Methods and Systems for Sports and Health Monitoring Based on Smart Wearable Devices
Smart wearable devices that combine multi-type sensor data acquisition with a federated learning framework solve the problems of single data sets and sensor drift, achieving high-precision sports and health monitoring and personalized feedback, thus improving the user experience.
Patent Information
- Application Number
- CN202611269688.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing smart wearable devices for sports and health monitoring suffer from limitations such as single data dimensions, susceptibility to interference, sensor drift, inconsistent data quality, and a lack of multi-model fusion capabilities, making it impossible to achieve accurate and personalized feedback and real-time interaction.
Multiple types of sensors are used to simultaneously collect motion, physiological, and environmental parameters. Combined with wavelet transform, moving average filtering, and normalization processing, and through a federated learning framework, attention-enhanced Bi-LSTM neural network, large model, and weighted fusion algorithm, the AET and ANT thresholds are dynamically estimated to generate personalized exercise suggestions, which are then presented through AR interaction and voice assistant.
It achieves full coverage of multi-dimensional data, eliminates interference noise and sensor drift, improves data quality and analysis accuracy, protects user privacy, provides personalized and real-time motion guidance, and enhances user experience.
Smart Images

Figure CN123120296A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent interactive technology, and in particular to a method and system for monitoring sports and health based on intelligent wearable devices. Background Technology
[0002] Currently, smart wearable devices, with their portability, are widely used for sports data collection and health monitoring. However, existing technologies still have many shortcomings, making it difficult to meet users' needs for precise and personalized monitoring. Existing monitoring methods mostly rely on single sensors to collect data, focusing only on one type of exercise or physiological parameter. This results in a limited data dimension, failing to comprehensively reflect the relationship between exercise and health. Furthermore, the collected data is susceptible to motion interference and sensor drift, leading to inconsistent data quality and significant biases in subsequent analysis results. In terms of data processing and analysis, traditional methods often employ single algorithms, lacking multi-model fusion capabilities, and largely rely on cloud-based data aggregation and analysis, posing a risk of user privacy leaks. They also struggle to accurately guide users in adjusting exercise intensity, and personalized feedback interaction methods are limited, failing to achieve real-time interaction and dynamic guidance. Summary of the Invention
[0003] The purpose of this invention is to solve the above problems by designing a method and system for monitoring sports and health based on smart wearable devices.
[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned exercise and health monitoring method based on smart wearable devices: By using multiple types of sensors built into the user's smart wearable device, motion parameter data, physiological parameter data and environmental parameter data are collected simultaneously during the user's exercise process to obtain multi-dimensional raw data; The wavelet transform algorithm is used to eliminate motion interference noise from the collected multi-dimensional raw data. After the sensor drift error is corrected by moving average filtering, the data is normalized to obtain the preprocessed raw data. Based on the federated learning framework, combined with attention-enhanced Bi-LSTM neural network, large model and weighted fusion algorithm, multi-dimensional fusion analysis is performed on preprocessed raw data. The user's AET aerobic threshold and ANT anaerobic threshold are estimated by dynamic DDFA algorithm to obtain sports health monitoring results. Personalized exercise suggestions are generated based on sports and health monitoring results and fed back to the user's smart wearable device. These suggestions are then presented through an AR interactive module for visualization and voice assistant broadcasting. The device also receives interactive commands from the user through natural interaction methods for real-time interaction.
[0005] Furthermore, in the above-mentioned sports and health monitoring method based on smart wearable devices, the step of simultaneously collecting motion parameter data, physiological parameter data, and environmental parameter data during the user's exercise process through multiple types of sensors built into the user's smart wearable device yields multi-dimensional raw data, including: Motion parameter data, including user's motion posture, motion intensity, motion trajectory and motion type, are collected through the three-axis accelerometer, gyroscope and geomagnetic sensor built into the user's smart wearable device; Physiological parameter data are collected based on photoelectric heart rate sensors, blood oxygen sensors, and body temperature sensors, including real-time heart rate, blood oxygen saturation, body surface temperature, and heart rate variability during exercise. Environmental parameter data, including ambient temperature, relative humidity, and atmospheric pressure, are collected using the temperature and humidity sensors and barometric pressure sensors built into the user's smart wearable device.
[0006] Furthermore, in the above-mentioned sports and health monitoring method based on smart wearable devices, the preprocessed raw data is obtained by using a wavelet transform algorithm to eliminate motion interference noise, correcting sensor drift error through moving average filtering, and then normalizing the collected multi-dimensional raw data, including: The wavelet transform algorithm is adopted, and the db4 wavelet basis function is selected to perform 3-5 levels of wavelet decomposition on the original data, decomposing the data into low-frequency components and high-frequency components. By setting an adaptive threshold, the high-frequency components are thresholded, and then the signal is reconstructed through inverse wavelet transform to obtain the original data with noise reduction. Based on the moving average filtering algorithm, the sliding window size is set to 5-10 sampling points. The original data for noise reduction is processed by moving average filtering. The filtered original data is obtained by calculating the mean of adjacent sampling points. The min-max normalization method is used to map the filtered raw data to the [0,1] interval to obtain the preprocessed raw data.
