Rehabilitation exercise risk assessment method and device

Abstract

The invention discloses a rehabilitation exercise risk assessment method and device, and relates to the technical field of exercise rehabilitation, and the method comprises the steps: building a body load baseline according to user basic information and static medical history, collecting the exercise posture data of a user through a wearable device, and determining and triggering exercise risk monitoring through three-level abnormal event judgment; and backtracking physiological parameter data, establishing a real-time state vector, comparing the real-time state vector with a body load baseline, outputting a real-time motion risk coefficient, matching a real-time grading intervention strategy, outputting a self-adaptive intervention instruction after an ultra-short-term alarm log is corrected, and driving motion behavior intervention correction. The technical problems of one-sided data and insufficient accuracy caused by the fact that a traditional rehabilitation exercise risk assessment method does not integrate dynamic exercise postures and real-time physiological parameters are solved, accurate and comprehensive assessment, graded intervention and closed-loop tracking of rehabilitation exercise risks are achieved, the exercise injury probability is effectively reduced, and the rehabilitation exercise risk assessment efficiency is improved. And safety and effectiveness of rehabilitation of the patient are guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sports rehabilitation, in particular to a rehabilitation exercise risk assessment method and device. BACKGROUND

[0002] In the biomedical engineering industry, rehabilitation exercise of cardiovascular patients is crucial to their health recovery and quality of life improvement, and exercise risk assessment is the core link to ensure rehabilitation safety. In the prior art, the assessment of rehabilitation exercise risk of cardiovascular patients mainly relies on static medical history such as disease course, medication records or single basic physiological parameters such as static heart rate, or through artificial observation of exercise state judgment. These methods have certain effect in stable medical scenes, but with the improvement of rehabilitation demand, traditional assessment methods have exposed many limitations in dynamic exercise. Due to the dynamic changes of the body state of cardiovascular patients during rehabilitation exercise and the complexity of the exercise scene, the traditional assessment method cannot accurately integrate the exercise posture and real-time physiological parameters to comprehensively detect the risk, resulting in one-sided data, inaccurate, and difficult to meet the demand for accurate assessment and effective intervention of rehabilitation exercise risk. SUMMARY

[0003] The present application provides a rehabilitation exercise risk assessment method and device, which solves the technical problem that the traditional rehabilitation exercise risk assessment method does not integrate dynamic exercise posture and real-time physiological parameters, resulting in one-sided data and insufficient accuracy.

[0004] In a first aspect, the present application provides a rehabilitation exercise risk assessment method, which comprises: constructing a body load baseline according to user basic information and static medical history data; collecting exercise posture data of the user through a wearable device, wherein the exercise posture data includes real-time center of gravity position projection trajectory and real-time posture oscillation amplitude; performing three-level abnormal event determination based on the exercise posture data, triggering exercise risk monitoring, and then performing physiological parameter data backtracking, wherein the physiological parameter data includes time series heart rate, time series heart rate variability, time series blood oxygen and time series blood pressure; constructing a real-time state vector based on the physiological parameter data; performing dynamic deviation comparison of the real-time state vector on the body load baseline, and outputting a real-time exercise risk coefficient; matching a real-time grading intervention strategy according to the real-time exercise risk coefficient; correcting the real-time grading intervention strategy by backtracking the ultra-short-term alarm log, and outputting an adaptive intervention instruction; and driving exercise behavior intervention correction by using the adaptive intervention instruction.

[0005] In a second aspect of the present application, a rehabilitation exercise risk assessment device is provided, comprising: a physical load baseline construction module for constructing a physical load baseline according to user basic information and static medical history data; an exercise posture data acquisition module for collecting exercise posture data of a user through a wearable device, wherein the exercise posture data comprises real-time center of gravity position projection trajectory and real-time posture oscillation amplitude; a physiological parameter data backtracking module for performing three-level abnormal event determination based on the exercise posture data, triggering exercise risk monitoring, and performing physiological parameter data backtracking, wherein the physiological parameter data comprises time-series heart rate, time-series heart rate variability, time-series blood oxygen, and time-series blood pressure; a real-time state vector construction module for constructing a real-time state vector based on the physiological parameter data; a real-time exercise risk coefficient acquisition module for performing dynamic deviation comparison of the real-time state vector on the physical load baseline to output a real-time exercise risk coefficient; a real-time hierarchical intervention strategy matching module for matching a real-time hierarchical intervention strategy according to the real-time exercise risk coefficient; an adaptive intervention instruction acquisition module for correcting the real-time hierarchical intervention strategy by backtracking super-short-term alarm logs to output adaptive intervention instructions; and an adaptive intervention instruction execution module for driving exercise behavior intervention correction using the adaptive intervention instructions.

[0006] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0007] The present application collects user static data basic information, medical history, and dynamic real-time data exercise posture and physiological parameters, processes to obtain risk-related data through construction of a dynamic risk classification model and dynamic deviation comparison, calculates a real-time exercise risk coefficient and matches a hierarchical intervention strategy, adjusts in combination with closed-loop tracking after intervention, thereby accurately assessing the risk situation in rehabilitation exercise, making the rehabilitation exercise risk assessment and intervention results more accurate and reliable, achieving accurate and comprehensive assessment of rehabilitation exercise risk, hierarchical intervention, and closed-loop tracking, effectively reducing the probability of exercise injury, and ensuring the safety and effectiveness of patient rehabilitation. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0009] Figure 1 is a flowchart of a rehabilitation exercise risk assessment method provided by an embodiment of the present application.

[0010] Figure 2This is a schematic diagram of the structure of a rehabilitation exercise risk assessment device provided in an embodiment of this application.

[0011] Figure labeling: Module 1: Body load baseline construction; Module 2: Movement posture data acquisition; Module 3: Physiological parameter data backtracking; Module 4: Real-time state vector construction; Module 5: Real-time movement risk coefficient acquisition; Module 6: Real-time graded intervention strategy matching; Module 7: Adaptive intervention instruction acquisition; Module 8: Adaptive intervention instruction execution. Detailed Implementation

[0012] This application provides a method and device for assessing rehabilitation exercise risk, which solves the technical problem that traditional rehabilitation exercise risk assessment methods do not integrate dynamic movement posture and real-time physiological parameters, resulting in incomplete data and insufficient accuracy.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] Example 1, as Figure 1 As shown, a method for assessing the risk of rehabilitation exercise, wherein the method includes:

[0016] A baseline of physical burden is constructed based on user basic information and static medical history data.

[0017] Specifically, first, a plurality of standard exercise scenes are divided according to the characteristics of exercise physiology, and then a plurality of groups of dominant monitoring parameters are defined for each scene. Subsequently, in combination with the dominant monitoring parameters, the user's static medical history data is directionally quantized to obtain a plurality of groups of scene reference values. Based on the scene reference values, a plurality of associated load dynamic baselines are constructed, and historical training data is called to perform covariance statistics based on the scene reference values to generate a plurality of associated covariance matrices. Finally, the plurality of standard exercise scenes, the plurality of associated load dynamic baselines, and the plurality of associated covariance matrices are associated and stored, thereby completing the configuration of the body load baseline.

[0018] The motion posture data of the user is collected by a wearable device, wherein the motion posture data includes a real-time center of gravity position projection trajectory and a real-time posture oscillation amplitude.

[0019] Optionally, when collecting the motion posture data of the user, a wearable device carrying a three-axis accelerometer and a three-axis gyroscope is selected and worn on the waist or wrist of the user or other parts that can reflect the overall posture of the body. First, the linear acceleration data of the user during the motion is obtained through the three-axis accelerometer of the device, and the angular velocity data is obtained through the three-axis gyroscope.

