An exercise index generation method and system for AI exercise health assessment

Abstract

The present application relates to a kind of movement index generation method and system for AI sports health assessment.The method collects the multi-source movement data of intelligent sports equipment and identifies aerobic exercise and anaerobic exercise;For aerobic data, identify effective steady-state interval, establish heart rate time response relationship and individualized heart rate-power response relationship, generate maximum oxygen uptake and aerobic capacity percentile;For anaerobic data, generate single maximum load or strength capacity representative value and anaerobic capacity percentile;Each movement index, percentile, data confidence, data source, generation time and missing mark are combined to form standardized movement ability characteristic data, and output to artificial intelligence sports health assessment module, to provide uniform and traceable input data for downstream assessment.

Description

Technical Field

[0001] This invention relates to the fields of intelligent sports equipment, sports data processing and artificial intelligence application data interface technology, and in particular to a method and system for generating sports indicators for AI-based sports health assessment. Background Technology

[0002] With the development of smart sports hardware and digital health technology, aerobic fitness equipment such as treadmills, exercise bikes, and rowing machines, as well as electronic resistance training devices and wearable heart rate monitoring devices, can now collect multi-dimensional data in real time, including speed, incline, resistance, power, heart rate, number of repetitions, movement speed, and training duration. Artificial intelligence-based sports and health assessment systems require standardized features that reflect a user's aerobic and anaerobic capacity. However, differences exist in data fields, units, sampling frequencies, and exercise types across different devices, making it difficult to directly use raw device data as a unified assessment input. Existing technologies mainly suffer from the following problems: 1. Multi-source exercise data are isolated from each other and lack unified exercise ability indicators and standardized output structure: Existing smart fitness equipment mainly records and displays data collected by itself. The power, resistance, speed, heart rate and movement data of different devices lack unified technical meaning, data quality identification and time information. Downstream evaluation modules need to repeatedly perform data cleaning and feature interpretation, and it is difficult to make horizontal comparisons between aerobic and anaerobic capacity.

[0003] 2. Insufficient stability and scenario adaptability in generating aerobic exercise capacity indicators: Maximum oxygen uptake (VO2max) is an important indicator for evaluating cardiopulmonary aerobic capacity. Some existing solutions estimate it using empirical formulas based solely on basic parameters such as age and resting heart rate; others are geared towards specific cycling scenarios, fitting or indirectly extrapolating relevant capacity parameters based on power and heart rate data. These solutions may not adequately identify the time intervals where power is relatively stable and the exercise load reaches an effective level, making them susceptible to the effects of heart rate increases, power fluctuations, and abnormal equipment data at the beginning of exercise. Furthermore, some solutions only output a single value, lacking information such as population percentiles, data confidence levels, equipment type, and calibration version, which can be used for downstream evaluation.

[0004] 3. Anaerobic strength performance indicators are limited in scope and difficult to adapt to home low-resistance equipment scenarios: Current strength performance assessments are mostly based on 1RM (one-repetition maximum load) or empirical formulas based on training resistance and repetitions. For home electronic strength training equipment with limited maximum resistance, users may not be able to cover the testing range close to their limit; there is also a lack of a unified way to combine measurable data such as movement speed, peak power, strength decay rate, and training completion rate. Furthermore, strength data generated by different movements and different equipment lacks consistent percentile representations and data confidence indicators.

[0005] 4. Lack of a unified interface between exercise indicators and downstream AI-based sports and health assessments: Existing systems may directly provide raw field data from equipment, manually input data, or single test results to subsequent assessment modules, lacking a unified description of exercise type, equipment source, reason for missing data, indicator generation time, calibration version, and data reliability, which affects the compatibility, traceability, and assessment stability of downstream AI models.

[0006] In summary, there is an urgent need for a method and system that can generate aerobic and anaerobic exercise indicators from multi-source intelligent sports equipment data, convert the indicators into standardized percentiles, and output them to an artificial intelligence sports and health assessment module according to a unified data structure. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for generating exercise indicators for AI-based sports health assessment. This method converts real-time or historical exercise data generated by different smart sports devices into standardized aerobic exercise indicators, anaerobic exercise indicators and their percentiles, and forms standardized exercise ability feature data with data confidence, data source and missing data markers, for use by downstream AI-based sports health assessment modules.

[0008] To achieve the above objectives, this technical solution provides a method for generating exercise indicators for AI-based sports and health assessment, comprising the following steps: S1: Obtain basic user information and collect multi-source motion data from at least one smart sports device; S2: Perform time synchronization, unit standardization, and abnormal data processing on the multi-source motion data, calculate the initial data quality value, and identify the motion type based on at least one of the following: device type and device identifier, as well as at least one of the following: motion type, data field combination, data continuity, and motion cycle characteristics. S3: When the exercise type includes aerobic exercise, calculate the average heart rate data within the first sliding window to obtain the first sliding window average; calculate the average power data or equivalent power data within the second sliding window to obtain the second sliding window average; calculate the power standard deviation of the original power data or equivalent power data within the third sliding window; determine the effective steady-state interval when the power standard deviation is less than a preset fluctuation threshold and the average value of the first sliding window is higher than a preset heart rate threshold; fit the heart rate time response relationship with the natural logarithm of the relative elapsed time within the effective steady-state interval as the independent variable and the corresponding first sliding window average value as the dependent variable to obtain the corrected heart rate corresponding to the preset target duration; establish an individualized heart rate-power saturation response relationship based on the corrected heart rate and the power representative value of the effective steady-state interval, reversely solve the target power corresponding to the target heart rate, and generate the maximum oxygen uptake as an aerobic exercise index based on the target power and the user's weight; update and obtain the aerobic data confidence level based on the duration of the effective steady-state interval, power stability, and the fitting quality of the heart rate time response relationship, combined with the initial data quality value. S4: When the exercise type includes anaerobic exercise, perform at least one of the following processes: generate a single maximum load based on a combination of training resistance and repetition count or different combinations of training resistance and movement speed; generate a representative value of strength ability based on at least two strength performance characteristics; use at least one of the obtained single maximum load and representative value of strength ability as an anaerobic exercise index; update and obtain the anaerobic data confidence level based on the number of effective repetitions, load coverage, movement integrity and the calculation quality of the estimated path used, combined with the initial data quality value. S5: Call the percentile standard library to convert the aerobic exercise index into aerobic capacity percentiles and the anaerobic exercise index into anaerobic capacity percentiles. S6: Based on the aforementioned sports indicators, corresponding percentiles, and data confidence levels, generate standardized sports ability feature data according to a preset data structure; S7: Output the standardized athletic ability characteristic data to the artificial intelligence sports and health assessment module as input data for the artificial intelligence sports and health assessment module.

[0009] Compared with existing technologies, this technical solution has the following characteristics and beneficial effects: This solution reduces the impact of initial exercise phase and power fluctuation data on the generation of aerobic indicators by using steady-state power conditions and heart rate time response relationships; it adapts to strength equipment with different resistance ranges and sensing capabilities through two paths: single maximum load and representative strength ability values; it provides a comparable and standardized expression of ability indicators for different units and different sports through a percentile standard library; and it provides structurally unified and traceable input data for downstream artificial intelligence sports health assessment by adding equipment type, generation time, missing data markers, and data confidence levels to the output data. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a method for generating exercise indicators for AI-based sports and health assessment according to an embodiment of this application; Figure 2 This is a schematic diagram of the standardized athletic ability characteristic data structure; Figure 3 This is a schematic diagram of the data interface between the sports index generation system and the downstream AI-powered sports and health assessment module. Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0011] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0012] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0013] Example 1 like Figure 1As shown, this solution provides a method for generating exercise indicators for AI-based sports and health assessment. It should be noted that "AI" in this specification is short for Artificial Intelligence. The core technology of this invention is the generation of exercise indicators and standardized exercise ability characteristic data; the specific model structure, training methods, and scoring algorithms of the downstream AI-based sports and health assessment module do not constitute a necessary limitation of this invention.

