A fall prevention prediction method and system for a rehabilitation training scene
By using multimodal data fusion and deep temporal coding networks, a fall risk prediction system was constructed, which solved the problem of insufficient accuracy in fall risk monitoring during rehabilitation training in existing technologies. This system enables individualized fall risk warnings and improves the safety and reliability of rehabilitation training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANCHANG UNIV
- Filing Date
- 2026-03-11
- Publication Date
- 2026-06-12
AI Technical Summary
Existing rehabilitation training techniques lack systematic fall risk monitoring programs, making it difficult to provide effective early warnings before falls occur. Moreover, most programs rely on single sensors or single-dimensional motion data, resulting in insufficient accuracy and adaptability of risk assessments. This is especially true in rehabilitation populations with significant individual differences, where false alarms and missed alarms are prone to occur.
By acquiring multimodal motion state data, including multi-source sensor motion state data and visual motion state data, preprocessing and feature extraction and fusion are performed to construct stability features and risk state vectors. A deep temporal coding network and preset weight coefficients are used to predict fall risk and output the fall risk judgment result.
It enables accurate identification and early warning of risk conditions before a fall occurs, improving the safety and reliability of the rehabilitation training process, adapting to the differences in movement among different individuals, and reducing the occurrence of false alarms and missed alarms.
Smart Images
Figure CN122201618A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent rehabilitation and sports safety analysis, specifically to a fall prevention prediction method and system for rehabilitation training scenarios. Background Technology
[0002] Rehabilitation training is widely used in medical settings such as postoperative rehabilitation, nerve injury rehabilitation, and functional recovery in the elderly. Because rehabilitation patients commonly experience muscle weakness, decreased balance, or weakened motor control, their movement stability is poor and the risk of falls is high during gait training, lower limb function training, and balance training.
[0003] Current research and application of rehabilitation training technologies mainly focus on movement recognition, training process assessment, and training effect feedback, but a systematic fall risk monitoring program for the rehabilitation period has not yet been formed. The few existing methods involving fall detection are mostly limited to post-event alarms or simple threshold-based judgments, lacking the ability to analyze the dynamic changes in human movement states, thus making it difficult to provide effective early warnings before falls occur. Furthermore, most solutions rely on single sensors or single-dimensional motion data, which cannot fully characterize the dynamic stability characteristics of the human body during rehabilitation, resulting in insufficient accuracy and adaptability in risk assessment. Especially in rehabilitation populations with significant individual differences, such methods are prone to false alarms and false negatives, limiting their reliable application in real-world scenarios. Summary of the Invention
[0004] This application aims to provide a fall prevention prediction method and system for rehabilitation training scenarios, which can accurately predict fall risks and thus improve the safety of the rehabilitation training process.
[0005] The technical solution of this application is implemented as follows: In a first aspect, embodiments of this application provide a fall prevention prediction method for rehabilitation training scenarios, the method comprising: Acquire multimodal motion state data during rehabilitation training; wherein, the multimodal motion state data includes multi-source sensor motion state data and visual motion state data; The multi-source sensor motion state data and visual motion state data are preprocessed to obtain a multimodal motion sequence; Based on the multimodal motion sequence, encoding is performed to obtain individual stable feature centers and stability features; and based on the individual stable feature centers and stability features, stability confidence and dynamic stability margin data are determined. Based on the stability confidence level and the dynamic stability margin data, a risk state vector is constructed; and based on the risk state vector and preset weight coefficients, the risk state of human fall is predicted, and the fall risk judgment result is determined.
[0006] In the above scheme, the preprocessing of the multi-source sensor motion state data and visual motion state data to obtain a multimodal motion sequence includes: The multi-source sensor motion state data and visual motion state data are aligned and resampled to obtain aligned multi-source sensor motion state data and aligned visual motion state data. Feature extraction and feature fusion are performed on the aligned multi-source sensor motion state data and the aligned visual motion state data to obtain a comprehensive motion state vector; Based on a predetermined individual scale factor, the comprehensive motion state vector is normalized to obtain the current normalized state vector. Anomaly analysis is performed based on the normalized state vector and the pre-acquired historical moment state vector to determine a reliable state; and based on the reliable state, the multimodal motion sequence is determined.