[0020] For obtaining the real-time center of gravity position projection trajectory, the Kalman filtering algorithm is first used to fuse the linear acceleration and angular velocity data, and the sensor error is corrected to obtain the pitch angle, roll angle and yaw angle of the user in the three-dimensional space. In combination with the geometric parameters such as the average distance of the center of gravity of the adult body from the waist or wrist in the existing ergonomics database, the real-time coordinates of the center of gravity of the user's body in the three-dimensional coordinate system are calculated according to the above attitude angles. Then, the three-dimensional coordinates are projected onto the ground two-dimensional plane according to the coordinate system setting, taking the initial wearing position of the wearable device as the origin and the horizontal direction as the X-axis and Y-axis, to obtain the projection point of the center of gravity in the two-dimensional plane. With continuous collection and calculation of the projection point during the user's motion, the continuous projection points are sequentially connected in time order, and the real-time center of gravity position projection trajectory is obtained.

[0021] For obtaining the real-time posture oscillation amplitude, the angular velocity sequence around each coordinate axis is extracted from the angular velocity data collected by the wearable device, and a sliding time window is used to set 1 second as a time window to segment the angular velocity sequence. The maximum and minimum values of the angular velocity in each time window are calculated, and the difference between the two values is obtained to obtain the angular velocity fluctuation range in the window. According to the corresponding relationship between the angular velocity fluctuation range and the posture oscillation degree in the field of kinematics, the angular velocity fluctuation range is quantized as the posture oscillation amplitude value. By continuously processing the angular velocity data through the sliding time window, the posture oscillation amplitude value corresponding to each window is continuously output, and the real-time posture oscillation amplitude is obtained.

[0022] Based on the motion posture data, three-level abnormal event determination is performed, and after triggering the motion risk monitoring, physiological parameter data is traced back, wherein the physiological parameter data includes time series heart rate, time series heart rate variability, time series blood oxygen and time series blood pressure.

[0023] In an embodiment of the present application, first, motion pattern analysis is carried out according to the spatial vector time sequence of the real-time center of gravity position projection trajectory, and a first set of alternative motion scenes is generated. Second, the frequency domain energy distribution features of the real-time posture oscillation amplitude are extracted, and motion pattern analysis is also carried out to obtain a second set of alternative motion scenes. Subsequently, intersection operation is performed on the first set of alternative motion scenes and the second set of alternative motion scenes to determine the real-time motion scene. Then, the determined real-time motion scene is used as a criterion to implement three-level abnormal event determination on the motion posture data, and a spatio-temporal coupled abnormal event is output. Finally, if the output spatio-temporal coupled abnormal event is not empty, the motion risk monitoring is triggered, and a triggering condition is established for subsequent physiological parameter data tracing back.

[0024] Based on the physiological parameter data, a real-time state vector is constructed.

[0025] Specifically, first, the acquired time series heart rate, time series heart rate variability, time series blood oxygen and time series blood pressure data are preprocessed. The original data are processed by using the moving average method, a fixed moving window is set, for example, 5 data points are taken as one window, the average value of the data in each window is calculated, and the original data in the window are replaced by the average value, so that random noise in the data is removed, and the smoothed physiological parameter time series data are obtained. Then, the smoothed physiological parameter time series data are processed by using Min-Max standardization. First, the maximum and minimum values of each physiological parameter in the historical data are counted, and then the current smoothed parameter data are substituted into the formula: standardized data = (current data - parameter minimum value) / (parameter maximum value - parameter minimum value), so that all physiological parameter data are mapped to the unified interval of 0-1, and the influence of different parameter units and magnitude differences is eliminated.

[0026] Then the time dimension sampling interval of the real-time state vector is determined. According to the real-time requirement of the rehabilitation exercise risk assessment, a fixed sampling interval of 1 second is set, and at each sampling time, the standardized time series heart rate, time series heart rate variability, time series blood oxygen and time series blood pressure data are extracted as the basic elements of the vector. Then arrange the elements in the preset parameter order. The fixed order of physiological parameters is set in advance as time series heart rate, time series heart rate variability, time series blood oxygen, and time series blood pressure. The four parameter standardized data extracted at each sampling time are arranged in this order in turn to form a four-dimensional data sequence. Finally, the integrity of the data sequence is verified. Check if there are missing values or abnormal values in the arranged four-dimensional data sequence, such as values outside the 0-1 interval. If there are, use the parameter value of the previous sampling time to complete the data sequence to ensure that the data sequence is complete and effective. The complete four-dimensional data sequence is the real-time state vector.

[0027] The real-time state vector is used for dynamic deviation comparison of the body load baseline, and a real-time exercise risk coefficient is output.

[0028] Specifically, first, according to the positioned real-time exercise scene, the corresponding adaptive load dynamic baseline and adaptive covariance matrix are matched and called in the pre-constructed body load baseline to provide reference data for deviation comparison. Then the real-time training data in the dynamic backtracking time window is called, and the real-time covariance matrix is constructed based on these data, and is dynamically fused with the adaptive covariance matrix to generate an adaptive covariance matrix that can reflect the current exercise state. Then the adaptive covariance matrix is used as a measurement matrix, and the Mahalanobis distance of the real-time state vector relative to the adaptive load dynamic baseline is calculated to quantify the dynamic deviation between the two. Finally, the Mahalanobis distance is processed for risk mapping to convert it into a real-time exercise risk coefficient that can be directly used for risk judgment, and then the dynamic deviation comparison and risk coefficient output are completed.

[0029] According to the real-time exercise risk coefficient, a real-time grading intervention strategy is matched.

[0030] Optionally, first, the grading interval of the real-time exercise risk coefficient is preset. Based on clinical rehabilitation exercise risk case data, the distribution range of the risk coefficient under different risk levels is counted, and the risk coefficient of 0-100 is divided into three intervals of low, medium and high, for example, 0-30 is the low risk interval, 31-70 is the medium risk interval, and 71-100 is the high risk interval. Each interval corresponds to a unique risk level label.

[0031] Then a hierarchical intervention strategy library is constructed. For each risk level, specific intervention measures are developed in combination with rehabilitation medicine exercise intervention specifications. For example, a low risk level corresponds to a strategy of maintaining the current exercise intensity and monitoring physiological parameters every 5 minutes, a medium risk level corresponds to a strategy of reducing exercise intensity by 20% and monitoring physiological parameters every 2 minutes, and a high risk level corresponds to a strategy of immediately pausing exercise and evaluating physiological parameter recovery. The risk level and the corresponding intervention strategy are associated and stored to form the hierarchical intervention strategy library.

[0032] Then the risk coefficient is matched with the level. After obtaining the real-time exercise risk coefficient, it is compared with the preset hierarchical interval one by one to determine the risk level corresponding to the interval to which the coefficient belongs. For example, when the real-time risk coefficient is 28, it is determined by comparison to belong to the low risk interval corresponding to the low risk level; when the coefficient is 65, it belongs to the medium risk interval corresponding to the medium risk level. Finally, the corresponding intervention strategy is called. According to the matched risk level, the intervention strategy associated with the level is located and called through level indexing in the hierarchical intervention strategy library, completing the matching of the real-time hierarchical intervention strategy.

[0033] The real-time hierarchical intervention strategy is corrected by backtracking the ultra-short-term alarm log, and an adaptive intervention instruction is output.

[0034] In an embodiment of the present application, first, the time-series hierarchical intervention strategy is called from the ultra-short-term alarm log. Then the time-series hierarchical intervention strategy is traversed, and the recurrence frequency of the current real-time hierarchical intervention strategy in the log is counted and obtained. Subsequently, according to the comparison result of the recurrence frequency and the preset recurrence frequency threshold, two cases are handled, if the recurrence frequency does not exceed the preset threshold, the intervention instruction of the current real-time hierarchical intervention strategy is directly matched as the adaptive intervention instruction output; if the recurrence frequency exceeds the preset threshold, the intervention instruction of the current real-time hierarchical intervention strategy is adjusted by N-level transition, and the adjusted intervention instruction is output as the adaptive intervention instruction, so as to realize dynamic correction of the intervention strategy and improve the pertinence and effectiveness of the intervention.

[0035] The adaptive intervention instruction is used to drive exercise behavior intervention correction.