[0007] Furthermore, in the aforementioned exercise and health monitoring method based on smart wearable devices, the federated learning framework, combined with attention-enhanced Bi-LSTM neural networks, large models, and weighted fusion algorithms, performs multi-dimensional fusion analysis on the preprocessed raw data. The dynamic DDFA algorithm is used to estimate the user's AET aerobic threshold and ANT anaerobic threshold to obtain the exercise and health monitoring results, including: Based on the federated learning framework, a client-server architecture is adopted, with smart wearable devices as clients and terminal devices as servers. The client is used for local data preprocessing and preliminary feature extraction, while the server is used for model parameter aggregation and updating.
[0008] Furthermore, in the aforementioned exercise and health monitoring method based on smart wearable devices, the federated learning framework, combined with attention-enhanced Bi-LSTM neural networks, large models, and weighted fusion algorithms, performs multi-dimensional fusion analysis on the preprocessed raw data. The dynamic DDFA algorithm is used to estimate the user's AET aerobic threshold and ANT anaerobic threshold to obtain the exercise and health monitoring results, including: Attention-enhanced Bi-LSTM neural networks are used to extract temporal features of data, capture the dynamic correlation between motion parameters and physiological parameters, and focus on key features through an attention mechanism; The ChatGLM large model is introduced to perform in-depth analysis of the extracted temporal features and identify user movement patterns and physiological states. A weighted fusion algorithm is used to weight and fuse the feature output of the attention-enhanced Bi-LSTM neural network with the analysis results of the large model to obtain a comprehensive feature vector.
[0009] Furthermore, in the aforementioned exercise and health monitoring method based on smart wearable devices, the federated learning framework, combined with attention-enhanced Bi-LSTM neural networks, large models, and weighted fusion algorithms, performs multi-dimensional fusion analysis on the preprocessed raw data. The dynamic DDFA algorithm is used to estimate the user's AET aerobic threshold and ANT anaerobic threshold to obtain the exercise and health monitoring results, including: The dynamic DDFA algorithm is used to estimate the user's AET aerobic threshold and ANT anaerobic threshold, and the relative relationship between the current exercise intensity and the AET aerobic threshold and ANT anaerobic threshold is output. By integrating comprehensive feature vectors, a multi-dimensional sports and health monitoring report is generated, including current exercise type, exercise intensity, real-time physiological status, AET aerobic threshold, ANT anaerobic threshold, exercise duration, energy consumption, and potential health risk warnings.
[0010] Furthermore, in the aforementioned sports and health monitoring method based on smart wearable devices, the step of generating personalized exercise suggestions based on the sports and health monitoring results, feeding these personalized exercise suggestions back to the user's smart wearable device, and presenting them through AR interaction module visualization and voice assistant broadcasting, while simultaneously receiving interactive commands issued by the user through natural interaction methods for real-time interaction, including: Based on the monitoring results, including the AET aerobic threshold, ANT anaerobic threshold, current exercise intensity, and user goals, personalized exercise suggestions are generated. Personalized exercise suggestions are synchronized to smart wearable devices and presented through an AR interactive module for visualization and voice assistant broadcast. The AR interactive module supports gesture swipes to switch between viewing details, and the broadcast frequency of the voice assistant is dynamically adjusted according to the exercise scenario.
[0011] Furthermore, in the sports and health monitoring system based on smart wearable devices, the sports and health monitoring system includes the following modules: The motion data acquisition module is used to simultaneously collect motion parameter data, physiological parameter data, and environmental parameter data of the user during exercise through multiple types of sensors built into the user's smart wearable device, so as to obtain multi-dimensional raw data. The data noise reduction and filtering module is used to eliminate motion interference noise from the acquired multi-dimensional raw data using wavelet transform algorithm, and then normalizes the data after correcting sensor drift error by moving average filtering to obtain preprocessed raw data. The sports and health monitoring module is used to perform multi-dimensional fusion analysis on preprocessed raw data based on the federated learning framework, combined with attention-enhanced Bi-LSTM neural network, large model and weighted fusion algorithm. It estimates the user's AET aerobic threshold and ANT anaerobic threshold through dynamic DDFA algorithm to obtain sports and health monitoring results. The sports virtual interaction module is used to generate personalized exercise suggestions based on sports and health monitoring results, and feed these personalized exercise suggestions back to the user's smart wearable device. The suggestions are then presented through the AR interaction module for visualization and voice assistant broadcasting, while also receiving interactive commands from the user through natural interaction methods for real-time interaction.
[0012] Furthermore, in the sports and health monitoring system based on smart wearable devices, the sports and health monitoring module includes the following sub-modules: The calculation submodule is used to estimate the user's AET aerobic threshold and ANT anaerobic threshold using the dynamic DDFA algorithm, and outputs the relative relationship between the current exercise intensity and the AET aerobic threshold and ANT anaerobic threshold. The integration submodule is used to integrate comprehensive feature vectors to generate multi-dimensional sports and health monitoring reports, including current exercise type, exercise intensity, real-time physiological status, AET aerobic threshold, ANT anaerobic threshold, exercise duration, energy consumption, and potential health risk warnings.