[0014] S1: Obtain basic user information and collect multi-source motion data from at least one smart sports device; S2: Perform time synchronization, unit standardization, and abnormal data processing on the multi-source motion data, calculate the initial data quality value, and identify the motion type based on the device type and data characteristics; S3: When the exercise type includes aerobic exercise, calculate the average heart rate data within the first sliding window to obtain the first sliding window average; calculate the average power data or equivalent power data within the second sliding window to obtain the second sliding window average; calculate the power standard deviation of the original power data or equivalent power data within the third sliding window; determine the effective steady-state interval when the power standard deviation is less than a preset fluctuation threshold and the average value of the first sliding window is higher than a preset heart rate threshold; fit the heart rate time response relationship with the natural logarithm of the relative elapsed time within the effective steady-state interval as the independent variable and the corresponding first sliding window average value as the dependent variable to obtain the corrected heart rate corresponding to the preset target duration; establish an individualized heart rate-power saturation response relationship based on the corrected heart rate and the power representative value of the effective steady-state interval, reversely solve the target power corresponding to the target heart rate, and generate the maximum oxygen uptake as an aerobic exercise index based on the target power and the user's weight; update and obtain the aerobic data confidence level based on the duration of the effective steady-state interval, power stability, and the fitting quality of the heart rate time response relationship, combined with the initial data quality value. S4: When the exercise type includes anaerobic exercise, perform at least one of the following processes: generate a single maximum load based on a combination of training resistance and repetition count or different combinations of training resistance and movement speed; generate a representative value of strength ability based on at least two strength performance characteristics; use at least one of the obtained single maximum load and representative value of strength ability as an anaerobic exercise index; update and obtain the anaerobic data confidence level based on the number of effective repetitions, load coverage, movement integrity and the calculation quality of the estimated path used, combined with the initial data quality value. S5: Call the percentile standard library to convert the aerobic exercise index into aerobic capacity percentiles and the anaerobic exercise index into anaerobic capacity percentiles. S6: Based on the aforementioned sports indicators, corresponding percentiles, and data confidence levels, generate standardized sports ability feature data according to a preset data structure; S7: Output the standardized athletic ability characteristic data to the artificial intelligence sports and health assessment module as input data for the artificial intelligence sports and health assessment module.

[0015] Regarding step S1 of this plan: Since the subsequent heart rate-power saturation response model needs to be built based on the maximum heart rate and resting heart rate, the user's basic information must include at least the maximum heart rate and resting heart rate. The maximum heart rate can be obtained from user input, historical tests, age formula initialization, existing maximum heart rate estimation models, or imported from third-party health platforms. The resting heart rate can be obtained from user input, imported from wearable devices, or calculated from historical resting state data.

[0016] Furthermore, basic user information can also include age, gender, weight, height, training experience, training goals, available training time, health restrictions, and family member identification. Among these, age, gender, and weight are used to match percentile standard libraries and correct exercise indicators; training goals, available training time, and health restrictions can serve as inputs for downstream AI-powered sports and health assessments or training recommendations; and family member identification can be used for data differentiation and independent record-keeping when multiple people share the device.

[0017] In real-world scenarios, users may perform different types of training on multiple smart fitness devices. This solution can generate aerobic and anaerobic exercise metrics from real-time and / or historical exercise data from at least one smart fitness device, and output them according to a unified data structure.

[0018] In some embodiments, smart exercise equipment includes one or any combination of treadmills, exercise bikes, rowing machines, elliptical trainers, electronic strength training devices, electronic resistance training devices, and wearable heart rate devices. For aerobic equipment that cannot directly output power, the equipment speed, incline, user weight, and the measured power corresponding to a reference power meter can be collected. An equivalent power mapping relationship can be established through least squares regression, piecewise interpolation, or a device calibration table, and the mapping relationship and its version can be recorded as calibration information.

[0019] Real-time or historical exercise data from various smart fitness devices can be collected via Bluetooth, proprietary protocols, Wi-Fi, HealthKit, Google Health Connect, Garmin Connect, or local sensors. The system can simultaneously record device type, device identifier, data source, and calibration version for data confidence calculation and traceability.

[0020] Regarding step S2 of this plan: Since the calculation logic and required data for ability indicators differ for different types of exercise, the system first identifies aerobic exercise, anaerobic exercise, or mixed exercise based on equipment attribute information (equipment type, equipment identifier) ​​and exercise data characteristics (action type, data field combination, data continuity, action cycle characteristics), and then executes the aerobic exercise indicator generation process and the anaerobic exercise indicator generation process respectively.

[0021] In some embodiments, the information used to identify the type of exercise includes device attribute information and exercise data features. The device attribute information includes at least one of device type and device identifier; the exercise data features include at least one of motion type, data field combination, data continuity, and motion cycle features. The presence of heart rate, speed, or power fields alone does not necessarily determine the type of exercise; the system can make a comprehensive judgment by combining the device attribute information and the exercise data features.

[0022] When the data is presented as continuous time-series heart rate and power, equivalent power, or speed data, it can be identified as aerobic exercise; when the data includes discrete movement cycles and at least one of resistance, repetition count, movement displacement, movement speed, or movement power, it can be identified as anaerobic exercise. If both aerobic and anaerobic exercise are identified simultaneously, steps S3 and S4 are executed respectively.

[0023] Before performing step S3 or step S4, the system preprocesses the multi-source motion data and calculates the initial data quality value based on the preprocessing results.

[0024] Specifically, the preprocessing operations are as follows: performing timestamp alignment, sampling frequency unification, data unit standardization, abnormal heart rate value removal, abnormal power or resistance filtering, speed and mileage continuity verification, exercise interruption status determination, missing data value marking, and data source reliability rating on multi-source motion data.

[0025] The initial data quality value can be calculated using the following formula: InitialQuality = u1 × C_complete + u2 × C_continuity + u3 × C_source + u4 × C_calibration. Where u1 to u4 are non-negative weights with a sum of 1, and each sub-item is normalized to 0 to 1; C_complete represents the completeness rate of key fields, C_continuity represents the percentage of effective motion time, C_source represents the reliability of the data source, and C_calibration represents the effectiveness of equipment calibration.

[0026] In steps S3 and S4, the system updates and obtains the confidence scores for aerobic and anaerobic data based on the initial data quality values, respectively. For example, the confidence score for aerobic data can be calculated as AerobicConfidence = v0 × InitialQuality + v1 × C_aerobic_stability + v2 × C_aerobic_fit, where C_aerobic_stability is determined based on the duration of the effective steady-state interval and the power standard deviation, and C_aerobic_fit is determined based on the fitting quality of the heart rate-time response relationship. The confidence level for anaerobic data can be calculated as AnaerobicConfidence = z0 × InitialQuality + z1 × C_motion_validity + z2 × C_load_coverage + z3 × C_path_quality, where C_motion_validity is determined based on the number of effective repetitions and the completeness of the movements, C_load_coverage represents the effective load coverage, and C_path_quality represents the computational quality of the repetition estimation path, load-velocity estimation path, or strength capacity representative value calculation path used. v0 to v2 and z0 to z3 are all non-negative weights, and the sum of the weights for each group is 1. Each sub-item is normalized to 0 to 1.

[0027] In some embodiments, when key heart rate data or power / equivalent power data is missing, corresponding aerobic exercise metrics are not generated; when data required to establish the combination of training resistance and repetitions or the load-speed relationship is missing, corresponding 1RM metrics are not generated. Data that is not tested, not supported by the device, or of insufficient quality is recorded as missing data and is not interpreted as a true low ability.

[0028] Regarding step S3 of this plan: Step S3 is used to generate aerobic exercise indicators. In one embodiment, maximum oxygen uptake is used to characterize the user's aerobic capacity and can serve as a unified capacity indicator across different aerobic exercise devices. Since the data fields output by different devices may differ, this solution allows the use of power data directly output by the device or equivalent power data generated based on speed, incline, weight, and device calibration parameters.

[0029] When identified as aerobic exercise, the system generates aerobic exercise indicators based on continuously collected heart rate data, power data or equivalent power data, and exercise duration data; speed and distance data can be used to assist in verifying the continuity of exercise, the status of equipment data transmission, and the validity of the test.

[0030] In this embodiment, the first sliding window is 5 seconds, the second sliding window is 60 seconds, and the third sliding window is 120 seconds. The system calculates the average heart rate data within the first sliding window to obtain the first sliding window average value HR5_t; calculates the average power data or equivalent power data within the second sliding window to obtain the second sliding window average value P60_t; calculates the window average value P120_t of the original power data or equivalent power data within the third sliding window, and calculates the power standard deviation within the third sliding window based on the window average value P120_t.

[0031] The continuous motion data used to perform steady-state candidate point determination should cover at least one complete third sliding window. Using longer periods of continuous motion data can generate multiple candidate sliding windows and help improve the stability of effective steady-state interval identification and heart rate time-response relationship fitting.

[0032] In the "Calculate the first sliding window average" step, all valid heart rate data within the current time and the previous 4 seconds are taken, and their arithmetic mean is calculated to obtain the first sliding window average HR5_t corresponding to the current time. If there are fewer than 5 valid heart rate data points before the current time, the calculation is performed according to the actual number of valid data points. The formula for calculating the first sliding window average is as follows: HR5_t = (1 / n) × ΣHR_j, where j = max(1, t-4)…t, n≤5; Where t is the current sampling time, HR_j is the raw heart rate data collected at the j-th sampling point, HR5_t is the average value of the first sliding window corresponding to time t, and n is the number of valid sampling points involved in the calculation. With a sampling frequency of 1Hz, the average value of the first sliding window is also called the 5-second rolling average heart rate.