[0036] Specifically, first, the adaptive intervention instruction is analyzed, and the intervention types contained in the instruction are extracted, such as adjusting exercise intensity, pausing exercise, changing action amplitude, and specific parameters such as target exercise intensity value, pause duration, and action amplitude adjustment ratio. The instruction is converted from text or code form to structured data that can be directly used for calculation, providing a clear basis for subsequent correction actions. Then the current exercise state is compared with the target state. By obtaining the user's current exercise data including real-time exercise intensity, action amplitude, and exercise duration, the state comparison algorithm is used to compare the current data with the target parameters obtained by instruction analysis one by one, calculate the difference value, and then determine the specific dimension and degree that need to be corrected.

[0037] Then the motion behavior correction control signal is generated. According to the calculated difference value and converting the difference value into a control signal recognizable by the execution device, for example, for the motion intensity difference, a device control code of reducing 20% resistance is generated; for the pause instruction, a pause signal of 3 seconds high level is generated, ensuring that the control signal matches the interface protocol of the execution device. Then the execution device is driven to output the correction prompt. The generated control signal is sent to the corresponding execution device, such as a smart fitness equipment, a wearable device, a rehabilitation APP, and after the execution device receives the signal, the execution device delivers the correction instruction to the user through a preset prompt mode, such as automatic adjustment of resistance of the equipment, vibration prompt of the wearable device, and voice broadcast of the APP, guiding the user to adjust the motion behavior.

[0038] Finally, the correction effect is monitored and adjusted in a closed loop. By continuously collecting the motion data of the user after the correction, it is judged whether the corrected motion state meets the requirements of the instruction target. If yes, the correction process is stopped; if not, for example, if the user does not reduce the intensity as prompted, the difference value is recalculated and a new control signal is generated, the execution device is prompted again, and the user's motion behavior reaches the instruction target state.

[0039] Further, the method provided by the embodiment of the present application comprises:

[0040] Based on the physiological characteristics of exercise, the standard exercise scenes are divided to obtain a plurality of standard exercise scenes; a plurality of groups of dominant monitoring parameters of the plurality of standard exercise scenes are defined; the static medical history data is directionally quantized and converted according to the plurality of groups of dominant monitoring parameters to obtain a plurality of groups of scene reference values; a plurality of associated load dynamic baselines are constructed based on the plurality of groups of scene reference values; covariance statistics of the historical training data are performed using the plurality of groups of scene reference values to construct a plurality of associated covariance matrices; the plurality of standard exercise scenes, the plurality of associated load dynamic baselines, and the plurality of associated covariance matrices are stored in association to complete the configuration of the body load baseline.

[0041] Specifically, first, the standard exercise scenes are divided based on the physiological characteristics of exercise. Referring to the mature classification logic of rehabilitation training in exercise therapy, the training is summarized according to the energy consumption, muscle involvement site and physiological response difference during exercise, such as classifying the training with joint activity as a joint activity degree scene, classifying the training with muscle strength improvement as a muscle strength training scene, and classifying the training focusing on body balance maintenance as a balance training scene. Through the summary and induction of the existing rehabilitation exercise types, a plurality of standard exercise scenes are obtained.

[0042] Then the dominant monitoring parameters of each standard exercise scene are defined, which are screened by the person skilled in the art in combination with clinical rehabilitation monitoring experience and biomechanical principles. In the balance training scene, according to the conventional indicators of balance evaluation, heart rate is selected as a physiological parameter to reflect the body load state, and the center of gravity offset is selected as a posture parameter to reflect the balance stability; in the muscle strength training scene, electromyographic signal is selected as a physiological parameter to monitor the muscle activation degree, and the joint activity angle is selected as a posture parameter to record the movement amplitude. For other standard exercise scenes, the core indicators are directly screened from the existing rehabilitation monitoring commonly used parameters, forming a monitoring combination containing physiological and posture parameters for each group.

[0043] Then the static medical history data is directionally quantized and converted by using a combination of assignment method and correction coefficient method. For example, for the stroke course, the assignment is made according to the time stage, with 1 year or less assigned as 1, 1-3 years assigned as 2, and 3 years or more assigned as 3; for the medication record, the correction coefficient is set according to the influence of the drug on the physiological indicators, such as the correction coefficient of 0.9 for taking conventional antihypertensive drugs on heart rate parameters. The assignment result is multiplied by the correction coefficient, and then adjusted in combination with the monitoring parameter type of the corresponding scene to obtain multiple scene reference values adapted to different scenes.

[0044] Subsequently, the associated load dynamic baseline is constructed, with the reference values of each standard exercise scene as the center, the historical load data of the same type of patients in the standard exercise scene is retrieved, the mean and standard deviation of these data are calculated to form the initial load baseline range. Then, the baseline load is fine-tuned in combination with the basic information of the current user such as age and weight, such as reducing the upper limit of the baseline load by 5% for every 5 kg overweight than the standard range, and finally obtaining multiple associated load dynamic baselines adapted to individuals.

[0045] Then the associated covariance matrix is constructed, the monitoring parameter sequences in the historical training data under the corresponding standard exercise scene are retrieved, such as heart rate and center of gravity offset data in the balance scene, the mean of each group of parameters is calculated, and then the deviation of each data from the mean is calculated. The sum of the heart rate deviation and the center of gravity offset deviation is multiplied by the number of data samples minus 1 to obtain the covariance value. Similarly, the covariances between all parameters are calculated to form the associated covariance matrix, which reflects the correlation between parameters.

[0046] Finally, a unified data table is created in the database, the standard exercise scene name is taken as the main index, and the associated load dynamic baseline data and the associated covariance matrix data under the scene are stored correspondingly, establishing a one-to-one correspondence between the standard exercise scene and the associated load dynamic baseline and the associated covariance matrix, realizing the structured storage and quick retrieval of data, and finally completing the configuration of the body load baseline.

[0047] By combining existing rehabilitation classification experience, clinical monitoring indicators, basic data quantification methods and statistical analysis methods, a body load baseline suitable for different exercise scenarios and individual characteristics is constructed, which provides accurate and practical basic data support for subsequent rehabilitation exercise risk assessment.

[0048] Further, the method provided by the embodiment of the application comprises:

[0049] According to the space vector time sequence of the real-time center of gravity position projection trajectory, motion mode analysis is performed, and a first candidate motion scene set is output; the frequency domain energy distribution characteristics of the real-time attitude oscillation amplitude are extracted for motion mode analysis, and a second candidate motion scene set is output; intersection solving is performed on the first candidate motion scene set and the second candidate motion scene set, and a real-time motion scene is located; the real-time motion scene is taken as a judgment reference, three-level abnormal event judgment is performed on the motion attitude data, and a space-time coupling abnormal event is output; if the space-time coupling abnormal event is a non-empty set, motion risk monitoring is triggered.

[0050] Optionally, first, the space vector time sequence is divided into a plurality of continuous sliding windows according to a fixed time interval of 1 second, direction angle mean values and length variance of the space vector in each window are calculated, and then a standard motion scene feature template library is called, the template library is established based on space vector features of common rehabilitation motion scenes such as balance training and gait training, and by calculating the cosine similarity of the window feature parameters and the feature of each standard scene template, the standard motion scenes with a similarity higher than a preset threshold value 0.8 are included in the first candidate motion scene set.

[0051] Then, the time domain data of the real-time attitude oscillation amplitude are sampled, the sampling frequency is set according to the performance of the wearable device, such as 50 Hz, fast Fourier transform is performed on the sampled data, and energy values corresponding to different frequency components are obtained. Subsequently, the frequency range is divided into a plurality of frequency bands, for example, 0-1 Hz, 1-5 Hz, 5-10 Hz, and the like, the proportion of the energy of each frequency band in the total energy is calculated, and a frequency domain energy distribution characteristic is formed. Then, the real-time calculated frequency domain energy distribution characteristics are converted into a feature vector, and the frequency domain energy distribution characteristic vectors of each scene in the standard motion scene template library are called. The cosine similarity algorithm is used to calculate the similarity values of the real-time feature vector and the feature vectors of each standard motion scene, a similarity threshold value is preset, such as 0.8, if the similarity value of a certain standard scene and the real-time feature is greater than or equal to the threshold value, it is determined that the feature matching degree meets the requirement, and the standard scene is included in the second candidate motion scene set.