[0013] Furthermore, in the sports and health monitoring system based on smart wearable devices, the sports virtual interaction module includes the following sub-modules: The generation submodule is used to generate personalized exercise suggestions based on the AET aerobic threshold, ANT anaerobic threshold, current exercise intensity and user goals in the monitoring results; The interactive submodule is used to synchronize personalized exercise suggestions to smart wearable devices. It is presented through the AR interactive module for visualization and the voice assistant for broadcast. The AR interactive module supports gesture swipe to switch and view details, and dynamically adjusts the broadcast frequency of the voice assistant according to the exercise scenario.
[0014] Its beneficial effects lie in achieving comprehensive multi-dimensional data coverage by simultaneously collecting motion, physiological, and environmental parameters using multiple types of sensors. This overcomes the limitations of traditional single-data collection, providing a more comprehensive and realistic reflection of the user's exercise and health status, and offering solid data support for subsequent analysis. A three-step preprocessing process—wavelet transform denoising, moving average filtering correction, and normalization—effectively eliminates interference noise and sensor drift errors, unifies data dimensions, significantly improves data quality, and ensures the accuracy of subsequent analysis results. Based on a federated learning framework, user data is stored and analyzed locally. Combined with encrypted communication protocols, user privacy is effectively protected. Furthermore, the fusion analysis of attention-enhanced Bi-LSTM neural networks and large models improves the accuracy of feature extraction and state recognition. The dynamic DDFA algorithm accurately estimates AET and ANT thresholds, providing a scientific basis for adjusting user exercise intensity. Personalized suggestions are generated based on user exercise goals and presented through dual channels of AR visualization and voice broadcasting, coupled with natural interaction methods for real-time response, enhancing the user experience. Attached Figure Description
[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.
[0016] Figure 1 This is a schematic diagram of the first embodiment of the sports and health monitoring method based on smart wearable devices in this invention; Figure 2 This is a schematic diagram of a second embodiment of the sports and health monitoring method based on a smart wearable device according to the present invention; Figure 3 This is a schematic diagram of the first embodiment of the sports and health monitoring system based on smart wearable devices in this invention. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms "one," "an," and "this" used herein may also include the plural forms. It should be further understood that the terminology used in this specification includes the presence of features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.
[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, a method for monitoring exercise and health based on smart wearable devices includes the following steps: Step 101: Through the multiple types of sensors built into the user's smart wearable device, synchronously collect motion parameter data, physiological parameter data and environmental parameter data during the user's exercise process to obtain multi-dimensional raw data; Specifically, in this embodiment, motion parameter data, including user posture, exercise intensity, exercise trajectory, and exercise type, are collected using the triaxial accelerometer, gyroscope, and geomagnetic sensor built into the user's smart wearable device; physiological parameter data, including real-time heart rate, blood oxygen saturation, body surface temperature, and heart rate variability during exercise, are collected using the photoelectric heart rate sensor, blood oxygen sensor, and body temperature sensor; and environmental parameter data, including ambient temperature, relative humidity, and atmospheric pressure, are collected using the temperature and humidity sensor and barometric pressure sensor built into the user's smart wearable device.
[0020] By utilizing multiple types of sensors built into the user's smart wearable device, and dynamically adjusting the sampling frequency according to the type of exercise (10Hz-50Hz for regular exercise, and 50Hz-100Hz for high-intensity exercise), three core data points are collected simultaneously during the user's exercise process. This ensures the integrity, timeliness, and synchronization of the data, ultimately resulting in a multi-dimensional raw data set. The specific data collected is as follows: Motion parameter data: collected through a three-axis accelerometer, gyroscope, and geomagnetic sensor, including user motion posture, pitch angle, roll angle, heading angle, motion intensity, peak acceleration, root mean square value, motion trajectory, cadence, stride length, step length, cumulative steps, and motion type, such as walking, running, cycling, and swimming. The data is initially determined based on the characteristics of the sensor data. Physiological parameter data are collected through photoelectric heart rate sensor, blood oxygen sensor, and body temperature sensor, including real-time heart rate, resting heart rate, peak heart rate during exercise, average heart rate, blood oxygen saturation, SpO2, body surface temperature, heart rate variability during exercise, and HRV. The heart rate data sampling frequency is no less than 10Hz to ensure the capture of real-time fluctuations in heart rate. Environmental parameter data is collected through the device's built-in temperature and humidity sensors and air pressure sensors, including ambient temperature, relative humidity, and atmospheric pressure, to help analyze the impact of environmental factors on the user's exercise status and physiological indicators. The sampling frequency is 1Hz-5Hz.
