is used for posture stability analysis; ψ is the rate of change of posture angle, which describes the user's rotation or tilt dynamics; According to the sports data in the feature vector, calculate the user's comprehensive sports performance score S m , the formula is: Where N is the number of time segments, Δt is the interval between each time segment; β i is the weight parameter, which is determined by optimizing the training data; According to the health data in the feature vector, calculate the health status score S h , the formula is: Among them, γ is a Gaussian weighting parameter used to adjust the contribution of muscle activity to the score; f(y) is the muscle activity distribution function, which reflects the user's muscle activation state; y is the muscle activity intensity; The user's overall score S is calculated by combining the sports performance score and health status score. The formula is: Among them, w m and w h are the weights of sports performance score and health status score, respectively, satisfying w m +w h =1;h j is the scenario-related health feature; η j is the corresponding scene feature weight; g(h j ) is the mapping function of scene features.

6. The sports health intelligent assessment method based on multi-sensor fusion as claimed in claim 5, characterized in that: The generating of personalized health feedback based on the evaluation result includes: h , Sports performance score S m , combining historical health data and specific scenario characteristics to classify user status levels; According to the value range of the comprehensive score S, the user's health status and sports performance level are divided; When S<T1, the user is judged to be in an abnormal state, a health risk warning is generated, the user is notified to stop the current exercise and health check suggestions are provided. The cause of the abnormality is described in the feedback content, and mitigation measures are recommended; When T1≤S≤T2, the user is judged to be in an optimizable state and motion optimization suggestions are provided; When S>T2, the user is judged to be in a good state and a state maintenance strategy is provided; Among them, T1 and T2 are preset scoring thresholds; Generate personalized feedback content, which is pushed to the user terminal in the form of text, graphics and voice, including specific health improvement suggestions, exercise plan adjustments or health maintenance strategies; Include unusual risk warnings in feedback to explain the risks and their impacts.

7. The method for intelligent sports health assessment based on multi-sensor fusion according to claim 6, characterized in that: The parameters of the dynamic update evaluation model include real-time collection of the user's motion data and health signals, which together with historical data form an extended training data set; Dynamically adjust the model parameter set P = [w m , w h , η j , β i ]; Use cross-validation methods to evaluate the performance of the updated model to ensure the accuracy and stability of the model; Analyze the changing trends of users’ characteristics in specific sports scenarios; Construct a scene feature set C = [c1, c2, ..., c n ] and adjust the model input structure through scene perception; The objective function is used to guide the optimization process of model parameters, and the model structure and parameter configuration are dynamically adjusted to meet the health needs of users.

8. A sports health intelligent assessment system based on multi-sensor fusion, used to implement the sports health intelligent assessment method based on multi-sensor fusion as claimed in any one of claims 1 to 7, characterized in that: include: Data acquisition and preprocessing module: collect multimodal motion data and physiological signals, and preprocess the collected data; Data fusion and feature extraction module: performs data fusion and extracts key sports and health features; Evaluation model building and analysis module: build a sports health intelligent evaluation model to analyze the user's health status and sports performance; Personalized feedback and early warning module: Generate personalized health feedback based on the evaluation results and provide real-time early warning; Model dynamic update and optimization module: Dynamically update the parameters of the evaluation model and optimize the model according to the user's specific sports scenarios and health needs.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for intelligent sports health assessment based on multi-sensor fusion described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligent sports health assessment based on multi-sensor fusion according to any one of claims 1 to 7 are implemented.

Images