[0052] Then, list all the standard motion scene names contained in the two candidate sets of motion scenes, and compare them one by one with the standard motion scenes in the two sets to filter out the scene elements that exist in both sets. If a unique standard motion scene is obtained after filtering, then the standard motion scene is the real-time motion scene; if multiple common scenes exist, further filtering can be done by adding feature matching dimensions, such as adding motion duration features, to finally determine the unique real-time motion scene.

[0053] Next, based on the located real-time motion scene, the corresponding spatial and temporal anomaly detection rules are retrieved. Then, spatial and temporal features are extracted from the motion posture data to provide a data foundation for anomaly detection. Following this, the retrieved spatial and temporal anomaly detection rules are used to perform spatiotemporal anomaly detection on the extracted two types of features through rule mapping. If any level of spatiotemporal coupling detection fails to meet the conditions, the corresponding spatiotemporal mismatch type is invoked and output as a spatiotemporal coupling anomaly event. This completes the core anomaly identification and result output stage in the three-level anomaly event judgment, a step that will be explained in detail later.

[0054] Finally, the number of elements in the output spatiotemporal coupling abnormal event set is counted. If the count result is greater than 0, that is, the set is not empty, the preset motion risk monitoring mechanism is triggered. This mechanism usually includes starting the high-frequency data recording function of the physiological parameter acquisition device and activating the local risk warning of the device, etc., to prepare for subsequent physiological parameter data backtracking.

[0055] By combining methods such as sliding window feature matching, fast Fourier transform, set operations, threshold hierarchical comparison, and set counting judgment, accurate positioning of real-time motion scenarios and effective triggering of motion risk monitoring are achieved, providing accurate scenario benchmarks and triggering conditions for subsequent rehabilitation exercise risk assessment.

[0056] Furthermore, the method provided in this application embodiment includes:

[0057] Based on the real-time motion scene, retrieve the spatial dimension anomaly judgment rules and the temporal dimension anomaly judgment rules; extract spatial dimension features and temporal dimension features from the motion posture data; use the spatial dimension anomaly judgment rules and the temporal dimension anomaly judgment rules to map and perform spatiotemporal anomaly detection of the spatial dimension features and temporal dimension features; if any level of spatiotemporal coupling detection fails, call the spatiotemporal mismatch type as the spatiotemporal coupling anomaly event output.

[0058] Specifically, a rehabilitation exercise scenario rule database is first pre-constructed. This database stores spatial dimension anomaly detection rules for different standard exercise scenarios, such as balance training, gait training, and muscle strength training. These rules include thresholds for center of gravity shift range and joint range limitations. Temporal dimension anomaly detection rules also exist, such as thresholds for postural oscillation duration and movement frequency ranges. Each standard exercise scenario is uniquely associated with one of these two types of rules. Once a real-time exercise scenario is determined, the rule database is searched using index matching. The spatial dimension and temporal dimension anomaly detection rules associated with that standard exercise scenario are then invoked, providing a basis for subsequent anomaly detection.

[0059] Next, spatial and temporal features are extracted from the motion posture data. For spatial features, based on information such as the real-time center of gravity projection trajectory and posture oscillation amplitude in the motion posture data, extraction is achieved by calculating parameters such as the spatial offset of the trajectory (i.e., the maximum offset distance relative to the initial position) and the spatial distribution range of posture oscillation (i.e., the range of spatial coordinate changes of key body points during oscillation). For temporal features, based on the temporal attributes of the motion posture data, a sliding time window with a 1-second interval is used to calculate the duration and frequency of posture oscillation within each window, such as the number of oscillations per unit time and the time interval of posture changes. This completes the extraction of temporal features, resulting in two types of dimensional feature data that can be used for anomaly detection.

[0060] Next, the spatial dimension features are analyzed to obtain three types of hierarchical feature parameters: frame-level spatiotemporal offset vector, periodic-level morphological distortion degree, and trend-level envelope expansion rate. Then, the temporal dimension features are analyzed to obtain three types of hierarchical feature parameters: frame-level instantaneous oscillation amplitude, periodic-level risk frequency band energy proportion, and trend-level dominant frequency migration rate. Subsequently, from the spatial dimension anomaly judgment rules, the corresponding judgment thresholds are called in a progressive order of frame-level, periodic-level, and trend-level, and spatial anomaly detection is performed on the three types of spatial hierarchical feature parameters obtained from the analysis based on these thresholds. Simultaneously, while performing spatial anomaly detection, the corresponding judgment thresholds are called in the temporal dimension anomaly judgment rules in the same progressive order, and spatiotemporal anomaly detection is performed on the three types of temporal hierarchical feature parameters obtained from the analysis based on these thresholds. Through hierarchical analysis and synchronous progressive detection, refined anomaly identification of spatiotemporal dimension features is achieved, ultimately completing the spatiotemporal anomaly detection process. This step will be explained in detail later.

[0061] Finally, a spatiotemporal mismatch type library is pre-built. This library defines the mismatch types corresponding to different coupling detection failures. For example, "spatial anomaly - temporal normality" corresponds to "spatial one-dimensional mismatch," "temporal anomaly - spatial normality" corresponds to "temporal one-dimensional mismatch," and "spatiotemporal anomaly but failure to meet coupling timing requirements" corresponds to "spatiotemporal timing mismatch," etc. When any level of coupling failure is detected, the specific combination of anomalies is analyzed, such as spatial anomaly only, temporal anomaly only, or spatiotemporal anomaly but timing mismatch. Based on this combination of anomalies, the corresponding mismatch type is matched in the spatiotemporal mismatch type library and output as a spatiotemporal coupling anomaly event.

[0062] By combining methods such as pre-defined rule database retrieval, basic feature extraction, threshold comparison coupling detection, and pre-defined mismatch type invocation, the system achieves accurate determination and output of spatiotemporal coupling abnormal events based on real-time motion scenarios, providing clear abnormal identification results to support rehabilitation exercise risk assessment.

[0063] Furthermore, the method provided in this application embodiment includes:

[0064] The spatial dimension features are analyzed to obtain the frame-level spatiotemporal offset vector, periodic morphological distortion degree, and trend-level envelope expansion rate. The temporal dimension features are analyzed to obtain the frame-level instantaneous oscillation amplitude, periodic risk frequency band energy ratio, and trend-level main frequency migration rate. The frame-level scene safety zone, periodic trajectory morphology correlation threshold, and trend-level envelope area expansion rate threshold are progressively called from the spatial dimension anomaly judgment rules to map and perform spatial anomaly detection of the frame-level spatiotemporal offset vector, periodic morphological distortion degree, and trend-level envelope expansion rate. During spatial anomaly detection, the frame-level oscillation amplitude interval, periodic energy entropy change threshold, and trend-level main frequency migration rate threshold are synchronously and progressively called to map and perform spatiotemporal anomaly detection of the frame-level instantaneous oscillation amplitude, periodic risk frequency band energy ratio, and trend-level main frequency migration rate.

[0065] Specifically, the spatial dimension feature data is first divided into continuous frames of fixed frame length. For each frame, its spatial offset relative to a reference frame is calculated, and combined with the offset direction, a frame-level spatiotemporal offset vector is obtained. Then, the motion period is determined using the autocorrelation method. The trajectory morphology within each period is extracted and its difference from the standard period trajectory morphology is calculated to obtain the period-level morphological distortion. Finally, the trend envelope of the spatial feature data is drawn using the sliding window method, and the ratio of the area difference to the time difference between adjacent window envelopes is calculated to obtain the trend-level envelope expansion rate.

[0066] Next, the time-dimensional feature data is sampled frame by frame, and the oscillation amplitude data corresponding to each frame is directly read as the instantaneous value of the frame-level oscillation amplitude. A Fast Fourier Transform is performed on the time series data for each motion cycle. After dividing the data into regular and risk bands, the ratio of the sum of energy in the risk bands to the total energy is calculated to obtain the period-level risk band energy proportion. The dominant frequency position is tracked through continuous periodic frequency domain analysis, and the ratio of the difference in dominant frequency between adjacent cycles to the time interval is calculated to obtain the trend-level dominant frequency migration rate.