[0021] Step 102: The collected multi-dimensional raw data is processed by wavelet transform algorithm to eliminate motion interference noise, and then normalized after correcting sensor drift error by moving average filtering to obtain preprocessed raw data. Specifically, in this embodiment, a wavelet transform algorithm is used, with the db4 wavelet basis function selected. The original data is decomposed into low-frequency and high-frequency components by 3-5 levels of wavelet decomposition. An adaptive threshold is set to perform threshold processing on the high-frequency components. The signal is then reconstructed through inverse wavelet transform to obtain the denoised original data. Based on the moving average filtering algorithm, the sliding window size is set to 5-10 sampling points. The denoised original data is processed by moving average filtering. The filtered original data is obtained by calculating the mean of adjacent sampling points. The min-max normalization method is used to map the filtered original data to the [0,1] interval to obtain the preprocessed original data.
[0022] To address issues such as motion interference noise, sensor drift error, and inconsistent data dimensions in the collected multi-dimensional raw data, a three-step preprocessing workflow of denoising, correction, and normalization is adopted to eliminate invalid interference, improve data quality, and obtain standardized preprocessed raw data. Motion interference noise elimination employs a wavelet transform algorithm, using the db4 wavelet basis function to perform 3-5 levels of wavelet decomposition on the raw data, decomposing the data into low-frequency components, effective signals, high-frequency components, and interference noise. By setting an adaptive threshold and dynamically adjusting it based on the data variance, threshold processing is performed on the high-frequency components, followed by soft threshold compression. Then, the signal is reconstructed through inverse wavelet transform, effectively eliminating interference noise generated by equipment shaking and muscle tremors during motion. The denoising effect is particularly optimized for physiological parameters that are easily affected by interference, such as heart rate and blood oxygen. Sensor drift error correction uses a moving average filtering algorithm, with the sliding window size set to 5-10 sampling points and dynamically adjusted according to the sampling frequency. The higher the sampling frequency, the larger the window size is appropriately increased. The denoised data stream is processed by sliding filtering, and the drift error caused by long-term operation of the sensor is corrected by calculating the average of adjacent sampling points. This avoids data deviation caused by slight offset of the device wearing position, sensor aging, etc., and ensures data stability. Data normalization was performed using the min-max normalization method, which mapped all corrected data to the [0,1] interval, eliminating the differences in the data units of different parameters, such as heart rate range of 60-180 beats / minute, cadence range of 60-180 steps / minute, and ambient temperature range of 0-40℃, thereby improving the accuracy of the analysis.
[0023] After preprocessing, the data is validated to remove outliers, such as heart rate exceeding 220 beats / minute or below 40 beats / minute, or blood oxygen below 90%. If the proportion of outliers exceeds 5%, a re-collection mechanism is triggered to ensure the validity of the preprocessed data.
[0024] Step 103: Based on the federated learning framework, combined with attention-enhanced Bi-LSTM neural network, large model and weighted fusion algorithm, perform multi-dimensional fusion analysis on the preprocessed raw data, estimate the user's AET aerobic threshold and ANT anaerobic threshold through dynamic DDFA algorithm, and obtain the sports health monitoring results. Specifically, in this embodiment, a client-server architecture is adopted based on the federated learning framework. The smart wearable device acts as the client, and the terminal device acts as the server. The client is used for local data preprocessing and preliminary feature extraction, while the server is used for model parameter aggregation and updating.
[0025] Attention-enhanced Bi-LSTM neural networks are used to extract temporal features from data, capture the dynamic correlation between motion parameters and physiological parameters, and focus on key features through an attention mechanism. A ChatGLM large model is introduced to perform in-depth analysis of the extracted temporal features and identify user motion patterns and physiological states. A weighted fusion algorithm is used to weight and fuse the feature output of the attention-enhanced Bi-LSTM neural network with the analysis results of the large model to obtain a comprehensive feature vector.
[0026] The system uses a dynamic DDFA algorithm to estimate the user's AET (Aerobic Equivalent Threshold) and ANT (Antibiotic Threshold), and outputs the relative relationship between the current exercise intensity and the AET and ANT thresholds. It integrates comprehensive feature vectors to generate a multi-dimensional exercise health monitoring report, including the current exercise type, exercise intensity, real-time physiological status, AET threshold, ANT threshold, exercise duration, energy consumption, and potential health risk warnings.