[0033] In the step of "calculating the second sliding window average value", all effective power data or equivalent power data within the current time and the previous 59 seconds are taken, and their arithmetic mean is calculated to obtain the second sliding window average value P60_t corresponding to the current time. When there are fewer than 60 sampling points of effective power data or equivalent power data before the current time, the calculation is performed according to the actual number of effective sampling points. The formula for calculating the second sliding window average value is as follows: P60_t=(1 / m)×ΣP_j,where j=max(1,t-59)…t,m≤60; Where t is the current sampling time, P_j is the raw power data or equivalent power data collected at the j-th sampling point, P60_t is the average value of the second sliding window corresponding to time t, and m is the number of valid sampling points involved in the calculation. When the sampling frequency is 1Hz, the average value of the second sliding window is also called the 60-second rolling average power.

[0034] In the step of "calculating the power standard deviation within the third sliding window", the raw power data or equivalent power data within the current time and the previous 119 seconds are used to form the third sliding window, and the window average value P120_t within the third sliding window is calculated first: P120_t=(1 / q)×ΣP_j,where j=t-119…t; Where t is the current sampling time, P_j is the raw power data or equivalent power data collected at the j-th sampling point, P120_t is the average value of the third sliding window corresponding to time t, and q is the number of effective sampling points participating in the calculation within the third sliding window. When the sampling frequency is 1Hz, q = 120 is preferred.

[0035] When the number of valid sampling points in the third sliding window does not reach the preset number, the current sampling time is not determined as a steady-state candidate point, and the insufficient data can be included in the initial data quality value or aerobic data confidence level; when the number of valid sampling points in the third sliding window reaches the preset number, the power standard deviation of the original power data or equivalent power data in the third sliding window is calculated based on the window average value P120_t.

[0036] The first sliding window average value is used to reduce instantaneous heart rate noise; the second sliding window average value is used to represent the current power level and can serve as a representative power value for the effective steady-state interval or as the basis for calculating the representative power value; the power standard deviation of the original power data or equivalent power data within the third sliding window is used to determine whether the power output is stable.

[0037] In an implementation where the second sliding window average sequence is sampled or resampled at 1Hz and FTP is greater than 0, the system can further calculate the normalized power and training stress score.

[0038] Optionally, after obtaining the average value of the second sliding window, the system can also calculate the normalized power (NP) and training stress score (TSS) based on the sequence of the average values ​​of the second sliding window. The specific calculation formula is as follows: NP_t = [(1 / t) × Σ(P60_r^4)]^(1 / 4), where the value of r ranges from 1 to t; TSS_t=t×NP_t² / (FTP²×3600)×100; Wherein, NP_t is the standardized power at time t, used to characterize the current training intensity; TSS_t is the training stress score at time t, used to characterize the cumulative training load; FTP is the functional threshold power, and FTP² is the square of the functional threshold power; t is the effective exercise duration from the start of this exercise to the current time, in seconds.

[0039] The standardized power (NP) and training stress score (TSS) are not necessary conditions for determining steady-state candidate points, effective steady-state intervals, or generating maximum oxygen uptake. They can be used as data quality references for aerobic exercise indicators or as additional inputs to downstream sports health assessment and training recommendation modules.

[0040] The formula for calculating the power standard deviation of the original power data or equivalent power data within the third sliding window is as follows: SD120_t=sqrt[(1 / q)×Σ(P_j-P120_t)^2] Where SD120_t is the standard deviation of the original power data or equivalent power data in the third sliding window corresponding to time t, sqrt() is the square root function, P_j is the original power or equivalent power data of the j-th sampling point, and P120_t is the average power of the window.

[0041] In some embodiments, the preset fluctuation threshold can be set to 10W, and the preset heart rate ratio can be set to 0.85; the above parameters can be adjusted according to the device type, sampling frequency, user group, or calibration results.

[0042] This approach filters out periods of relatively stable power output by using the power standard deviation within the third sliding window and excludes low-intensity warm-up intervals by using a heart rate threshold, thereby reducing the impact of power fluctuations and low-intensity data on model parameter calculations.

[0043] In the step of "determining an effective steady-state interval when the power standard deviation is less than a preset fluctuation threshold and the average value of the first sliding window is higher than a preset heart rate threshold", sampling points that simultaneously satisfy the conditions of a power standard deviation less than a preset fluctuation threshold and an average value of the first sliding window higher than a preset heart rate threshold are determined as steady-state candidate points. When a preset number of consecutive steady-state candidate points are reached or a preset duration is reached, the effective steady-state interval is determined.

[0044] Once the current time period is confirmed to be within the effective steady-state range, the system constructs a logarithmic heart rate response curve to obtain the corrected heart rate corresponding to the preset duration. Then, it substitutes the representative power value of the effective steady-state range and the corrected heart rate into the heart rate-power saturation response model to solve for the user's corresponding power-heart rate response coefficient. Finally, it calculates the target steady-state power and maximum oxygen uptake corresponding to the target heart rate. This process reduces the impact of a sustained increase in heart rate and power instability at the beginning of exercise on the estimation results.

[0045] A continuous sampling interval that satisfies the steady-state condition is designated as the effective steady-state interval, and the starting point of the effective steady-state interval is taken as the relative time starting point. For example, for time t, a candidate sliding window is constructed using 120 seconds of continuous power data from t-119 to t. When the standard deviation of the original power data within this window is less than a preset fluctuation threshold, and the corresponding 5-second rolling average heart rate is higher than a preset proportion of the maximum heart rate, time t is marked as a steady-state candidate point. When the number of consecutive steady-state candidate points is not less than a preset number or the duration is not less than a preset duration, the corresponding time period is determined as the effective steady-state interval; the preset duration can be 10 to 60 seconds.

[0046] In one implementation, the number of preset steady-state candidate points, the duration threshold, and the fitting quality threshold are determined based on the device type, sampling frequency, and calibration data. When the effective steady-state interval is insufficient or the fitting quality does not meet the conditions, the system does not generate the maximum oxygen uptake, or outputs a low-confidence result and prompts for retesting.

[0047] The relative elapsed time of each sampling point within the effective steady-state interval relative to the starting point of the effective steady-state interval is transformed by natural logarithm. The natural logarithm of the relative elapsed time is used as the independent variable, and the average value of the first sliding window of the corresponding sampling point is used as the dependent variable to fit the heart rate time response relationship.

[0048] In one implementation, the least squares method is used to perform linear fitting on the independent and dependent variables. The number of data points involved in the fitting should be no less than a preset number, and the coefficient of determination R² of the fitting result should not be lower than a preset threshold, or the mean square error of the fitting result should not be higher than a preset threshold. When the number of data points involved in the fitting or the fitting quality does not meet the preset conditions, a corrected heart rate is not generated, or the corresponding aerobic exercise index is marked as low confidence.

[0049] For the j-th sampling point within the effective steady-state interval, the independent variable is defined as: x_j = ln(τ_j); Wherein, τ_j is the relative elapsed time of the j-th sampling point relative to the starting point of the effective steady-state interval. τ_j starts taking values ​​from a preset initial time greater than 0; in one embodiment with a sampling frequency of 1Hz, τ_j starts taking values ​​from 1 second to avoid the natural logarithm being undefined.

[0050] The dependent variable is defined as: y_j = HR5_j; Wherein, HR5_j is the average value of the first sliding window corresponding to the j-th sampling point within the effective steady-state interval; in the embodiment where the first sliding window is 5 seconds and the sampling frequency is 1Hz, HR5_j is the 5-second rolling average heart rate corresponding to the j-th sampling point.

[0051] The heart rate time response relationship is expressed as follows: y = slope × x + intercept; Where slope represents the rate of change of the average value of the first sliding window with the natural logarithm of the relative elapsed time, and intercept represents the fitting intercept.

[0052] The slope can be calculated using the following formula: slope=[N×Σ(x_j×y_j)-Σx_j×Σy_j] / [N×Σ(x_j²)-(Σx_j)²]; Where N is the number of data points involved in the fitting, and the calculation should satisfy that the denominator is not 0.

[0053] The fitting intercept can be calculated according to the following formula: intercept=(Σy_j-slope×Σx_j) / N.

[0054] Substituting the preset target duration T_adj into the heart rate time response relationship, we obtain the corrected heart rate AdjHR corresponding to the preset target duration: AdjHR=slope×ln(T_adj)+intercept.

[0055] Wherein, T_adj is a preset target duration greater than 0. In a preferred embodiment, T_adj is 300 seconds, and the corrected heart rate AdjHR is used to characterize the theoretical heart rate of the user when exercising continuously for 300 seconds at the power level corresponding to the effective steady-state interval.

[0056] When the corrected heart rate does not meet the conditions that the resting heart rate is less than the corrected heart rate and the corrected heart rate is less than the maximum heart rate, or when the fitting result does not meet the preset validity conditions, the corrected heart rate is determined to be invalid and will not participate in the subsequent parameter calculation of the individualized heart rate-power saturation response relationship.