[0067] Then, based on the spatial dimension anomaly judgment rules, three thresholds are progressively applied to perform spatial anomaly detection. The frame-level scene safe zone is determined by collecting a large amount of normal frame-level spatiotemporal offset vector data under standard motion scenes of the same type, statistically analyzing the distribution range of these data, and taking the 95% confidence interval as the frame-level scene safe zone under that scene; the periodic trajectory morphology correlation threshold is determined by collecting trajectory morphology data of normal motion cycles, calculating the correlation coefficient between these trajectories and standard trajectories, and taking the minimum value among all correlation coefficients as the periodic trajectory morphology correlation threshold; the trend-level envelope area expansion rate threshold is determined by statistically analyzing all data on the envelope area expansion rate under normal motion trends, and taking the maximum value among them as the trend-level envelope area expansion rate threshold.

[0068] During spatial anomaly detection, the thresholds mentioned above are applied in a progressive order: frame-level, periodic-level, and trend-level. First, the frame-level scene safe zone is called, and the real-time calculated frame-level spatiotemporal offset vector is compared with this safe zone. If the offset vector exceeds the safe zone range, a frame-level spatial anomaly is determined. Next, the periodic-level trajectory morphology correlation threshold is called, and the correlation coefficient between the real-time periodic trajectory morphology and the standard trajectory is calculated. If this coefficient is less than the correlation threshold, a periodic-level spatial anomaly is determined. Finally, the trend-level envelope area expansion rate threshold is called, and the real-time trend-level envelope expansion rate is compared with this rate threshold. If the expansion rate is greater than the rate threshold, a trend-level spatial anomaly is determined.

[0069] Subsequently, during spatial anomaly detection, three thresholds are progressively invoked simultaneously, and the spatiotemporal anomaly detection process is executed. The frame-level oscillation amplitude range is determined by collecting frame-level oscillation amplitude data under normal motion scenarios, statistically analyzing the reasonable distribution range of these data, and setting this range as the frame-level oscillation amplitude range. The periodic-level energy entropy change threshold is determined by calculating the energy entropy value changes of a large number of normal motion cycles, and taking the maximum value among these changes as the periodic-level energy entropy change threshold. The trend-level main frequency migration rate threshold is determined by statistically analyzing all data on the main frequency migration rate under normal motion trends, and taking the maximum value among them as the trend-level main frequency migration rate threshold.

[0070] While spatial anomaly detection progresses from frame-level to period-level to trend-level, the time dimension also synchronously applies the aforementioned thresholds at the same progressive level: When spatial anomaly detection reaches the frame level, the frame-level oscillation amplitude range is synchronously applied, and the instantaneous value of the real-time frame-level oscillation amplitude is compared with this range. If the instantaneous value exceeds the range, a frame-level spatiotemporal anomaly is determined. When spatial anomaly detection reaches the period level, the period-level energy entropy change threshold is synchronously applied, and the entropy change of the real-time period-level risk frequency band energy proportion is calculated. If the change is greater than the entropy change threshold, a period-level spatiotemporal anomaly is determined. When spatial anomaly detection reaches the trend level, the trend-level main frequency migration rate threshold is synchronously applied, and the real-time trend-level main frequency migration rate is compared with this rate threshold. If the migration rate is greater than the rate threshold, a trend-level spatiotemporal anomaly is determined. Through this progressive detection with spatiotemporal synchronization, layered and accurate identification of motion attitude anomalies is achieved.

[0071] Furthermore, the method provided in this application embodiment includes:

[0072] Using the temporal hierarchy attribute of the spatiotemporal coupled abnormal event and the real-time motion scene as two-dimensional retrieval conditions, the retrospective time reference value is retrieved from the physiological retrospective rule mapping library; based on the coupling coefficient of the spatiotemporal coupled abnormal event, a window length correction coefficient is calculated; the retrospective time reference value is adjusted using the window length correction coefficient to obtain a dynamic retrospective time window; after triggering motion risk monitoring, the physiological parameter data is retrospectively retrieved using the dynamic retrospective time window as a retrospective constraint.

[0073] Specifically, when pre-constructing a physiological retrospective rule mapping library, it is necessary to determine the retrospective time benchmark value based on a large amount of historical rehabilitation exercise data and clinical validation results. First, collect physiological parameter retrospective data from different rehabilitation patients in various real-time exercise scenarios such as balance training and gait training, at the frame, period, and trend levels. Statistical analysis is then performed to determine the minimum retrospective duration that can completely cover abnormal precursor data under different combinations of "time level attribute - real-time exercise scenario." Next, combined with expert advice on the effective retrospective duration for each combination, the retrospective time benchmark value corresponding to each combination is determined. For example, in the balance training scenario, the frame level corresponds to 1 second, the period level to 3 seconds, and the trend level to 5 seconds; in the gait training scenario, the frame level corresponds to 1.5 seconds, the period level to 4 seconds, and the trend level to 6 seconds, and each combination is uniquely associated with its corresponding benchmark value. Once the time level attribute and real-time exercise scenario of the spatiotemporally coupled abnormal event are obtained, the library is searched using a two-dimensional matching method to locate the retrospective time benchmark value corresponding to that combination and complete the retrieval.

[0074] Next, the preset range of the coupling coefficient is 0 to 1. The larger the value, the higher the degree of abnormal coupling. A basic correction ratio is set, for example, the correction coefficient is 1 when the coupling coefficient is 0, and the correction coefficient is 1.6 when the coupling coefficient is 1. The actual obtained coupling coefficient is substituted into the preset linear formula, such as window length correction coefficient = 1 + 0.6 × coupling coefficient, and the window length correction coefficient is obtained through numerical calculation.

[0075] The retrieved backtracking time base value is then multiplied by the calculated window length correction factor to obtain the adjusted time length. For example, if the backtracking time base value is 4 seconds and the window length correction factor is 1.3, the two are multiplied to obtain 5.2 seconds. This duration is the range of the dynamic backtracking time window, that is, backtracking 5.2 seconds from the moment the abnormal event occurred.

[0076] After triggering exercise risk monitoring, physiological parameter data is retrieved back using a dynamic backtracking time window as a constraint. This physiological parameter data includes time-series heart rate, time-series heart rate variability, time-series blood oxygen saturation, and time-series blood pressure. The real-time collected physiological parameter data is stored in a database with timestamps. The specific acquisition process can be achieved through synchronous real-time acquisition via wearable devices: the wearable device is equipped with corresponding physiological sensors that continuously collect heart rate, blood oxygen saturation, and blood pressure data during the user's exercise process. Through signal processing, heart rate variability information is extracted from the raw signals, and various parameter data are recorded at fixed time intervals, forming a continuous data sequence containing a time dimension, thus obtaining time-series heart rate, time-series heart rate variability, time-series blood oxygen saturation, and time-series blood pressure.

[0077] When exercise risk monitoring is triggered, a time range determined by a dynamic backtracking time window is used. For example, if the time of the abnormality is T, the backtracking range in the above example is from T-5.2 seconds to T. Physiological parameter data within this range is retrieved from the database by timestamp matching. The retrieved data is then used to provide historical physiological data support for subsequent risk assessment.

[0078] Furthermore, the method provided in this application embodiment includes:

[0079] Based on the real-time motion scenario, the adaptive load dynamic baseline and adaptive covariance matrix are invoked according to the body load baseline; real-time training data within the dynamic backtracking time window are retrieved to construct a real-time covariance matrix, and an adaptive covariance matrix is ​​generated by dynamically fusing the adaptive covariance matrix and the real-time covariance matrix; the adaptive covariance matrix is ​​used as a metric matrix to calculate the Mahalanobis distance of the real-time state vector relative to the adaptive load dynamic baseline; the Mahalanobis distance is used for risk mapping, and the real-time motion risk coefficient is output.