[0027] It adopts a client-server, C / S architecture, with smart wearable devices as clients and mobile phones as servers. The client is responsible for local data preprocessing and preliminary feature extraction, while the server is responsible for model parameter aggregation and updating. There is no need to upload the user's original data to the cloud, thus protecting user privacy. At the same time, the client and server transmit model parameters through encrypted communication protocols, such as AES encryption, to avoid parameter leakage. Attention-enhanced Bi-LSTM neural network: used to extract temporal features of data, focusing on capturing the dynamic correlation between motion parameters and physiological parameters, such as the change pattern of heart rate with cadence and exercise intensity. It focuses on key features through attention mechanism, such as heart rate data at the peak of exercise and abnormal blood oxygen fluctuation data, to improve the targeting of feature extraction and avoid interference from invalid features. We introduce a lightweight sports and health model, adapt to the computing power of wearable devices, and perform in-depth analysis on the extracted time-series features to identify user exercise patterns, such as steady-state running, interval running, and hill climbing, as well as physiological states, such as fatigue level and hypoxia. At the same time, we combine user historical exercise data and basic health information, such as age, gender, height, and weight, to optimize the analysis accuracy. The feature outputs of the attention-enhanced Bi-LSTM neural network are weighted and fused with the analysis results of the large model. The weights are dynamically adjusted according to the data credibility. For example, the credibility of physiological parameters is higher than that of environmental parameters, so the weight accounts for 60%-70%; the weight of motion parameters accounts for 20%-25%; and the weight of environmental parameters accounts for 5%-10%. The comprehensive feature vector is obtained by weighted summation. Based on comprehensive feature vectors, the dynamic DDFA algorithm is adopted to dynamically adjust algorithm parameters, such as the slope of heart rate change and threshold judgment threshold, by monitoring the correlation between heart rate changes, heart rate variability and exercise intensity during the user's exercise in real time. This accurately estimates the user's AET, aerobic threshold (the critical point of maximum exercise intensity when aerobic metabolism is dominant) and ANT, and anaerobic threshold (the critical point when anaerobic metabolism begins to dominate during exercise). At the same time, it outputs the relative relationship between the current exercise intensity and AET and ANT, such as below AET, between AET and ANT, and above ANT. By integrating the above analysis results, a multi-dimensional sports health monitoring report is generated, including the current type of exercise, exercise intensity, real-time physiological status, heart rate, blood oxygen, body temperature, AET / ANT threshold, exercise duration, energy consumption, and potential health risk warnings, such as abnormal heart rate, low blood oxygen, and over-exercising.
[0028] Step 104: Generate personalized exercise suggestions based on the results of sports and health monitoring, and feed the personalized exercise suggestions back to the user's smart wearable device. The suggestions are presented through the AR interaction module for visualization and voice assistant broadcasting, and the device can also receive interactive commands issued by the user through natural interaction for real-time interaction.
[0029] Specifically, in this embodiment, personalized exercise suggestions are generated based on the AET aerobic threshold, ANT anaerobic threshold, current exercise intensity, and user goals from the monitoring results. The personalized exercise suggestions are synchronized to the smart wearable device and presented through the AR interaction module for visualization and voice assistant broadcast. The AR interaction module supports gesture swipes to switch and view details, and the broadcast frequency of the voice assistant is dynamically adjusted according to the exercise scenario.
[0030] Based on the above sports and health monitoring results, combined with the user's preset exercise goals, such as fat loss, muscle gain, endurance improvement, basic health information, and historical exercise data, personalized exercise suggestions are generated. These suggestions are fed back to the user's smart wearable device through multiple channels and support real-time interaction. Personalized exercise suggestions are generated as follows: Based on the AET / ANT thresholds in the monitoring results, the current exercise intensity, and the user's goals, targeted suggestions are generated. For example: if the current exercise intensity is lower than AET, it is recommended to appropriately increase the cadence / speed to maintain aerobic metabolism; if it is between AET and ANT, it is recommended to maintain the current intensity and avoid over-exercising; if it is higher than ANT, it is recommended to reduce the intensity to prevent excessive anaerobic metabolism that could lead to fatigue or injury. At the same time, combined with energy consumption and exercise duration, reasonable exercise duration, rest intervals, and timing for hydration / energy replenishment are suggested. We recommend presenting personalized exercise suggestions in multiple formats: synchronize these suggestions to smart wearable devices and present them in two ways: visual display via an AR interactive module, where monitoring results and suggestions are presented intuitively on the device screen in the form of dynamic charts, such as heart rate curves, exercise intensity bar charts, and AET / ANT threshold labels, with support for gesture swipes to switch and view details; and voice assistant broadcast, which dynamically adjusts the broadcast frequency according to the exercise scenario, with concise and clear broadcast content. The device supports users to issue interactive commands through natural interaction methods, including voice commands and gesture commands. After receiving the command, the device responds quickly through local computing power, adjusts the displayed content or broadcast mode in real time, and synchronizes the user interaction records to the terminal device for subsequent optimization of the accuracy of personalized suggestions.
[0031] Its beneficial effects lie in achieving comprehensive multi-dimensional data coverage by simultaneously collecting motion, physiological, and environmental parameters using multiple types of sensors. This overcomes the limitations of traditional single-data collection, providing a more comprehensive and realistic reflection of the user's exercise and health status, and offering solid data support for subsequent analysis. A three-step preprocessing process—wavelet transform denoising, moving average filtering correction, and normalization—effectively eliminates interference noise and sensor drift errors, unifies data dimensions, significantly improves data quality, and ensures the accuracy of subsequent analysis results. Based on a federated learning framework, user data is stored and analyzed locally. Combined with encrypted communication protocols, user privacy is effectively protected. Furthermore, the fusion analysis of attention-enhanced Bi-LSTM neural networks and large models improves the accuracy of feature extraction and state recognition. The dynamic DDFA algorithm accurately estimates AET and ANT thresholds, providing a scientific basis for adjusting user exercise intensity. Personalized suggestions are generated based on user exercise goals and presented through dual channels of AR visualization and voice broadcasting, coupled with natural interaction methods for real-time response, enhancing the user experience.