[0057] This approach uses relative elapsed time within the effective steady-state interval, rather than an absolute time index calculated from the start of the entire exercise, ensuring a consistent time reference for heart rate time responses across different effective steady-state intervals. The corrected heart rate is used to characterize the theoretical heart rate when exercising continuously to the preset target duration at the corresponding stable power level, and to reduce the impact of an unstable heart rate during the initial stage of exercise on the estimation of maximum oxygen uptake.

[0058] Heart rate reserve is calculated based on the user's maximum heart rate and resting heart rate, and the power-heart rate response coefficient is calculated based on heart rate reserve, resting heart rate, power representative value of effective steady-state interval, and corrected heart rate.

[0059] Specifically, when constructing the heart rate-power saturation response model, data validity check rules are set: maxHR shall be greater than restingHR, the adjusted heart rate AdjHR shall satisfy restingHR<AdjHR<maxHR, the general model constant B shall be greater than 0, the representative power value P_s is not equal to the fixed power offset O, R_adj shall be greater than 0, and the denominator for regression calculation is not 0; a data point that fails to meet any of the conditions is determined to be invalid and does not participate in solving for the response coefficient k.

[0060] In some embodiments, the individualized heart rate-power saturation response model records power and its corresponding predicted heart rate value.

[0061] In some embodiments, the individualized heart rate-power saturation response relationship is HR(P)=restingHR+HRR / [1+B×exp(-k×(P-O))], where HRR=maxHR-restingHR; the power-heart rate response coefficient k is calculated according to R_adj=HRR / (AdjHR-restingHR)-1 and k=-ln(R_adj / B) / (P_s-O); The target heart rate TargetHR=α×maxHR is set, and the target power corresponding to a single valid steady-state sample is calculated according to R_target=HRR / (TargetHR-restingHR)-1 and P_target=-ln(R_target / B) / k+O; The maximum oxygen uptake power PVO2max is determined based on one or more P_targets that have passed the validity check, and the maximum oxygen uptake is calculated according to VO2max=C0+C1×PVO2max / weight; Wherein, P_s is the representative power value corresponding to the valid steady-state interval, AdjHR is the adjusted heart rate, weight is the user's body weight, maxHR is the user's maximum heart rate, restingHR is the user's resting heart rate, B, O, α, C0 and C1 are preset or calibrated parameters, and maxHR>restingHR, restingHR<AdjHR<maxHR, B>0, P_s≠O, R_adj>0, R_target>0, k is a finite positive value, and P_target>0.

[0062] For example, the calculation formula for heart rate reserve is: HRR=maxHR-restingHR; where HRR is the heart rate reserve, maxHR is the user's maximum heart rate, and restingHR is the user's resting heart rate.

[0063] Set a fixed power offset O=hrOffset to calibrate the power reference; set a universal model constant B=hrFormulaConstantB to control the shape of the heart rate-power response curve.

[0064] The constructed heart rate-power saturation response model is as follows: HR(P)=restingHR+HRR / (1+B×exp(-k×(PO))).

[0065] Where HR(P) is the predicted heart rate value corresponding to power P, k is the power-heart rate response coefficient, and exp() is the natural exponential function.

[0066] The formula for calculating the heart rate reserve adjustment parameter is: R_adj=HRR / (AdjHR-restingHR)-1.

[0067] Where R_adj is the dimensionless response parameter corresponding to the corrected heart rate, and AdjHR is the corrected heart rate.

[0068] The formula for calculating the power-heart rate response coefficient is: k = -ln(R_adj / B) / (P_s-O).

[0069] Wherein, P_s is the power representative value of the effective steady-state interval, which can be the 60-second rolling average power of the corresponding interval, the interval average power, or the power obtained by steady-state mass weighting.

[0070] Set the target heart rate to a preset proportion of the maximum heart rate, and then calculate the target steady-state power corresponding to that target heart rate.

[0071] For example: TargetHR = α × maxHR, where α can be from 0.90 to 0.98, preferably 0.95; R_target=HRR / (TargetHR-restingHR)-1; Where TargetHR is the target heart rate, and R_target is the dimensionless response parameter corresponding to the target heart rate, not the power value.

[0072] The target steady-state power is calculated as follows: P_target = -ln(R_target / B) / k + O, where P_target is the target steady-state power corresponding to a single effective steady-state sample; B is a universal model constant; k is the power-heart rate response coefficient; and O is a fixed power offset.

[0073] For one or more effective steady-state intervals, calculate P_target. The maximum effective value, median, upper quantile value, maximum value after removing outliers, or representative value obtained by weighting according to steady-state quality can be used as the maximum oxygen uptake power PVO2max.

[0074] In a preferred embodiment: PVO2max = max(P_target over valid samples), where valid samples are steady-state samples that have passed data validity checks; the calculation should also satisfy restingHR. <TargetHR<maxHR、R_target> 0 and k are finite positive values ​​and P_target>0, and abnormal target power can be eliminated according to the equipment calibration range.

[0075] In the above validity check, if k is not a finite positive value, the target power exceeds the calibration range of the device or algorithm, the number of fitted data points is insufficient, or the regression denominator is 0, then the corresponding steady-state sample is excluded.

[0076] Maximum oxygen uptake is converted as follows: VO2max = C0 + C1 × (PVO2max / weight), where the unit of VO2max is ml / kg / min, and weight is kilogram; C0 is the basic oxygen uptake parameter, and C1 is a comprehensive parameter that includes power-oxygen uptake conversion, unit conversion and equipment calibration factors.

[0077] C0 and C1 can be calibrated for different equipment types by matching equipment power data with laboratory gas metabolism test results. If the equipment or input uses pounds as the unit of weight, it should be converted to kilograms before being substituted into the formula to avoid repeated use of weight unit conversion and equipment correction factors.

[0078] In one optional implementation, the baseline oxygen uptake parameter and the power-oxygen uptake combined parameter can be set separately according to the equipment type, and the maximum oxygen uptake is calculated according to the following formula: VO2max=C0_d+C1_d×(PVO2max / weight) Wherein, d represents the type of exercise equipment, and C0_d and C1_d are the baseline oxygen uptake parameter and power-oxygen uptake combined parameter corresponding to the type of exercise equipment, respectively. C0_d and C1_d can be determined for specific equipment categories or models by regression, piecewise fitting, or calibration of paired samples of laboratory gas metabolism test results with the power or equivalent power data output by the equipment.

[0079] In another alternative implementation, the comprehensive parameter C1 can be decomposed into the product of the base power-oxygen uptake conversion parameter C1_base and the device type calibration parameter C_device, with the maximum oxygen uptake calculated according to the following formula: VO2max=C0+C1_base×(PVO2max / weight)×C_device C_device is a device type calibration parameter greater than 0, which can be determined based on the device category, device model, and calibration data. When the power data or equivalent power data used already includes the device type calibration factor, C_device is set to 1 or is no longer used separately to avoid repeated calibration of the device type factor.

[0080] In some embodiments, when devices such as treadmills and elliptical trainers cannot directly output power, speed, incline, user weight, and the measured power corresponding to a reference power meter can be collected. An equivalent power mapping relationship can be established through least squares regression, piecewise interpolation, or a device calibration table. The method for generating the equivalent power, the reference data range, and the calibration version are recorded together with the aerobic exercise indicators; when the input exceeds the calibration range, the data confidence level can be reduced or the aerobic exercise indicators can be not generated.

[0081] In some embodiments, step S3 further acquires speed and mileage to assist in verifying the confidence level of the VO2 max data. Speed ​​and mileage are not the primary variables determining VO2 max, but they can be used to verify exercise continuity, equipment transmission status, and test validity. For example, if power is consistently greater than a preset threshold but speed is 0, or if mileage does not increase over time, the system can determine that the equipment data is abnormal and reduce the confidence level of the aerobic data.

[0082] Regarding step S4 of this plan: When identified as anaerobic exercise, the system acquires data from electronic strength testing equipment, including resistance, number of repetitions, number of training sets, movement speed, strength curve, peak power, average power, and range of motion, to generate anaerobic exercise indicators. These anaerobic exercise indicators may include a single-repetition maximum load (1RM) and / or a representative value for strength capacity.

[0083] Furthermore, when the user has not performed a true maximum weight test, the maximum load for a single repetition can be estimated based on the combination of training resistance and repetition count; when the device can collect movement speed under different training resistances, the maximum load for a single repetition can be estimated based on the load-speed relationship; when the device's maximum resistance is low or the effective load coverage is insufficient, a representative value of strength ability can be calculated based on at least two strength performance characteristics.

[0084] In some embodiments, the maximum single load is calculated in at least one of the following ways: based on the training resistance W, the number of repetitions R completed, and an empirical coefficient K greater than 0, according to 1RM=W×(1+R / K); or, by collecting the centripetal phase movement speed under different training resistances, fitting the load-speed relationship V=ab×W, where a and b are model fitting parameters, and under the condition that b>0 and a>Vmin, calculating according to the preset minimum effective movement speed Vmin according to 1RMv=(a-Vmin) / b.