[0080] In one embodiment, the body load baseline is first stored according to standard exercise scenarios, and an index table corresponding to the scenarios and data is established. Each standard exercise scenario is associated with a unique adaptive load dynamic baseline and an adaptive covariance matrix in the index table. The adaptive load dynamic baseline contains the normal fluctuation range data of physiological parameters under that scenario, and the adaptive covariance matrix reflects the historical correlation between physiological parameters under that scenario. When a real-time exercise scenario is determined, the entry corresponding to that scenario is located in the index table using an index lookup method, and the corresponding adaptive load dynamic baseline and adaptive covariance matrix are directly retrieved from the body load baseline.

[0081] Next, time-series heart rate, time-series heart rate variability, time-series blood oxygen, and time-series blood pressure data were collected within the time window. A sample covariance calculation method was used: first, the mean of each physiological parameter was calculated separately; then, the deviation of each data point from the corresponding mean was calculated; the deviations of all parameters were organized into a deviation matrix; the real-time covariance matrix was obtained by multiplying the transpose of the deviation matrix by itself and then dividing by the data sample size minus 1. When dynamically fusing the adaptive covariance matrix and the real-time covariance matrix, a weighted average method was used. The adaptive covariance matrix was set to a weight of 0.6 to reflect the stability of historical data, and the real-time covariance matrix was set to a weight of 0.4 to reflect the current motion state. The elements at corresponding positions of the two matrices were multiplied by their respective weights and then summed to generate the adaptive covariance matrix.

[0082] Next, the baseline vector is extracted from the adaptive load dynamic baseline. This baseline vector consists of normal baseline values ​​of physiological parameters under this scenario. The deviation vector is obtained by subtracting the baseline vector from the real-time state vector. Then, the adaptive covariance matrix is ​​inverted to obtain the inverse matrix. The deviation vector is then multiplied by the inverse matrix, followed by multiplication by the transpose of the deviation vector. Finally, the square root of the calculation result is taken, and the resulting value is the Mahalanobis distance of the real-time state vector relative to the adaptive load dynamic baseline. The larger this distance, the greater the deviation of the real-time physiological state from the normal baseline.

[0083] Finally, when using Mahalanobis distance for risk mapping, a pre-defined mapping rule is established: Mahalanobis distances of 0-2 correspond to a low-risk range, 2-4 to a medium-risk range, and above 4 to a high-risk range. This range is then linearly mapped to a risk coefficient range of 0-100, meaning a Mahalanobis distance of 0 corresponds to a risk coefficient of 0, and a distance of 4 corresponds to a risk coefficient of 100. Based on the calculated Mahalanobis distance, a specific risk coefficient is determined through linear interpolation. For example, a Mahalanobis distance of 1 corresponds to a risk coefficient of 25, and a distance of 3 corresponds to a risk coefficient of 75. The final output is this real-time motion risk coefficient.

[0084] By combining the weighted matrix calculated from the covariance of indexed matching samples with Mahalanobis distance calculation and linear risk mapping, the system achieves accurate quantification of the deviation between real-time physiological state and normal baseline, and outputs a real-time exercise risk coefficient that can be directly used for risk assessment, providing accurate data support for matching subsequent rehabilitation exercise risk intervention strategies.

[0085] Furthermore, the method provided in this application embodiment includes:

[0086] The time-series hierarchical intervention strategy is retrieved from the ultra-short-term alarm log; the time-series hierarchical intervention strategy is traversed to obtain the recurrence frequency of the real-time hierarchical intervention strategy; if the recurrence frequency does not exceed the preset recurrence frequency threshold, the intervention instruction of the real-time hierarchical intervention strategy is matched as the adaptive intervention instruction output; if the recurrence frequency exceeds the recurrence frequency threshold, the intervention instruction of the real-time hierarchical intervention strategy is jumped across N levels as the adaptive intervention instruction output.

[0087] Optionally, the ultra-short-term alarm logs are first stored in a structured format in the database. Each ultra-short-term alarm log entry contains information such as the intervention strategy identifier, execution timestamp, and intervention level. First, a short-term time range is set, such as the most recent hour. All alarm log entries related to the current rehabilitation exercise assessment within this time range are retrieved using the database's time index. The corresponding graded intervention strategies are extracted from these entries and arranged in chronological order of execution to form a time-series graded intervention strategy list.

[0088] Next, a unique identifier for the current real-time graded intervention strategy is determined, such as the strategy code. Then, by iterating through each graded intervention strategy in the time-series graded intervention strategy list, the identifier of each graded intervention strategy in the list is compared with the identifier of the real-time graded intervention strategy. Each time a matching identifier is found, a count is accumulated. After the iteration is completed, the accumulated count result is the recurrence frequency of the real-time graded intervention strategy.

[0089] Then, based on routine experience in clinical rehabilitation exercise risk intervention, the recurrence frequency threshold was preset to 3 times. The calculated recurrence frequency was compared with the preset frequency threshold using a numerical comparison method. If the recurrence frequency was less than or equal to the preset frequency threshold, it indicated that the current real-time graded intervention strategy could still effectively address the risk and no adjustment was needed. The intervention instruction corresponding to the real-time graded intervention strategy was directly retrieved from the graded intervention strategy library and output as an adaptive intervention instruction.

[0090] If the recurrence frequency exceeds the preset recurrence frequency threshold, a rule for setting the value of N is first preset. For example, if the recurrence frequency exceeds the threshold once, N=1; if it exceeds it twice, N=2; and at most, N=3. Simultaneously, a hierarchical system for intervention strategies is established, such as low-risk intervention → medium-risk intervention → high-risk intervention → emergency suspension intervention. After determining the specific value of N through numerical calculation, based on the current real-time hierarchical intervention strategy's position in the hierarchical system, it jumps up or down by N levels, retrieves the intervention instruction corresponding to the new level, and outputs it as an adaptive intervention instruction. For example, if the current strategy is low-risk and N=1, it jumps to a medium-risk intervention strategy and retrieves its instruction.

[0091] By using structured log retrieval, traversal, numerical comparison, and level transition, dynamic correction of real-time graded intervention strategies based on ultra-short-term alarm logs is achieved, outputting adaptive intervention instructions that are more in line with current risk changes, thereby improving the timeliness and accuracy of rehabilitation exercise risk intervention.

[0092] Furthermore, the method provided in this application embodiment includes:

[0093] After the adaptive intervention instructions are used to drive the correction of movement behavior, a closed-loop tracking intervention for movement rehabilitation risk is performed.

[0094] In one embodiment, real-time data after corrected movement behavior is first continuously collected. Wearable devices and physiological parameter monitoring devices are used to simultaneously collect the user's movement posture data and physiological parameter data, recording data at fixed time intervals of 1 second per record to form a continuous tracking data sequence. Next, the intervention effect is evaluated based on the tracking data sequence. Preset target parameter ranges for rehabilitation exercises, such as target intervals for exercise intensity and normal thresholds for physiological parameters, are used. A data comparison method is employed to compare the real-time collected movement posture data and physiological parameter data with the target parameter range one by one to determine whether the current data falls within the target range. If all data are within the target range, the intervention effect is deemed satisfactory; if any data exceeds the target range, the intervention effect is deemed unsatisfactory, and the abnormal data type is recorded.

[0095] If the intervention is effective, maintain the current adaptive intervention instructions and continue monitoring. Keep the current intervention parameters of the execution device, such as exercise intensity and movement correction prompts, and continuously collect subsequent data to ensure the exercise state remains stable within the target range and to avoid risk recurrence. If the intervention is ineffective, analyze the cause of the anomaly and adjust the intervention instructions. Through data backtracking, examine the exercise stage and physiological state corresponding to the abnormal data to determine the cause of the failure, such as excessively high exercise intensity or incomplete movement correction. Adjust the parameters of the adaptive intervention instructions based on the cause, such as further reducing exercise intensity or increasing the frequency of movement correction prompts, and generate new adaptive intervention instructions.

[0096] Finally, the adjusted adaptive intervention instructions are sent to the execution device to drive the user to correct their movement behavior again. Then, the process returns to the above steps of continuous data collection, forming a cycle of collection-evaluation-adjustment-re-collection, until the tracking data is stable within the target range for a long period of time, thus completing the closed-loop tracking intervention for sports rehabilitation risks.