[0032] Please see Figure 2In the sports and health monitoring method based on smart wearable devices, the preprocessed raw data is obtained by using wavelet transform algorithm to eliminate motion interference noise, correcting sensor drift error through moving average filtering, and then normalizing. The process includes the following steps. Step 201: Using the wavelet transform algorithm, select the db4 wavelet basis function to perform 3-5 levels of wavelet decomposition on the original data, decompose the data into low-frequency components and high-frequency components, and perform threshold processing on the high-frequency components by setting an adaptive threshold. Then, reconstruct the signal through inverse wavelet transform to obtain the original data with noise reduction. Step 202: Based on the moving average filtering algorithm, set the sliding window size to 5-10 sampling points, perform moving average filtering on the original data to be denoised, and obtain the filtered original data by calculating the mean of adjacent sampling points; Step 203: Using the min-max normalization method, the filtered raw data is mapped to the [0,1] interval to obtain the preprocessed raw data.
[0033] The above describes embodiments of the exercise and health monitoring method based on smart wearable devices of the present invention. Please refer to [link / reference]. Figure 3 In a sports and health monitoring system based on smart wearable devices, the system includes the following modules: The motion data acquisition module is used to simultaneously collect motion parameter data, physiological parameter data, and environmental parameter data of the user during exercise through multiple types of sensors built into the user's smart wearable device, so as to obtain multi-dimensional raw data. The data noise reduction and filtering module is used to eliminate motion interference noise from the acquired multi-dimensional raw data using wavelet transform algorithm, and then normalizes the data after correcting sensor drift error by moving average filtering to obtain preprocessed raw data. The sports and health monitoring module is used to perform multi-dimensional fusion analysis on preprocessed raw data based on the federated learning framework, combined with attention-enhanced Bi-LSTM neural network, large model and weighted fusion algorithm. It estimates the user's AET aerobic threshold and ANT anaerobic threshold through dynamic DDFA algorithm to obtain sports and health monitoring results. The sports virtual interaction module is used to generate personalized exercise suggestions based on sports and health monitoring results, and feed these personalized exercise suggestions back to the user's smart wearable device. The suggestions are then presented through the AR interaction module for visualization and voice assistant broadcasting, while also receiving interactive commands from the user through natural interaction methods for real-time interaction.
[0034] Specifically, this embodiment also includes smart wearable devices such as AI watches, smart headphones, and AR glasses, such as MetaRay-Ban. These smart wearable devices utilize multiple built-in sensors to simultaneously collect motion parameter data, physiological parameter data, and environmental parameter data during the user's exercise process. Motion parameter data is collected by an inertial measurement unit, including acceleration, angular velocity, gait cycle, and stride frequency and length. Physiological parameter data includes heart rate, heart rate variability, blood oxygen saturation, and skin temperature. Heart rate data is collected using photoplethysmography (PPG) and undergoes motion artifact elimination. Environmental parameter data includes ambient temperature and humidity, altitude, and air pressure. Simultaneously, a health monitoring module on the device enables real-time collection and preliminary analysis of physiological parameters, linking with a voice assistant module and an AR interaction module to provide data support for subsequent interactive feedback. The collected multi-dimensional raw data undergoes noise reduction, normalization, and outlier removal. Wavelet transform algorithm is used to eliminate motion interference noise, and moving average filtering is used to correct sensor drift error. Simultaneously, based on the user's static physiological baseline and motion scenario, the sensor acquisition accuracy is dynamically calibrated. Dynamic calibration includes figure-eight posture calibration of the triaxial accelerometer and adaptive calibration of the optical heart rate sensor's wearing position. The calibration process can be triggered through the device's natural interaction module, responding to user operation commands. Employing a federated learning framework, combined with attention-enhanced BiLSTM neural networks, large models, and weighted fusion algorithms, this system performs multi-dimensional fusion analysis on preprocessed exercise, physiological, and environmental data to achieve user exercise pattern recognition, exercise intensity assessment, and health status quantification. Specifically, the system uses a dynamic DDFA algorithm to estimate users' aerobic thresholds (AET), anaerobic thresholds (ANT), and lactate thresholds, replacing traditional laboratory lactate testing methods. The large model supports chat, storytelling, and answering users' exercise and health-related questions, enhancing the interactive experience. Based on the results of the fusion analysis, a personalized threshold model is constructed by combining the user's exercise level and past health data to determine in real time whether the user has excessive exercise, sports injury, or health abnormality risk. When the detected index exceeds the corresponding threshold, a graded warning message is generated and simultaneously reminded through four methods: wearable device vibration, voice assistant broadcast, AR interactive display, and push notification from the associated terminal APP. Users can query the warning details and related health advice through natural interaction. The system feeds back exercise and health monitoring results, threshold assessment reports, and personalized exercise suggestions to user terminals and wearable devices, simultaneously presenting them through AR interactive modules for visualization and voice assistant broadcasting. It also receives interactive commands from users via natural interaction methods, including far-field voice, multi-turn voice, dialect voice, touch, facial expressions, and body movements. Based on user commands, it adaptively adjusts sensor sampling frequency, data fusion weights, and warning thresholds to dynamically match monitoring strategies with user exercise status and health needs. Personalized exercise suggestions are generated based on user exercise effect assessments and recovery time predictions, including adjustments to exercise intensity, rest duration, and guidance on proper movement. Users can further consult relevant details through the large model.