[0007] In the above scheme, the encoding based on the multimodal motion sequence to obtain individual stable feature centers and stability features includes: Multiple different temporal enhancement operators are applied to the multimodal motion sequence to generate multiple view sequences; The multiple view sequences are encoded using a deep temporal coding network to obtain latent features; Based on the latent features, the individual stable feature center is determined; The stability features are obtained by encoding the multimodal motion sequence using a pre-determined feature encoding model.
[0008] In the above scheme, determining the stability confidence and dynamic stability margin data based on the individual stable feature center and the stability feature includes: Stability is calculated based on the individual stable feature center and the stability feature to determine the stability confidence level; Based on the individual stability feature center, the stability feature, and the preset stability domain radius, the dynamic stability margin is determined; Based on the dynamic stability margin, calculate the rate of change of stability margin; The dynamic stability margin and the rate of change of the stability margin are determined as the dynamic stability margin data.
[0009] In the above scheme, constructing a risk state vector based on the stability confidence level and the dynamic stability margin data includes: The dynamic stability margin in the dynamic stability margin data is processed by a normalization function to determine the instantaneous stability margin risk vector. The stability margin change rate in the dynamic stability margin data is processed by a normalization function to obtain a stability trend change risk vector. Based on the stability confidence level, the individual steady-state pattern deviation risk vector is determined; The risk state vector is determined based on the instantaneous stability margin risk vector, the stable trend change risk vector, and the individual steady-state mode deviation risk vector.
[0010] In the above scheme, the step of predicting the human fall risk state based on the risk state vector and preset weight coefficients, and determining the fall risk assessment result, includes: The risk potential function value is determined based on the risk state vector and the preset weight coefficients; The risk state of a human body falling is predicted by using the risk potential function value and a preset time weight, and the result of the fall risk assessment is determined.
[0011] In the above scheme, the step of predicting the human fall risk state and determining the fall risk assessment result by using the risk potential function value and a preset time weight includes: By using the risk potential function value and the preset time weight, the risk state of human fall is predicted, and the fall risk function value is obtained. The fall risk probability is obtained by mapping the fall risk function value to the preset risk function value corresponding to the preset fall risk threshold; If the probability of falling is greater than the preset risk threshold, then the risk assessment result is determined to be that there is a risk of falling. If the probability of falling is less than or equal to the preset fall risk threshold, then the fall risk judgment result is determined to be that there is no fall risk.
[0012] Secondly, embodiments of this application provide a fall prevention prediction system for rehabilitation training scenarios. This fall prevention prediction system for rehabilitation training scenarios includes: an acquisition module, a preprocessing module, a determination module, and a prediction module, wherein... The acquisition module is used to acquire multimodal motion state data during rehabilitation training; wherein, the multimodal motion state data includes multi-source sensor motion state data and visual motion state data; The preprocessing module is used to preprocess the multi-source sensor motion state data and visual motion state data to obtain a multimodal motion sequence; The determining module is used to encode the multimodal motion sequence to obtain individual stable feature centers and stability features; and to determine stability confidence and dynamic stability margin data based on the individual stable feature centers and stability features. The prediction module is used to construct a risk state vector based on the stability confidence level and the dynamic stability margin data; and to predict the human fall risk state based on the risk state vector and preset weight coefficients, and determine the fall risk judgment result.
[0013] Thirdly, embodiments of this application provide a fall prevention prediction device for rehabilitation training scenarios, comprising: a processor and a memory; wherein, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in the first aspect.
[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing executable instructions for causing a processor to perform the method described in the first aspect.