[0085] Specifically, if multi-source motion data shows that the current user has not performed a true maximum weight test, the maximum load per repetition is estimated based on the number of repetitions completed under training resistance using an empirical conversion formula, as follows: 1RM = W × (1 + R / K), where W is the training resistance or equivalent load, and W is converted to a preset unit such as kilogram or pound before calculation; R is the number of repetitions (unit: times); K is an empirical coefficient greater than 0, with a suggested value range of 25-40, and a commonly recommended value of 30.48.

[0086] The aforementioned empirical conversion formula is a submaximal load estimation method. It allows users to generate a 1RM estimate based on daily training resistance and repetitions without directly challenging their maximum load. For beginners or users with exercise limitations, the system can decide whether to output this indicator based on data confidence and health restrictions.

[0087] In addition, if the smart sports equipment has the ability to collect motion speed, power, or force curves, then the maximum load per repetition can be estimated based on the load-speed or power conversion formula, combined with the average speed, peak speed, average power, and peak power during the concentric phase under different training resistances. The load-speed or power conversion formula is as follows: Vj = displacementj / concentric_timej, where Vj is the average velocity during the centripetal phase of the j-th training group, in meters per second; displacementj is the displacement during the centripetal phase of the j-th training group, in meters; and concentric_timej is the time during the centripetal phase of the j-th training group, in seconds. Fitting model: V = a - b W, where V is the average velocity during the centripetal phase, in meters per second; W is the training resistance, in kilograms; and a and b are the model fitting parameters. The estimated value is 1RMv = (a - Vmin) / b, where 1RMv is the estimated maximum load per action in kilograms; Vmin is the preset minimum effective action speed in meters per second, which can be determined by the action type, equipment configuration, and user level. The calculation should satisfy b > 0 and a > Vmin.

[0088] When both empirical formula estimates and load-velocity estimates are available, the confidence level can be determined based on load coverage, fit quality, action integrity, and data source reliability, and the two estimates can be weighted and fused. If the confidence level of any estimation path is lower than the threshold, the path is not used or the result is marked as low confidence.

[0089] Let the empirical formula estimate be 1RM_rep and the load-velocity estimate be 1RM_vel, with corresponding confidence levels of C_rep and C_vel. Then, the fusion can be performed as 1RM=(C_rep×1RM_rep+C_vel×1RM_vel) / (C_rep+C_vel), where C_rep+C_vel>0.

[0090] In some embodiments, when the maximum resistance of the smart sports device is lower than a preset resistance threshold, the representative value of the strength ability is calculated based on at least two of the following: maximum force per unit body weight, peak power per unit body weight, movement speed stability, training completion rate, and fatigue decay rate. The fatigue decay rate and training completion rate are limited to 0 to 1, and each indicator is normalized using N(x) = min(1, max(0, (xL) / (UL))), where L and U are the lower and upper reference limits corresponding to movement type, gender, age group, device type, or user group, and U > L. The movement speed stability is calculated based on the coefficient of variation of movement speed and limited to 0 to 1. When any sub-indicator is missing, the weights of the remaining sub-indicators are re-normalized.

[0091] Specifically, when the maximum resistance of the smart exercise device is relatively low, it is not necessary to force the output of the maximum load in a single session. Instead, it calculates the fatigue decay value and training completion rate based on multi-source motion data, and combines these with other measurable indicators to form a representative value of strength ability. The specific calculation formula is as follows: FR = clip(1 - average power of the last few sets of movements / average power of the first few sets of movements, 0, 1) ER=clip(number of completed sets / target number of sets, 0, 1) Q_strength = a1 N (maximum strength / body weight) + a2 N(peak power / body weight) + a3 N (velocity stability) + a4 N (training completion rate) + a5 N(1-fatigue attenuation rate); Where a1 + a2 + a3 + a4 + a5 = 1 Wherein, FR is the fatigue decay rate, ER is the training completion rate, clip() is used to limit the value to 0 to 1, N() is the normalization function, and a1 to a5 are the weight coefficients of each indicator. When calculating FR, the average power of the first few sets of actions should be greater than 0. The normalization function can be N(x)=min(1,max(0,(xL) / (UL))), and U>L; L and U are the lower and upper reference limits corresponding to action type, gender, age group, device type, or user group. Action speed stability can be calculated using 1-CV(v) and limited to 0 to 1, where CV(v) is the coefficient of variation of effective repetitive action speed; when a certain sub-indicator is missing, the weights of the remaining sub-indicators are renormalized.

[0092] Regarding step S5 of this plan: This solution utilizes a pre-defined percentile standard library. This library stores population reference data corresponding to exercise indicators according to gender, age group, weight range, indicator type, indicator name, device type, and percentile threshold. It can be constructed from publicly available norms, enterprise-built databases, quality-controlled historical data of App users, laboratory calibration data, or mixed data.

[0093] In some embodiments, the percentile standard library stores reference data according to gender, age group, indicator type, indicator name, and percentile threshold; when an exercise indicator falls between two adjacent percentile thresholds, linear interpolation, nearest neighbor, or interval mapping is used to obtain the corresponding percentile; when there are multiple anaerobic exercise indicators, the anaerobic capacity percentiles corresponding to each exercise are weighted or unweighted and aggregated to obtain the comprehensive anaerobic capacity percentile, and the aerobic capacity percentile and the comprehensive anaerobic capacity percentile are retained respectively.

[0094] In some embodiments, aerobic exercise metrics can be subdivided according to equipment type or testing method, such as VO2 max for running and VO2 max for cycling; anaerobic exercise metrics can be subdivided according to movement name and metric definition, such as relative 1RM or strength capacity representative values ​​for squats, deadlifts, bench presses or chest presses, overhead presses, lat pulldowns, and bicep curls. After obtaining basic user information and aerobic and anaerobic exercise metrics, the system queries the corresponding aerobic and anaerobic capacity percentiles in the percentile standard library based on user grouping and metric definitions.

[0095] When the anaerobic exercise index is 1RM, the relative strength value 1RM / body weight can be calculated first, and then the corresponding percentile can be queried according to the exercise type and user group; when the anaerobic exercise index is a strength capacity representative value, a dedicated norm table consistent with the calculation definition of the strength capacity representative value is used for querying.

[0096] If the aerobic or anaerobic exercise index falls between two adjacent percentile thresholds in the percentile standard library, the corresponding percentile can be obtained by linear interpolation, nearest neighbor, or interval mapping.

[0097] When a user completes multiple anaerobic exercises, the anaerobic capacity percentiles corresponding to each exercise can be weighted equally, by exercise category, or by data confidence level to form the overall anaerobic capacity percentile (StrengthP). The aerobic capacity percentile (CardioP) and the overall anaerobic capacity percentile (StrengthP) are retained separately; the aerobic capacity percentile is not included in the overall strength percentile.

[0098] If certain movements cannot be completed by the current smart sports equipment, functionally equivalent movements can be used as substitutes; when a certain movement indicator is missing, the comprehensive anaerobic capacity percentile is generated only based on the existing movements, and the missing data is marked to record reasons such as not being tested, not being supported by the equipment, or insufficient data quality.

[0099] Step S6 of this scheme: Generate standardized athletic ability characteristic data.

[0100] The system combines aerobic exercise indicators, anaerobic exercise indicators, corresponding percentiles, data confidence levels, and metadata according to a preset data structure. The standardized exercise ability characteristic data includes at least the indicator identifier, indicator value, indicator unit, data source, indicator generation time, data confidence level, and missing data marker, and includes at least one of the following: equipment type, maximum oxygen uptake power, calibration version, and data version; the indicator values ​​may include maximum oxygen uptake, aerobic capacity percentile, 1RM for each movement, representative strength ability value, and anaerobic capacity percentile.

[0101] Specifically, missing data tags are used to distinguish between reasons for missing data, such as untested, unsupported by equipment, insufficient data quality, and unauthorized user access; the indicator generation time indicates the timeliness of the indicator; the calibration version identifies the calibration parameter version of the equipment or algorithm; and the data version identifies the standardized data structure version. By simultaneously retaining the original indicator value, percentile, and data confidence level, downstream modules can identify the meaning, source, and reliability of the indicator.

[0102] Regarding step S7 of this solution: Output standardized athletic ability characteristic data to the artificial intelligence sports and health assessment module.

[0103] Standardized athletic ability characteristic data can be output to the AI-powered sports and health assessment module via mobile application internal functions, device interfaces, application programming interfaces (APIs), database records, message queues, or network services. This AI-powered sports and health assessment module can be located on a mobile terminal, sports equipment, server, or cloud platform.