[0097] In summary, the rehabilitation exercise risk assessment method provided in this application has the following technical effects:

[0098] This application integrates basic user information, static medical history, and dynamic real-time data to construct a body load baseline, collect movement posture and physiological parameters, and outputs a risk coefficient through three-level abnormality judgment, data backtracking, and Mahalanobis distance calculation. It then matches and corrects intervention strategies, drives closed-loop tracking after movement behavior correction, and achieves real-time assessment and precise intervention of rehabilitation exercise risks. This improves the reliability of assessment and the pertinence of intervention, achieving precise and comprehensive assessment, graded intervention, and closed-loop tracking of rehabilitation exercise risks. It effectively reduces the probability of sports injuries and ensures the safety and effectiveness of patient rehabilitation.

[0099] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a rehabilitation exercise risk assessment device, the device comprising:

[0100] Body load baseline construction module 1 is used to construct a body load baseline based on user basic information and static medical history data.

[0101] The motion posture data acquisition module 2 is used to collect the user's motion posture data through a wearable device. The motion posture data includes the real-time center of gravity projection trajectory and the real-time posture oscillation amplitude.

[0102] The physiological parameter data backtracking module 3 performs a three-level abnormal event determination based on the exercise posture data. After triggering exercise risk monitoring, it performs physiological parameter data backtracking, wherein the physiological parameter data includes time-series heart rate, time-series heart rate variability, time-series blood oxygen, and time-series blood pressure.

[0103] Real-time state vector construction module 4, which constructs a real-time state vector based on the physiological parameter data.

[0104] The real-time exercise risk coefficient acquisition module 5 is used to dynamically compare the deviation of the real-time state vector from the body load baseline and output the real-time exercise risk coefficient.

[0105] The real-time graded intervention strategy matching module 6 is used to match real-time graded intervention strategies based on the real-time motion risk coefficient.

[0106] The adaptive intervention instruction acquisition module 7 is used to backtrack the ultra-short-term alarm log to correct the real-time hierarchical intervention strategy and output adaptive intervention instructions.

[0107] An adaptive intervention instruction execution module 8 is used to drive the correction of movement behavior intervention using the adaptive intervention instruction.

[0108] Furthermore, the body load baseline construction module 1 is used to perform the following steps:

[0109] Exercise scenarios are segmented based on exercise physiological characteristics to obtain multiple standard exercise scenarios; multiple sets of dominant monitoring parameters for these standard exercise scenarios are defined; the static medical history data is quantified and transformed according to the multiple sets of dominant monitoring parameters to obtain multiple sets of scenario benchmark values; multiple associated load dynamic baselines are constructed based on the multiple sets of scenario benchmark values; the covariance of the retrieved historical training data is statistically analyzed using the multiple sets of scenario benchmark values ​​to construct multiple associated covariance matrices; the multiple standard exercise scenarios, multiple associated load dynamic baselines, and multiple associated covariance matrices are stored together to complete the configuration of the body load baseline.

[0110] Furthermore, the physiological parameter data backtracking module 3 is used to perform the following steps:

[0111] Based on the spatial vector time series of the real-time center of gravity projection trajectory, motion mode analysis is performed to output a first set of candidate motion scenarios; the frequency domain energy distribution characteristics of the real-time attitude oscillation amplitude are extracted for motion mode analysis to output a second set of candidate motion scenarios; the intersection of the first and second set of candidate motion scenarios is solved to locate the real-time motion scenario; using the real-time motion scenario as the criterion, a three-level anomaly event determination is performed on the motion attitude data to output a spatiotemporal coupling anomaly event; if the spatiotemporal coupling anomaly event is a non-empty set, motion risk monitoring is triggered.

[0112] Furthermore, the physiological parameter data backtracking module 3 is used to perform the following steps:

[0113] Based on the real-time motion scene, retrieve the spatial dimension anomaly judgment rules and the temporal dimension anomaly judgment rules; extract spatial dimension features and temporal dimension features from the motion posture data; use the spatial dimension anomaly judgment rules and the temporal dimension anomaly judgment rules to map and perform spatiotemporal anomaly detection of the spatial dimension features and temporal dimension features; if any level of spatiotemporal coupling detection fails, call the spatiotemporal mismatch type as the spatiotemporal coupling anomaly event output.

[0114] Furthermore, the physiological parameter data backtracking module 3 is used to perform the following steps:

[0115] Using the temporal hierarchy attribute of the spatiotemporal coupled abnormal event and the real-time motion scene as two-dimensional retrieval conditions, the retrospective time reference value is retrieved from the physiological retrospective rule mapping library; based on the coupling coefficient of the spatiotemporal coupled abnormal event, a window length correction coefficient is calculated; the retrospective time reference value is adjusted using the window length correction coefficient to obtain a dynamic retrospective time window; after triggering motion risk monitoring, the physiological parameter data is retrospectively retrieved using the dynamic retrospective time window as a retrospective constraint.

[0116] Furthermore, the real-time motion risk coefficient acquisition module 5 is used to perform the following steps:

[0117] Based on the real-time motion scenario, the adaptive load dynamic baseline and adaptive covariance matrix are invoked according to the body load baseline; real-time training data within the dynamic backtracking time window are retrieved to construct a real-time covariance matrix, and an adaptive covariance matrix is ​​generated by dynamically fusing the adaptive covariance matrix and the real-time covariance matrix; the adaptive covariance matrix is ​​used as a metric matrix to calculate the Mahalanobis distance of the real-time state vector relative to the adaptive load dynamic baseline; the Mahalanobis distance is used for risk mapping, and the real-time motion risk coefficient is output.

[0118] Furthermore, the adaptive intervention instruction acquisition module 7 is used to perform the following steps:

[0119] The time-series hierarchical intervention strategy is retrieved from the ultra-short-term alarm log; the time-series hierarchical intervention strategy is traversed to obtain the recurrence frequency of the real-time hierarchical intervention strategy; if the recurrence frequency does not exceed the preset recurrence frequency threshold, the intervention instruction of the real-time hierarchical intervention strategy is matched as the adaptive intervention instruction output; if the recurrence frequency exceeds the recurrence frequency threshold, the intervention instruction of the real-time hierarchical intervention strategy is jumped across N levels as the adaptive intervention instruction output.

[0120] Furthermore, the adaptive intervention instruction execution module 8 is used to perform the following steps:

[0121] After the adaptive intervention instructions are used to drive the correction of movement behavior, a closed-loop tracking intervention for movement rehabilitation risk is performed.

[0122] Furthermore, the physiological parameter data backtracking module 3 is used to perform the following steps:

[0123] The spatial dimension features are analyzed to obtain the frame-level spatiotemporal offset vector, periodic morphological distortion degree, and trend-level envelope expansion rate. The temporal dimension features are analyzed to obtain the frame-level instantaneous oscillation amplitude, periodic risk frequency band energy ratio, and trend-level main frequency migration rate. The frame-level scene safety zone, periodic trajectory morphology correlation threshold, and trend-level envelope area expansion rate threshold are progressively called from the spatial dimension anomaly judgment rules to map and perform spatial anomaly detection of the frame-level spatiotemporal offset vector, periodic morphological distortion degree, and trend-level envelope expansion rate. During spatial anomaly detection, the frame-level oscillation amplitude interval, periodic energy entropy change threshold, and trend-level main frequency migration rate threshold are synchronously and progressively called to map and perform spatiotemporal anomaly detection of the frame-level instantaneous oscillation amplitude, periodic risk frequency band energy ratio, and trend-level main frequency migration rate.

[0124] The rehabilitation exercise risk assessment device provided in this embodiment of the invention can execute a rehabilitation exercise risk assessment method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0125] Although this application makes various references to certain modules in the apparatus according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not intended to limit the scope of protection of this invention.