[0015] This application provides a method and system for fall prevention prediction in rehabilitation training scenarios. The method includes: acquiring multimodal motion state data during rehabilitation training; wherein the multimodal motion state data includes multi-source sensor motion state data and visual motion state data; preprocessing the multi-source sensor motion state data and visual motion state data to obtain a multimodal motion sequence; encoding the multimodal motion sequence to obtain an individual stability feature center and stability features; determining stability confidence and dynamic stability margin data based on the individual stability feature center and the stability features; constructing a risk state vector based on the stability confidence and the dynamic stability margin data; and predicting the human fall risk state based on the risk state vector and preset weight coefficients to determine the fall risk judgment result. In the above scheme, multi-source motion state data is preprocessed to obtain multimodal motion sequences; based on the multimodal motion sequences, encoding is performed to obtain individual stability feature centers and stability features, and individualized stability feature parameters that can reflect the differences in motion among different individuals are extracted; based on the individual stability feature centers and stability features, stability confidence and dynamic stability margin data are determined; based on the stability confidence and dynamic stability margin data, a risk state vector is constructed to quantitatively assess the stability changes during human movement; then, based on the risk state vector and preset weight coefficients, the fall risk state of the human body is predicted and judged, and the fall risk judgment result is output; this enables early identification and warning of risk states before a fall occurs, and accurate fall risk prediction can be performed, thereby improving the safety of the rehabilitation training process. Attached Figure Description
[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application. Obviously, the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.
[0017] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0018] Figure 1 An optional flowchart illustrating a fall prevention prediction method for rehabilitation training scenarios provided in this application embodiment. Figure 1 ; Figure 2 An optional flowchart illustrating a fall prevention prediction method for rehabilitation training scenarios provided in this application embodiment. Figure 2 ; Figure 3 An optional flowchart illustrating a fall prevention prediction method for rehabilitation training scenarios provided in this application embodiment. Figure 3 ; Figure 4 A schematic diagram of a fall prevention prediction system for rehabilitation training scenarios provided in this application embodiment; Figure 5 This is a schematic diagram of the structure of a fall prevention prediction device for a rehabilitation training scenario, provided in an embodiment of this application. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of this application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.
[0020] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this application is for the purpose of describing embodiments of this application only and is not intended to be limiting of this application.
[0021] In the following description, references to "some embodiments," "this embodiment," "this application embodiment," and examples, etc., describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subset of all possible embodiments and may be combined with each other without conflict.
[0022] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.
[0023] This application provides a fall prevention prediction method for rehabilitation training scenarios. Figure 1 An optional flowchart illustrating a fall prevention prediction method for rehabilitation training scenarios provided in this application embodiment. Figure 1 , will combine Figure 1 The steps shown are explained.
[0024] S101. Acquire multimodal motion state data during rehabilitation training; wherein, multimodal motion state data includes multi-source sensor motion state data and visual motion state data.
[0025] In some embodiments of this application, the multimodal motion state data includes multi-source sensor motion state data and visual motion state data.
[0026] In some embodiments of this application, a fall prevention prediction method for rehabilitation training scenarios is adapted to rehabilitation training scenarios.
[0027] In some embodiments of this application, a fall prevention prediction method for rehabilitation training scenarios is adapted to a fall prevention prediction system for rehabilitation training scenarios.
[0028] In some embodiments of this application, multi-source motion state data is acquired through multiple sensors, and visual motion state data is acquired through an image acquisition device.
[0029] S102. Preprocess the multi-source sensor motion state data and visual motion state data to obtain a multimodal motion sequence.
[0030] In some embodiments of this application, multi-source sensor motion state data and visual motion state data are aligned and resampled to obtain aligned multi-source sensor motion state data. The system integrates aligned multi-source sensor motion state data and aligned visual motion state data; it performs feature extraction and feature fusion on the aligned multi-source sensor motion state data and aligned visual motion state data to obtain a comprehensive motion state vector; it performs scale normalization processing on the comprehensive motion state vector based on a pre-determined individual scale factor to obtain the current normalized state vector; it performs anomaly analysis based on the normalized state vector and the pre-acquired historical moment state vectors to determine the credible state; and it determines the multimodal motion sequence based on the credible state.
[0031] S103. Based on the multimodal motion sequence, encode to obtain the individual stable feature center and stability feature; and based on the individual stable feature center and stability feature, determine the stability confidence and dynamic stability margin data.