[0104] It is important to emphasize that this invention does not use the artificial intelligence model itself as a means of generating exercise indicators, nor does it limit the model structure, training method, weights, or scoring algorithms of downstream artificial intelligence sports and health assessment modules. The technical problem solved by this invention is to generate aerobic and anaerobic exercise indicators with unified meaning, unified format, percentile expression, and confidence level labels from multi-source sports equipment data, and to provide a data interface suitable for downstream artificial intelligence assessment.

[0105] The following content regarding comprehensive exercise status, health score, training suggestions, and equipment parameter adjustments are all optional downstream application examples of standardized exercise ability characteristic data and do not constitute necessary limitations for steps S1 to S7.

[0106] In one downstream application embodiment, the AI-powered sports and health assessment module can construct a sports performance status based on the CardioP percentile of aerobic capacity and the StrengthP percentile of comprehensive anaerobic capacity; in scenarios where only strength improvement is a focus, StrengthP can also be used alone.

[0107] In other downstream application embodiments, to more comprehensively assess a user's exercise status, the AI-powered sports and health assessment module can combine CardioP, StrengthP, and features such as health percentile, training load, recovery ability, training goals, and data confidence. Each dimension of the indicator can be first converted to percentiles or standardized values ​​before comprehensive evaluation, rather than directly adding the original indicators with different dimensions and units.

[0108] CardioP, the percentile of cardiorespiratory fitness, reflects an individual's percentile ranking of aerobic capacity within a corresponding population. It can be obtained directly from the maximum oxygen uptake and its percentile generated according to this invention, or further combined with information such as resting heart rate in downstream modules to form extended cardiorespiratory fitness characteristics.

[0109] In order to incorporate training objectives, the downstream AI sports and health assessment module can also receive features such as standardized training load, training objective priority, and StrengthP, among which the training objective-related features can be quantified to 0 to 100 or other uniform numerical ranges.

[0110] In some embodiments, training objective metrics include standardized training load or training objective priority, wherein standardized training load can integrate standardized power (NP), training stress score (TSS), training volume and intensity range, and convert them to a uniform numerical range; training objective priority is taken from the training objectives in the user's basic information and is used by downstream modules to adjust the focus of evaluation or suggestions.

[0111] The downstream AI-powered sports and health assessment module can also receive personal health indicators, recovery characteristics, and athletic ability characteristics such as StrengthP to generate sports and health assessment results.

[0112] In some embodiments, personal health indicators include standardized recovery capacity and health limitation markers, wherein standardized recovery capacity can be converted to a uniform numerical range by combining sleep quality, heart rate variability, subjective fatigue feedback and exercise performance data; and health limitation markers are used to indicate user-reported exercise limitations and can serve as constraint information for downstream training recommendations to help reduce the risk of inappropriate training recommendations.

[0113] Data confidence level reflects the completeness and reliability of exercise metrics and physiological data. When device data is missing, exercise segments are invalid, or calibration versions are mismatched, data confidence level decreases; downstream scoring modules can reduce the weight of corresponding features or prompt retesting based on data confidence level.

[0114] In some downstream application scenarios, the overall state can be quantized and represented using the following vector form: StateVector=[Aerobic Capacity Percentile (CardioP), Overall Anaerobic Capacity Percentile (StrengthP), Standardized Training Load (TrainingLoad), Standardized Recovery (RecoveryP), Training Consistency (ConsistencyP), Data Confidence (Confidence), Training Goal Priority (GoalPriority), Health Limitation Flag (LimitationFlag)].

[0115] Downstream modules can calculate the degree of deviation for percentile-type indicators, using the 50th percentile as the neutral level; for non-percentile indicators such as training load, recovery ability, data confidence, and health limit markers, corresponding standardization or constraint processing methods are adopted respectively.

[0116] For capability-related indicators such as CardioP and StrengthP, return weights and insufficient penalty weights can be set according to different situations of being above or below the neutral level; for risk-related indicators, weights that match the risk direction can be set.

[0117] In one downstream scoring embodiment, a combined state coupling term and a constraint term can be introduced, as exemplified by the following formula: Composite = Composite_base + λ1 min(D_cardio, D_strength) - λ2 OverloadRisk + λ3 RecoveryBalance OverloadRisk = max(0, TrainingLoad - RecoveryP) RecoveryBalance=max(0,RecoveryP-β1×(1-ConsistencyP)-β2×LimitationFlag); Wherein, Composite represents the score for downstream comprehensive athletic performance, and Composite_base is the base comprehensive score; λ1, λ2, λ3, β1, and β2 are calibrable weights; D_cardio and D_strength are derived from CardioP and StrengthP, respectively; OverloadRisk represents the degree to which the training load exceeds recovery capacity; RecoveryBalance is a recovery balance feature generated and constrained to non-negative values ​​based on recovery capacity, training persistence, and health limitation markers. This formula is only an example of downstream application of the athletic indicators of this invention and does not limit the specific scoring algorithm of the artificial intelligence athletic health assessment module.

[0118] The aforementioned coupling terms allow downstream modules to simultaneously consider aerobic capacity, anaerobic capacity, training load, and recovery status; when the training load is too high and the recovery capacity is insufficient, or when there are health limitations, the downstream modules can lower the recommended level of high-intensity training.

[0119] In other downstream application embodiments, the AI-powered sports and health assessment module can determine the level of exercise status and suggest directions based on the overall condition.

[0120] Specifically, the correspondence between the exercise status level and the recommended direction can be found in Table 1 below: Table 1 Exemplary Description .

[0122] Furthermore, the downstream AI-powered sports and health assessment module can generate personalized training feedback schemes based on the overall condition according to preset rules or models.

[0123] For example, when StrengthP is low and RecoveryP is also weak, downstream modules can recommend low-resistance technical training or restorative strength training; when CardioP is insufficient but TrainingLoad is already high, low-intensity zone 2 training or restorative aerobic training can be recommended; when the confidence level of the exercise index data is insufficient, a retest can be prompted first.

[0124] In another application, the downstream module can select the corresponding training mode from the training template library based on StrengthP, IRM, training objective, and training status.

[0125] The training template library may include templates such as 5-3-1 Standard, 5-5-5 Standard, muscle building, body shaping, fat loss, VO2max enhancement, endurance enhancement, and recovery training. Examples of selection rules are shown in Table 2.

[0126] Table 2: Rules for Selecting Training Modes .

[0128] During training, downstream modules can also dynamically adjust training parameters based on real-time heart rate response, power stability, speed or incline completion, movement speed, strength decay, RPE feedback, and training completion rate.

[0129] For example, in treadmill training, the speed or incline can be adjusted based on the difference between the actual heart rate and the target heart rate zone; in cycling or rowing training, the resistance, target rotation speed, or target power can be adjusted based on heart rate response and power stability; in strength training, the resistance, repetitions, or rest time for the next set can be adjusted based on the decrease in movement speed, the rate of strength decay, and the number of interruptions. These adjustments are performed by downstream modules and do not alter the exercise metrics and standardized feature data generated by this invention.

[0130] Optional applications of downstream AI-powered sports and health assessment may also include the following: The AI-powered sports and health assessment module can acquire users' body and sleep indicators, and combine them with the aerobic and anaerobic exercise indicators and their percentiles generated by this invention to generate body indicator percentiles, sleep percentiles, AI sports and health scores, health span scores, risk warnings, or recovery status assessments.

[0131] Physical and sleep indicators may include glycated hemoglobin (HbA1c) or average blood glucose from continuous glucose monitoring, total cholesterol, high-density lipoprotein cholesterol, non-high-density lipoprotein cholesterol, total sleep duration, deep sleep, rapid eye movement sleep, sleep score, resting heart rate, and heart rate variability.

[0132] These metrics can be entered manually by users or imported through wearable devices such as Apple Health, Google Health Connect, Garmin Connect, and Ora, or by medical examination institutions.

[0133] Body metrics and sleep metrics are not necessary inputs for generating aerobic and anaerobic exercise metrics in this invention; they can be used by downstream AI-powered sports and health assessment modules in conjunction with the standardized exercise ability characteristic data output by this invention for sports and health management, trend analysis, training tips, risk alerts, recovery assessments, or follow-up reminders. The downstream assessment results should not be used as disease diagnoses, treatment decisions, or medical conclusions.

[0134] When a downstream module lacks any metric, the cause of the missing data can be identified by marking the missing data in the standardized athletic ability characteristic data. The module can then choose not to use the metric, use a neutral value, or adopt other missing data handling methods. Downstream processing should not misinterpret untested or unsupported metric as a true low-level metric.

[0135] Example 2 Referring to Embodiment 1, this solution provides a sports indicator generation system for AI-based sports health assessment, including an information acquisition module, a preprocessing and type recognition module, an aerobic indicator generation module, an anaerobic indicator generation module, a percentile conversion module, a standardized feature generation module, and a feature output module.

[0136] The information acquisition module is used to obtain basic user information and collect multi-source motion data from at least one smart sports device.