[0035] Multiple types of sensors include inertial measurement units (IMUs), PPG heart rate sensors, blood oxygen sensors, skin temperature sensors, barometric pressure sensors, and temperature and humidity sensors. The inertial measurement unit consists of an accelerometer, a gyroscope, and a magnetometer. The sampling frequency is dynamically adjusted according to the intensity of the exercise, with a sampling frequency of 200Hz for low-intensity exercise and increasing to 1000Hz for high-intensity exercise. AI watches, smart headphones, and AR glasses, such as the Meta Ray-Ban, can achieve synchronous acquisition of sensor data, complementing each other and improving the comprehensiveness of monitoring.
[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and equivalents.
Claims
1. A method for monitoring exercise and health based on smart wearable devices, characterized in that, The exercise health monitoring method includes the following steps: By using multiple types of sensors built into the user's smart wearable device, motion parameter data, physiological parameter data and environmental parameter data are collected simultaneously during the user's exercise process to obtain multi-dimensional raw data; The wavelet transform algorithm is used to eliminate motion interference noise from the collected multi-dimensional raw data. After the sensor drift error is corrected by moving average filtering, the data is normalized to obtain the preprocessed raw data. Based on the federated learning framework, combined with attention-enhanced Bi-LSTM neural network, large model and weighted fusion algorithm, multi-dimensional fusion analysis is performed on preprocessed raw data. The user's AET aerobic threshold and ANT anaerobic threshold are estimated by dynamic DDFA algorithm to obtain sports health monitoring results. Personalized exercise suggestions are generated based on sports and health monitoring results and fed back to the user's smart wearable device. These suggestions are then presented through an AR interactive module for visualization and voice assistant broadcasting. The device also receives interactive commands from the user through natural interaction methods for real-time interaction.
2. The exercise and health monitoring method based on a smart wearable device as described in claim 1, characterized in that, The system uses multiple types of sensors built into the user's smart wearable device to simultaneously collect motion parameter data, physiological parameter data, and environmental parameter data during the user's exercise process, obtaining multi-dimensional raw data, including: Motion parameter data, including user's motion posture, motion intensity, motion trajectory and motion type, are collected through the three-axis accelerometer, gyroscope and geomagnetic sensor built into the user's smart wearable device; Physiological parameter data are collected based on photoelectric heart rate sensors, blood oxygen sensors, and body temperature sensors, including real-time heart rate, blood oxygen saturation, body surface temperature, and heart rate variability during exercise. Environmental parameter data, including ambient temperature, relative humidity, and atmospheric pressure, are collected using the temperature and humidity sensors and barometric pressure sensors built into the user's smart wearable device.
3. The exercise and health monitoring method based on a smart wearable device as described in claim 1, characterized in that, The collected multi-dimensional raw data is processed by using wavelet transform algorithm to eliminate motion interference noise, and then normalized after correcting sensor drift error by moving average filtering to obtain preprocessed raw data, including: The wavelet transform algorithm is adopted, and the db4 wavelet basis function is selected to perform 3-5 levels of wavelet decomposition on the original data, decomposing the data into low-frequency components and high-frequency components. By setting an adaptive threshold, the high-frequency components are thresholded, and then the signal is reconstructed through inverse wavelet transform to obtain the original data with noise reduction. Based on the moving average filtering algorithm, the sliding window size is set to 5-10 sampling points. The original data for noise reduction is processed by moving average filtering. The filtered original data is obtained by calculating the mean of adjacent sampling points. The min-max normalization method is used to map the filtered raw data to the [0,1] interval to obtain the preprocessed raw data.
4. The exercise and health monitoring method based on a smart wearable device as described in claim 1, characterized in that, The federated learning framework, combined with attention-enhanced Bi-LSTM neural networks, large models, and weighted fusion algorithms, performs multi-dimensional fusion analysis on preprocessed raw data. The dynamic DDFA algorithm estimates the user's AET (Aerobic Equivalent Threshold) and ANT (Anterobic Threshold) to obtain exercise health monitoring results, including: Based on the federated learning framework, a client-server architecture is adopted, with smart wearable devices as clients and terminal devices as servers. The client is used for local data preprocessing and preliminary feature extraction, while the server is used for model parameter aggregation and updating.