[0032] In some embodiments of this application, multiple different temporal enhancement operators are applied to a multimodal motion sequence to generate multiple view sequences; these view sequences are encoded using a deep temporal coding network to obtain latent features; based on the latent features, individual stable feature centers are determined; the multimodal motion sequence is encoded using a pre-determined feature coding model to obtain stability features. Stability is calculated based on the individual stable feature centers and stability features to determine the stability confidence; the dynamic stability margin is determined based on the individual stable feature centers, stability features, and a preset stability domain radius; the stability margin change rate is calculated based on the dynamic stability margin; and the dynamic stability margin and the stability margin change rate are defined as dynamic stability margin data.
[0033] S104. Based on stability confidence and dynamic stability margin data, construct a risk state vector; and based on the risk state vector and preset weight coefficients, predict the risk state of human falls and determine the fall risk assessment result.
[0034] In some embodiments of this application, the dynamic stability margin in the dynamic stability margin data is processed using a normalization function to determine the instantaneous stability margin risk vector; the rate of change of stability margin in the dynamic stability margin data is processed using a normalization function to obtain the stable trend change risk vector; based on the stability confidence level, the individual steady-state pattern deviation risk vector is determined; based on the instantaneous stability margin risk vector, the stable trend change risk vector, and the individual steady-state pattern deviation risk vector, the risk state vector is determined. Based on the risk state vector and preset weight coefficients, the risk potential function value is determined; using the risk potential function value and preset time weights, the human fall risk state is predicted, and the fall risk judgment result is determined.
[0035] For example, such as Figure 2 As shown, a fall prevention prediction method for rehabilitation training scenarios can also be implemented through the following steps: S1. Obtain historical multi-source motion state data during rehabilitation training.
[0036] S2. Construct a unified multimodal temporal motion state representation.
[0037] S3. Modeling of stable individual characteristics based on historical data.
[0038] S4. Dynamic stability margin analysis modeling.
[0039] S5. Predict and judge the risk of human fall within a preset time window.
[0040] S6. Output the judgment result.
[0041] It is understood that the embodiments of this application preprocess multi-source motion state data to obtain multimodal motion sequences; based on the multimodal motion sequences, encoding is performed to obtain individual stability feature centers and stability features, and individualized stability feature parameters that can reflect the differences in motion among different individuals are extracted; based on the individual stability feature centers and stability features, stability confidence and dynamic stability margin data are determined; based on the stability confidence and dynamic stability margin data, a risk state vector is constructed to quantitatively assess the stability changes during human movement; then, based on the risk state vector and preset weight coefficients, the fall risk state of the human body is predicted and judged, and the fall risk judgment result is output; thus, the risk state can be identified and warned in advance before a fall occurs, and the fall risk can be accurately predicted, thereby improving the safety of the rehabilitation training process.
[0042] In some embodiments of this application, S102 can be implemented by S201-S204, as follows: S201. Align and resample the multi-source sensor motion state data and visual motion state data to obtain aligned multi-source sensor motion state data and aligned visual motion state data.
[0043] S202. Perform feature extraction and feature fusion on the aligned multi-source sensor motion state data and the aligned visual motion state data to obtain a comprehensive motion state vector.
[0044] S203. Based on the predetermined individual scale factor, the comprehensive motion state vector is scaled and normalized to obtain the current normalized state vector.
[0045] S204. Based on the normalized state vector and the pre-acquired historical moment state vector, perform anomaly analysis to determine the credible state; and based on the credible state, determine the multimodal motion sequence.
[0046] For example, during rehabilitation training, multimodal information such as multi-source sensor motion state data and visual motion state data are fused along a continuous time axis to construct a unified, continuous, and robust representation of human motion state. Assume the system contains a total of... The modality, the first The original observation data of each mode at discrete time are ; m represents the mode number, with a value range from 1 to M.
[0047] Due to differences in sampling rates, communication delays, and timestamp drift across different modalities, a unified time axis is first established { Based on}, the modal data are aligned and resampled to obtain aligned observation data: in, This represents the alignment / interpolation operator for the m-th modality, such as nearest neighbor, linear interpolation, or a timestamp-based synchronization strategy. After alignment, a feature vector of uniform dimension is extracted for each modality. in, This is the feature extraction function.