[0126] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for assessing the risk of rehabilitation exercise, characterized in that, The method includes: A baseline of physical burden is constructed based on user basic information and static medical history data; The user's motion posture data is collected through wearable devices, wherein the motion posture data includes the real-time center of gravity position projection trajectory and the real-time posture oscillation amplitude. Based on the aforementioned motion posture data, a three-level abnormal event determination is performed. After triggering motion risk monitoring, physiological parameter data is backtracked, including time-series heart rate, time-series heart rate variability, time-series blood oxygen, and time-series blood pressure. A real-time state vector is constructed based on the physiological parameter data; The real-time state vector is used to dynamically deviate from the body load baseline and output a real-time motion risk coefficient. Based on the real-time motion risk coefficient, a real-time graded intervention strategy is matched. The real-time hierarchical intervention strategy is corrected by reviewing the ultra-short-term alarm logs, and adaptive intervention instructions are output. The adaptive intervention command is used to drive the correction of movement behavior intervention.

2. The rehabilitation exercise risk assessment method as described in claim 1, characterized in that, Based on user basic information and static medical history data, a physical burden baseline is constructed. The method includes: Based on the characteristics of exercise physiology, various standard exercise scenarios are divided. Define multiple sets of dominant monitoring parameters for the various standard motion scenarios; Based on the multiple sets of dominant monitoring parameters, the static medical history data is subjected to targeted quantitative transformation to obtain multiple sets of scenario benchmark values; Multiple dynamic baselines of associated loads are constructed based on the aforementioned sets of scenario benchmark values; The covariance of the retrieved historical training data is statistically analyzed using the multiple sets of scene benchmark values ​​to construct multiple correlation covariance matrices. The system associates and stores the various standard exercise scenarios, multiple associated load dynamic baselines, and multiple associated covariance matrices to complete the configuration of the body load baseline.

3. The rehabilitation exercise risk assessment method as described in claim 2, characterized in that, Based on the aforementioned motion posture data, a three-level abnormal event determination is performed. After triggering motion risk monitoring, physiological parameter data is retrospectively analyzed. The method includes: Based on the spatial vector time sequence of the real-time center of gravity position projection trajectory, motion mode analysis is performed, and a first set of candidate motion scenarios is output. Extract the frequency domain energy distribution characteristics of the real-time attitude oscillation amplitude for motion pattern analysis, and output a second set of alternative motion scenarios; The intersection of the first set of candidate motion scenes and the second set of candidate motion scenes is calculated to locate the real-time motion scene; Using the real-time motion scene as the criterion, the motion posture data is subjected to a three-level abnormal event judgment, and spatiotemporal coupling abnormal events are output. If the spatiotemporal coupling anomaly event is a non-empty set, then motion risk monitoring is triggered.

4. The rehabilitation exercise risk assessment method as described in claim 3, characterized in that, Using the real-time motion scene as the criterion, the motion posture data is subjected to a three-level anomaly event determination, and spatiotemporal coupled anomaly events are output. The method includes: Based on the real-time motion scene, retrieve the spatial dimension anomaly judgment rules and the temporal dimension anomaly judgment rules; Spatial and temporal features are extracted from the motion posture data; The spatial dimension anomaly determination rule and the temporal dimension anomaly determination rule are used to map the spatial dimension features and temporal dimension features to perform spatiotemporal anomaly detection; If any level of spatiotemporal coupling detection fails, the spatiotemporal mismatch type is invoked as the spatiotemporal coupling exception event output.

5. The rehabilitation exercise risk assessment method as described in claim 4, characterized in that, The method further includes: Using the temporal hierarchical attributes of the spatiotemporal coupled abnormal events and the real-time motion scene as two-dimensional retrieval conditions, the retrospective time reference value is retrieved from the physiological retrospective rule mapping library. Calculate the window length correction coefficient based on the coupling coefficient of the spatiotemporal coupling anomaly event; The backtracking time base value is adjusted using the window length correction coefficient to obtain a dynamic backtracking time window; After triggering exercise risk monitoring, the physiological parameter data is retrieved back using the dynamic backtracking time window as a backtracking constraint.

6. The rehabilitation exercise risk assessment method as described in claim 5, characterized in that, The method involves dynamically comparing the real-time state vector with the body load baseline to output a real-time motion risk coefficient. Based on the real-time motion scenario, the adaptive load dynamic baseline and adaptive covariance matrix are invoked in the body load baseline matching. Real-time training data within the dynamic backtracking time window is retrieved to construct a real-time covariance matrix, and an adaptive covariance matrix is ​​generated by dynamically fusing the adaptive covariance matrix and the real-time covariance matrix. Using the adaptive covariance matrix as a metric matrix, calculate the Mahalanobis distance of the real-time state vector relative to the adaptive load dynamic baseline; The Mahalanobis distance is used for risk mapping, and the real-time motion risk coefficient is output.

7. The rehabilitation exercise risk assessment method as described in claim 1, characterized in that, The method involves retrospectively analyzing ultra-short-term alarm logs to correct the real-time graded intervention strategy and outputting adaptive intervention instructions. Retrieve time-series hierarchical intervention strategies from the ultra-short-term alarm logs; The time-series hierarchical intervention strategies are traversed to obtain the recurrence frequency of the real-time hierarchical intervention strategies; If the recurrence frequency does not exceed the preset recurrence frequency threshold, then the intervention instruction of the real-time hierarchical intervention strategy is matched as the adaptive intervention instruction output. If the recurrence frequency exceeds the recurrence frequency threshold, the intervention instruction of the real-time hierarchical intervention strategy for the N-level transition is output as the adaptive intervention instruction.

8. The rehabilitation exercise risk assessment method as described in claim 1, characterized in that, After the adaptive intervention instructions are used to drive the correction of movement behavior, a closed-loop tracking intervention for movement rehabilitation risk is performed.

9. The rehabilitation exercise risk assessment method as described in claim 4, characterized in that, The method for detecting spatiotemporal anomalies in spatial and temporal features by mapping the spatial and temporal anomaly determination rules includes: By analyzing the spatial dimension features, we can obtain the frame-level spatiotemporal offset vector, the periodic-level morphological distortion degree, and the trend-level envelope expansion rate. By analyzing the time-dimensional features, we can obtain the instantaneous value of frame-level oscillation amplitude, the energy proportion of periodic risk frequency bands, and the trend-level main frequency migration rate. The spatial dimension anomaly determination rules progressively call the frame-level scene safe zone, the periodic trajectory morphology correlation threshold, and the trend-level envelope area expansion rate threshold, and map them to perform spatial anomaly detection of the frame-level spatiotemporal offset vector, the periodic morphological distortion degree, and the trend-level envelope expansion rate. During spatial anomaly detection, the frame-level oscillation amplitude range, periodic-level energy entropy change threshold, and trend-level main frequency migration rate threshold are synchronously and progressively invoked to perform spatiotemporal anomaly detection of the instantaneous value of the frame-level oscillation amplitude, the energy proportion of the periodic-level risk frequency band, and the trend-level main frequency migration rate.

10. A rehabilitation exercise risk assessment device, characterized in that, The apparatus for implementing the rehabilitation exercise risk assessment method according to any one of claims 1-9, the apparatus comprising: The physical workload baseline construction module is used to construct a physical workload baseline based on the user's basic information and static medical history data; The motion posture data acquisition module is used to collect the user's motion posture data through a wearable device, wherein the motion posture data includes the real-time center of gravity position projection trajectory and the real-time posture oscillation amplitude. The physiological parameter data backtracking module performs a three-level abnormal event determination based on the exercise posture data. After triggering exercise risk monitoring, it performs physiological parameter data backtracking, wherein the physiological parameter data includes time-series heart rate, time-series heart rate variability, time-series blood oxygen, and time-series blood pressure. A real-time state vector construction module constructs a real-time state vector based on the physiological parameter data; The real-time exercise risk coefficient acquisition module is used to dynamically compare the deviation of the real-time state vector from the body load baseline and output the real-time exercise risk coefficient. A real-time graded intervention strategy matching module is used to match real-time graded intervention strategies based on the real-time motion risk coefficient. The adaptive intervention instruction acquisition module is used to retrospectively analyze ultra-short-term alarm logs to correct the real-time hierarchical intervention strategy and output adaptive intervention instructions. An adaptive intervention instruction execution module is used to drive the correction of movement behavior intervention using the adaptive intervention instructions.

Images