5. The exercise and health monitoring method based on a smart wearable device as described in claim 1, characterized in that, The federated learning framework, combined with attention-enhanced Bi-LSTM neural networks, large models, and weighted fusion algorithms, performs multi-dimensional fusion analysis on preprocessed raw data. The dynamic DDFA algorithm estimates the user's AET (Aerobic Equivalent Threshold) and ANT (Anterobic Threshold) to obtain exercise health monitoring results, including: Attention-enhanced Bi-LSTM neural networks are used to extract temporal features of data, capture the dynamic correlation between motion parameters and physiological parameters, and focus on key features through an attention mechanism; The ChatGLM large model is introduced to perform in-depth analysis of the extracted temporal features and identify user movement patterns and physiological states. A weighted fusion algorithm is used to weight and fuse the feature output of the attention-enhanced Bi-LSTM neural network with the analysis results of the large model to obtain a comprehensive feature vector.
6. The exercise and health monitoring method based on a smart wearable device as described in claim 1, characterized in that, The federated learning framework, combined with attention-enhanced Bi-LSTM neural networks, large models, and weighted fusion algorithms, performs multi-dimensional fusion analysis on preprocessed raw data. The dynamic DDFA algorithm estimates the user's AET (Aerobic Equivalent Threshold) and ANT (Anterobic Threshold) to obtain exercise health monitoring results, including: The dynamic DDFA algorithm is used to estimate the user's AET aerobic threshold and ANT anaerobic threshold, and the relative relationship between the current exercise intensity and the AET aerobic threshold and ANT anaerobic threshold is output. By integrating comprehensive feature vectors, a multi-dimensional sports and health monitoring report is generated, including current exercise type, exercise intensity, real-time physiological status, AET aerobic threshold, ANT anaerobic threshold, exercise duration, energy consumption, and potential health risk warnings.
7. The exercise and health monitoring method based on a smart wearable device as described in claim 1, characterized in that, The system generates personalized exercise suggestions based on sports and health monitoring results, feeds these suggestions back to the user's smart wearable device, and presents them through AR interaction module visualization and voice assistant broadcasting. Simultaneously, it receives interactive commands from the user via natural interaction methods for real-time interaction, including: Based on the monitoring results, including the AET aerobic threshold, ANT anaerobic threshold, current exercise intensity, and user goals, personalized exercise suggestions are generated. Personalized exercise suggestions are synchronized to smart wearable devices and presented through an AR interactive module for visualization and voice assistant broadcast. The AR interactive module supports gesture swipes to switch between viewing details, and the broadcast frequency of the voice assistant is dynamically adjusted according to the exercise scenario.
8. A sports and health monitoring system based on smart wearable devices, characterized in that, The sports and health monitoring system includes the following modules: The motion data acquisition module is used to simultaneously collect motion parameter data, physiological parameter data, and environmental parameter data of the user during exercise through multiple types of sensors built into the user's smart wearable device, so as to obtain multi-dimensional raw data. The data noise reduction and filtering module is used to eliminate motion interference noise from the acquired multi-dimensional raw data using wavelet transform algorithm, and then normalizes the data after correcting sensor drift error by moving average filtering to obtain preprocessed raw data. The sports and health monitoring module is used to perform multi-dimensional fusion analysis on preprocessed raw data based on the federated learning framework, combined with attention-enhanced Bi-LSTM neural network, large model and weighted fusion algorithm. It estimates the user's AET aerobic threshold and ANT anaerobic threshold through dynamic DDFA algorithm to obtain sports and health monitoring results. The sports virtual interaction module is used to generate personalized exercise suggestions based on sports and health monitoring results, and feed these personalized exercise suggestions back to the user's smart wearable device. The suggestions are then presented through the AR interaction module for visualization and voice assistant broadcasting, while also receiving interactive commands from the user through natural interaction methods for real-time interaction.
9. The sports and health monitoring system based on intelligent wearable devices as described in claim 8, characterized in that, The sports and health monitoring module includes the following sub-modules: The calculation submodule is used to estimate the user's AET aerobic threshold and ANT anaerobic threshold using the dynamic DDFA algorithm, and outputs the relative relationship between the current exercise intensity and the AET aerobic threshold and ANT anaerobic threshold. The integration submodule is used to integrate comprehensive feature vectors to generate multi-dimensional sports and health monitoring reports, including current exercise type, exercise intensity, real-time physiological status, AET aerobic threshold, ANT anaerobic threshold, exercise duration, energy consumption, and potential health risk warnings.
10. The sports and health monitoring system based on intelligent wearable devices as described in claim 8, characterized in that, The motion virtual interaction module includes the following sub-modules: The generation submodule is used to generate personalized exercise suggestions based on the AET aerobic threshold, ANT anaerobic threshold, current exercise intensity, and user goals in the monitoring results. The interactive submodule is used to synchronize personalized exercise suggestions to smart wearable devices. It is presented through the AR interactive module for visualization and the voice assistant for broadcast. The AR interactive module supports gesture swipe to switch and view details, and dynamically adjusts the broadcast frequency of the voice assistant according to the exercise scenario.