[0048] Subsequently, the modal features at the same time are fused to construct a comprehensive motion state vector of the human body at time t: in, This represents the feature fusion function, implemented here using feature concatenation: Furthermore, considering the scale differences among individuals in body size, stride length, and range of motion (ROM), a scale normalization is performed on the overall status to reduce the interference of individual scale on subsequent stability / assessment indicators. Let the individual scale factor be... The current normalized state vector is calculated as follows: It can be obtained from height, limb length, stride statistics, or individual baseline motion calibration.
[0049] Due to sensor noise, occlusion, sudden changes in action, packet loss, or delays, state sequences may contain missing values, short-term distortions, or transient outliers. To ensure the continuity and reliability of the timing representation, the module introduces a robust mechanism of "anomaly detection + rollback substitution".
[0050] Based on the assumption of time continuity, the predicted value for the current time is obtained by using the credible state of the previous time step or a historical window estimate. : in, Take the moving average.
[0051] Anomaly detection is performed on the deviation between the current normalized state and the predicted value: when > If a modality missing marker exists, that moment is determined to be abnormal / missing; where, is The threshold is adaptively set based on historical statistics.
[0052] Define the credible state that will ultimately be used for subsequent analysis. : Finally, a length of [length] was adopted. The sliding time window will continuously The reliable state vectors at each time step are stacked to form a multimodal motion sequence: The output X(t) contains both multimodal information and temporal evolution structure, and can be directly used as input for subsequent stability analysis. For example... Figure 3 As shown, the construction of multimodal motion sequences includes the following steps: S11. Input multi-source motion state data of the human body during rehabilitation training.
[0053] S12. Perform time alignment and feature extraction on multi-source motion state data to obtain multiple features.
[0054] S13. Perform multimodal fusion on multiple features to obtain a comprehensive motion state vector.
[0055] S14. Perform individual-scale normalization on the comprehensive motion state vector to obtain the current normalized state vector.
[0056] S15. Perform anomaly detection and backoff mechanism on the current normalized state vector to obtain a reliable state.
[0057] S16. Output the matrix of the reliable state through a sliding window to obtain the multimodal motion sequence.
[0058] In some embodiments of this application, encoding based on multimodal motion sequences in S103 to obtain individual stable feature centers and stability features can be achieved through S301-S304, as follows: S301. Apply multiple different temporal enhancement operators to the multimodal motion sequence to generate multiple view sequences.
[0059] S302. Encode multiple view sequences using a deep temporal coding network to obtain latent features.
[0060] S303. Based on latent features, determine the center of stable features for an individual.
[0061] S304. Encode the multimodal motion sequence using a pre-determined feature coding model to obtain stability features.
[0062] For example, this study explores the intrinsic consistency features of stable motion states under different time scales and perturbation conditions by employing a self-supervised feature representation learning strategy based on temporal contrastive learning. This is achieved through the analysis of multimodal motion sequences. Applying different temporal enhancement operators Generate multiple view sequences that maintain stable semantics (i.e., multiple view sequences): Temporal augmentation operators include, but are not limited to, temporal clipping, temporal warping, amplitude perturbation, and multimodal channel occlusion, used to simulate real-world training scenarios such as velocity variations, amplitude fluctuations, and local perception loss.
[0063] A depth-time coding network that inputs multiple view sequences with shared parameters Obtain the corresponding latent feature representation: , By introducing consistency constraints based on stability semantics, the feature representations of different views generated from multimodal motion sequences are brought closer together, while the feature representations of different samples (i.e., multimodal motion sequences from different time periods) are distinguished. Its contrastive learning objective function is defined as the multi-view temporal contrastive loss: in, This represents the set of positive sample view pairs. Represents the view with anchor points The corresponding set of negative samples, sim(·) is the cosine similarity function, This is the temperature coefficient.
[0064] Through the above training process, the encoding network is constrained to learn a latent feature space that is insensitive to changes in action form, speed, and noise disturbances, but highly consistent with stable semantics.