[0137] The preprocessing and type identification module is used to perform time synchronization, unit standardization, abnormal data processing and missing data marking on the multi-source motion data, calculate the initial data quality value, and identify aerobic exercise, anaerobic exercise or mixed exercise based on at least one of the following: device type and device identifier, and at least one of the following: action type, data field combination, data continuity and action cycle characteristics.

[0138] The aerobic index generation module is used to identify effective steady-state intervals based on heart rate data and power data or equivalent power data, establish heart rate-time response relationships and individualized heart rate-power response relationships, generate maximum oxygen uptake and maximum oxygen uptake power, and update and obtain the confidence level of aerobic data based on the duration of the effective steady-state interval, power stability and the fitting quality of the heart rate-time response relationship, combined with the initial data quality value.

[0139] The anaerobic index generation module is used to perform at least one of the following processes: generating a single maximum load based on a combination of training resistance and repetition count or different combinations of training resistance and movement speed; generating a representative value of strength ability based on at least two strength performance characteristics; using at least one of the obtained single maximum load and representative value of strength ability as an anaerobic exercise index, and updating and obtaining the anaerobic data confidence level based on the number of effective repetitions, load coverage, movement integrity and the calculation quality of the estimation path used, combined with the initial data quality value.

[0140] The percentile conversion module is used to call the percentile standard library to generate aerobic capacity percentiles and anaerobic capacity percentiles respectively; when there are multiple anaerobic exercise indicators, a comprehensive anaerobic capacity percentile can be generated and retained separately from the aerobic capacity percentile.

[0141] The standardized feature generation module is used to combine indicator identifier, indicator value, indicator unit, data source, indicator generation time, data confidence level and missing data marker, and combine at least one of equipment type, calibration version and data version to generate standardized motion capability feature data.

[0142] The feature output module is used to output the standardized athletic ability feature data to the artificial intelligence sports and health assessment module through a local function interface, application programming interface, database, message queue, or network interface.

[0143] In some downstream application embodiments, the AI-powered sports and health assessment module can further output one or more of the following: current exercise status, main weaknesses, suggested training type, suggested training intensity, suggested training duration, target heart rate, target power, target resistance, exercise list, number of sets, number of repetitions, rest time, expected improvement direction, and retest prompts. The above assessment and training feedback are downstream applications of standardized exercise ability characteristic data and do not constitute necessary steps in the exercise indicator generation process of this invention.

[0144] Output methods may include app pages, voice prompts, training course recommendations, device control parameters, reports, or home health centers. Figure 2 The data structure of standardized athletic ability characteristic data is shown. Figure 3 The data interface between the exercise index generation system and the downstream AI-powered sports and health assessment module is shown.

[0145] In scenarios where multiple people share a device, the system can identify users through account selection, NFC, Bluetooth binding, facial or voiceprint recognition, weight characteristics, or training habits, and associate each user's multi-source exercise data with the corresponding user profile. Each user's exercise metrics, percentiles, and data confidence scores are calculated separately to avoid data mixing between different family members.

[0146] Example 3 This embodiment also provides an electronic device, for reference. Figure 4 It includes a memory 402 and a processor 401. The memory 402 stores a computer program, and the processor 401 is configured to run the computer program to execute the method for generating exercise indicators for AI-based sports and health assessment as described in any of the above embodiments.

[0147] The processor 401 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0148] The memory 402 may include a large-capacity memory for data or instructions, and is used to store or cache multi-source motion data, percentile standard library, calibration parameters, standardized motion capability characteristic data, and computer program instructions executed by the processor 401.

[0149] The processor 401 implements the motion index generation method in the above embodiments by reading and executing computer program instructions stored in the memory 402.

[0150] Optionally, the electronic device may further include a transmission device 403 and an input / output device 404, wherein the transmission device 403 is connected to the processor 401, and the input / output device 404 is connected to the processor 401.

[0151] The transmission device 403 can receive data from smart sports devices via wired networks, wireless networks, Bluetooth, Wi-Fi or other communication methods, and output standardized sports ability characteristic data to the artificial intelligence sports health assessment module.

[0152] Input / output device 404 is used to input or output information. In this embodiment, the input information may include basic user information and multi-source exercise data, and the output information may include maximum oxygen uptake, single maximum load, representative strength ability value, corresponding percentile, data confidence level, and standardized exercise ability characteristic data.

[0153] Optionally, processor 401 may be configured to perform the following steps via a computer program: S1: Obtain basic user information and collect multi-source motion data from at least one smart sports device; S2: Perform time synchronization, unit standardization, and abnormal data processing on the multi-source motion data, calculate the initial data quality value, and identify the motion type based on the device type of the intelligent motion device and the data characteristics of the multi-source motion data; S3: When the exercise type includes aerobic exercise, calculate the average value of heart rate data within the first sliding window, the average value of power data or equivalent power data within the second sliding window, and the standard deviation of power data or equivalent power data within the third sliding window; determine an effective steady-state interval when the standard deviation of power is less than a preset fluctuation threshold and the average value of the first sliding window is higher than a preset heart rate threshold; fit the heart rate-time response relationship with the natural logarithm of the relative elapsed time within the effective steady-state interval as the independent variable and the corresponding average heart rate as the dependent variable to obtain the corrected heart rate corresponding to the preset target duration; establish an individualized heart rate-power saturation response relationship based on the corrected heart rate and the power representative value of the effective steady-state interval, reversely solve the target power corresponding to the target heart rate, and generate the maximum oxygen uptake as an aerobic exercise index based on the target power and the user's weight; update and obtain the aerobic data confidence level based on the duration of the effective steady-state interval, power stability, and the fitting quality of the heart rate-time response relationship, combined with the initial data quality value. S4: When the exercise type includes anaerobic exercise, perform at least one of the following processes: calculate the maximum load per repetition based on a combination of training resistance and repetition count, or based on a combination of different training resistance and movement speed; calculate a representative value of strength ability based on at least two of the following: maximum strength per unit body weight, peak power per unit body weight, movement speed stability, training completion rate, and fatigue decay rate; use at least one of the obtained maximum load per repetition and representative value of strength ability as an anaerobic exercise indicator; update and obtain the anaerobic data confidence level based on the number of effective repetitions, load coverage, movement integrity, and the calculation quality of the estimated path used, combined with the initial data quality value. S5: Call the preset percentile standard library to convert the aerobic exercise index into aerobic capacity percentile and the anaerobic exercise index into anaerobic capacity percentile. S6: Based on aerobic exercise indicators, anaerobic exercise indicators, aerobic capacity percentiles, anaerobic capacity percentiles and corresponding data confidence levels, generate standardized exercise capacity feature data according to a preset data structure. S7: Output the standardized athletic ability characteristic data as input data to the artificial intelligence sports and health assessment module.

[0154] This embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described motion index generation method. The storage medium may include a semiconductor memory, a magnetic storage medium, or an optical storage medium.

[0155] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0156] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products), including software routines, applets, and / or macros, can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer-executable components configured to perform embodiments when the program is run. One or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted that any block in the logical flow of the figures may represent a program step, or interconnected logical circuitry, blocks and functions, or a combination of program steps and logical circuitry, blocks and functions. The software may be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0157] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0158] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for generating exercise indicators for AI-based sports and health assessment, characterized in that, Includes the following steps: S1: Obtain basic user information and collect multi-source motion data from at least one smart sports device; S2: Perform time synchronization, unit standardization, and abnormal data processing on the multi-source motion data, calculate the initial data quality value, and identify the motion type based on at least one of the device type and device identifier of the intelligent motion device, as well as at least one of the motion type, data field combination, data continuity, and motion cycle characteristics. S3: When the exercise type includes aerobic exercise, calculate the average value of heart rate data within the first sliding window to obtain the average value of the first sliding window; calculate the average value of power data or equivalent power data within the second sliding window to obtain the average value of the second sliding window; calculate the power standard deviation of the original power data or equivalent power data within the third sliding window; determine the effective steady-state interval when the power standard deviation is less than a preset fluctuation threshold and the average value of the first sliding window is higher than a preset heart rate threshold; Using the natural logarithm of the relative elapsed time within the effective steady-state interval as the independent variable and the corresponding first sliding window average as the dependent variable, a heart rate-time response relationship is fitted to obtain the corrected heart rate corresponding to the preset target duration. Based on the corrected heart rate and the power representative value of the effective steady-state interval, an individualized heart rate-power saturation response relationship is established. The target power corresponding to the target heart rate is calculated in reverse. Based on the target power and the user's weight, the maximum oxygen uptake is generated as an aerobic exercise indicator. The aerobic data confidence level is updated and obtained by combining the initial data quality value with the duration of the effective steady-state interval, power stability, and the fitting quality of the heart rate-time response relationship. S4: When the type of exercise includes anaerobic exercise, perform at least one of the following processes: calculate the maximum load for a single repetition based on a combination of training resistance and repetition count, or based on a combination of different training resistance and movement speed; A representative value of strength ability is calculated based on at least two of the following: maximum force per unit body weight, peak power per unit body weight, movement speed stability, training completion rate, and fatigue decay rate. At least one of the obtained single maximum load and strength capacity representative values ​​is used as the anaerobic exercise index; the anaerobic data confidence level is updated and obtained based on the number of effective repetitive movements, load coverage, movement integrity and the calculation quality of the estimated path used, combined with the initial data quality value. S5: Call the preset percentile standard library to convert the aerobic exercise index into aerobic capacity percentile and the anaerobic exercise index into anaerobic capacity percentile. S6: Based on the aerobic exercise index, the anaerobic exercise index, the aerobic capacity percentile, the anaerobic capacity percentile, and the corresponding data confidence, generate standardized exercise ability feature data according to a preset data structure. S7: Output the standardized athletic ability characteristic data to the artificial intelligence sports and health assessment module as input data for the artificial intelligence sports and health assessment module.

2. The method according to claim 1, characterized in that, The user's basic information includes at least age, gender, weight, maximum heart rate, and resting heart rate; the smart sports equipment includes one or more of the following: treadmill, exercise bike, rowing machine, elliptical trainer, electronic strength training equipment, electronic resistance training equipment, and wearable heart rate device; the multi-source sports data includes real-time sports data and / or historical sports data.

3. The method according to claim 1, characterized in that, The exercise type is identified based on at least one of the device type and device identifier, and at least one of the action type, data field combination, data continuity, and action cycle characteristics; when the multi-source exercise data includes continuous time-series heart rate data and power data, equivalent power data, or speed data that can be used to generate equivalent power data, it is identified as aerobic exercise; when the multi-source exercise data includes discrete action cycle and at least one of resistance, repetition count, action displacement, action speed, or action power, it is identified as anaerobic exercise; when the identification conditions for both aerobic and anaerobic exercise are met simultaneously, it is identified as mixed exercise, and the aerobic exercise index and anaerobic exercise index are generated respectively.

4. The method according to claim 1, characterized in that, The first sliding window is 5 seconds, the second sliding window is 60 seconds, and the third sliding window is 120 seconds. Sampling points that simultaneously satisfy the power standard deviation being less than a preset fluctuation threshold and the average value of the first sliding window being higher than a preset heart rate threshold are determined as steady-state candidate points. When the number of consecutive steady-state candidate points reaches a preset number or the duration reaches a preset duration, the effective steady-state interval is determined. Logarithmic regression is performed with the natural logarithm of the relative elapsed time within the effective steady-state interval as the independent variable and the corresponding first sliding window average as the dependent variable. The corrected heart rate is generated when the number of data points involved in the fitting and the fitting quality meet preset conditions.

5. The method according to claim 1, characterized in that, The individualized heart rate-power saturation response relationship is HR(P)=restingHR+HRR / [1+B×exp(-k×(PO))], where HRR=maxHR-restingHR; the power-heart rate response coefficient k is calculated according to R_adj=HRR / (AdjHR-restingHR)-1 and k=-ln(R_adj / B) / (P_s-O); Set the target heart rate TargetHR=α×maxHR, and calculate the target power corresponding to a single effective steady-state sample according to R_target=HRR / (TargetHR-restingHR)-1 and P_target=-ln(R_target / B) / k+O; The maximum oxygen uptake power PVO2max is determined based on one or more valid P_targets, and the maximum oxygen uptake is calculated according to VO2max=C0+C1×PVO2max / weight; Wherein, P_s is the power representative value corresponding to the effective steady-state interval, AdjHR is the corrected heart rate, weight is the user's weight, maxHR is the user's maximum heart rate, restingHR is the user's resting heart rate, B is the curve shape parameter of the heart rate-power saturation response relationship, O is the power offset parameter, α is the target heart rate ratio, C0 is the basic oxygen uptake parameter, and C1 is the power-oxygen uptake comprehensive parameter; B, O, α, C0 and C1 are preset parameters or parameters determined according to calibration data, and 0 < α < 1, maxHR > restingHR, restingHR < AdjHR < maxHR, restingHR < TargetHR < maxHR, B > 0, P_s ≠ O, R_adj > 0, R_target > 0, k is a finite positive value and P_target > 0.

6. The method according to claim 1, characterized in that, The maximum load per repetition is calculated in at least one of the following ways: based on the training resistance W, the number of repetitions R, and an empirical coefficient K greater than 0, it is calculated according to 1RM=W×(1+R / K); or, the centripetal phase movement speed is collected under different training resistances, and the load-speed relationship V=ab×W is fitted, where a and b are model fitting parameters, and under the condition that b>0 and a>Vmin, it is calculated according to the preset minimum effective movement speed Vmin according to 1RMv=(a-Vmin) / b.

7. The method according to claim 1, characterized in that, When the maximum resistance of the intelligent sports device is lower than a preset resistance threshold, the representative value of the strength ability is calculated based on at least two of the following: maximum force per unit body weight, peak power per unit body weight, movement speed stability, training completion rate, and fatigue decay rate. The fatigue decay rate and training completion rate are limited to 0 to 1. All indicators are normalized using N(x) = min(1, max(0, (xL) / (UL))), where L and U are the lower and upper reference limits corresponding to the movement type, gender, age group, device type, or user group, and U > L. The movement speed stability is calculated based on the coefficient of variation of the movement speed and limited to 0 to 1. When any sub-indicator is missing, the weights of the remaining sub-indicators are re-normalized.

8. The method according to claim 1, characterized in that, The percentile standard library stores reference data according to gender, age group, indicator type, indicator name, and percentile threshold. When an exercise indicator falls between two adjacent percentile thresholds, the corresponding percentile is obtained by linear interpolation, nearest neighbor, or interval mapping. When there are multiple anaerobic exercise indicators, the anaerobic capacity percentiles corresponding to each exercise are weighted or unweighted and aggregated to obtain the comprehensive anaerobic capacity percentile. The aerobic capacity percentile and the comprehensive anaerobic capacity percentile are retained separately.

9. The method according to claim 1, characterized in that, The standardized exercise capacity characteristic data includes at least one of the following: indicator identifier, indicator value, indicator unit, data source, indicator generation time, data confidence level, and missing data marker, and includes at least one of the following: device type, calibration version, and data version; wherein, the indicator value includes at least one of the following: maximum oxygen uptake, aerobic capacity percentile, single maximum load, representative value of strength capacity, and anaerobic capacity percentile.

10. The method according to claim 1, characterized in that, The standardized athletic ability characteristic data is output to the AI-powered sports and health assessment module through a local function interface, application programming interface, database, message queue, or network interface. The AI-powered sports and health assessment module can then identify and call the corresponding athletic indicators based on at least one of the following: indicator identifier, indicator unit, data source, indicator generation time, data confidence level, missing data marker, device type, calibration version, and data version contained in the standardized athletic ability characteristic data.

11. A system for generating exercise indicators for AI-based sports and health assessment, characterized in that, include: The information acquisition module is used to acquire basic user information and collect multi-source motion data from at least one smart sports device; The preprocessing and type identification module is used to perform time synchronization, unit standardization and abnormal data processing on the multi-source motion data, calculate the initial data quality value, and identify the motion type based on at least one of the following: device type and device identifier, as well as at least one of the following: motion type, data field combination, data continuity and motion cycle characteristics. The aerobic index generation module is used to calculate the average value of heart rate data in a first sliding window, the average value of power data or equivalent power data in a second sliding window, and the power standard deviation of raw power data or equivalent power data in a third sliding window when the exercise is identified as aerobic exercise; and to determine the effective steady-state interval based on the power standard deviation and the average value of the first sliding window. The corrected heart rate is obtained by fitting the heart rate time response relationship with the natural logarithm of the relative elapsed time within the effective steady-state interval as the independent variable and the corresponding first sliding window average as the dependent variable. Based on the corrected heart rate and power representative values, an individualized heart rate-power saturation response relationship is established, the target power is solved in reverse to generate the maximum oxygen uptake, and the confidence level of aerobic data is updated. The anaerobic performance indicator generation module is used to perform at least one of the following processes when the exercise is identified as anaerobic: generate the maximum single load based on a combination of training resistance and repetitions or different combinations of training resistance and movement speed; generate a representative value of strength ability based on at least two strength performance characteristics. At least one of the obtained single maximum load and strength capacity representative values ​​is used as the anaerobic exercise index. The anaerobic data confidence is updated and obtained based on the number of effective repetitions, load coverage, movement integrity and the calculation quality of the estimated path used, combined with the initial data quality value. The percentile conversion module is used to convert the aerobic exercise index and the anaerobic exercise index into aerobic capacity percentiles and anaerobic capacity percentiles, respectively. The standardized feature generation module is used to generate standardized motion capability feature data according to a preset data structure. And a feature output module, used to output the standardized athletic ability feature data to the artificial intelligence sports and health assessment module.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 10.

13. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 10.

Images