[0065] After completing the stability representation learning, this application further constructs an individualized stability domain model within the latent feature space. To this end, a deep first-class stability domain constraint concept is introduced, imposing compactness constraints on the latent feature representations of stable samples to ensure their distribution within a minimal enclosing region centered on the stable feature center. Let the individual stable feature center be: in, This indicates the number of stable training samples.
[0066] By minimizing the squared distance from the latent features of stable samples to the center, and jointly optimizing the parameters of the feature extraction network, a deep first-class stability region constraint is formed, the objective function of which can be expressed as: During the training phase, the final feature encoding model is obtained by jointly optimizing the comparative learning loss and the stability region constraint loss: in, These are the weighting coefficients used to balance the characterization of discriminability and the compactness of the stable region.
[0067] During the online analysis phase, an encoding network that shares parameters with the training phase is used to analyze the multimodal motion state sequence within the current time window. Encode it to obtain its potential stability characteristics: Calculate the degree of deviation of this feature from the center of the individual's stable domain: Furthermore, to enhance the probabilistic interpretability of the deviation metric, this application constructs an individualized stability scoring function based on the deviation distribution of stable samples, mapping the deviation degree to the stability confidence degree: in, The scaling parameter is estimated from historical stable samples.
[0068] Furthermore, a continuity constraint is imposed on the instantaneous stability confidence score in the time dimension. By weighted averaging or integrating the scores within a preset time window, a smooth stability feature reflecting the evolutionary trend of individual motion stability is obtained. Smooth stability features can be used for online rehabilitation training stability assessment, abnormal state early warning, and individual training progress analysis.
[0069] In some embodiments of this application, determining the stability confidence and dynamic stability margin data based on the individual stability feature center and stability features in S103 can be achieved through S401-S404, as follows: S401. Calculate stability based on individual stable feature centers and stability features, and determine the stability confidence level.
[0070] S402. Determine the dynamic stability margin based on the individual stability feature center, stability features, and preset stability domain radius.
[0071] S403. Calculate the rate of change of stability margin based on dynamic stability margin.
[0072] S404. The dynamic stability margin and the rate of change of stability margin are defined as dynamic stability margin data.
[0073] For example, to further quantify the stability of human motion states and their dynamic evolution trends, an individualized stability domain and a corresponding dynamic stability margin model are constructed within the obtained potential stability feature space.
[0074] Based on the distribution of historical stable training samples in the latent feature space, a set of individualized stable domains is constructed: in, The individual stable feature centers obtained in the second part, The radius of the stability region is determined by the maximum deviation or statistical quantile of the stable samples. (Set) It represents the boundary of the stability region and is used to characterize the critical region where the human body's motion state transitions from stability to potential instability.
[0075] During the online analysis phase, for the motion state sequence within the current time window, the encoding network outputs its potential stability representation. Define dynamic stability margin. This represents the safety margin of the current state relative to the stability boundary: when When the value is greater than 0, it indicates that the current motion state is within the stable region, and the larger the value, the higher the stability margin; when → When, it indicates that the motion state is gradually approaching the stable boundary; when < When this occurs, it indicates that the motion state has crossed the stability boundary and entered a potentially unstable or high-risk area.
[0076] Considering the dynamic characteristics of human motion, relying solely on instantaneous stability margin is insufficient to reflect the trend of stability changes. Therefore, a model is further developed to examine the temporal variation characteristics of the stability margin. The rate of change of the stability margin is defined as: when When the absolute value of the value is less than 0 and continues to increase, it indicates that the human body's motion state is approaching the stability boundary at a relatively fast speed, and the stability is showing a downward trend; when When the value is greater than 0, it indicates that the motion state has returned to the stable region.
[0077] To enhance the ability to distinguish between short-term fluctuations and persistent trends, within a preset time window... Statistical analysis is performed on the stability margin series to calculate the minimum stability margin within the window: By combining the mean, variance, and rate of change within the window, a dynamic stability margin feature vector is constructed that reflects the degree of change in human motion stability, which is used to describe the trend of stability evolution.
[0078] Furthermore, to characterize the potential risk of the human motion state reaching the stability boundary in the near future, a time margin index based on the changing trend is introduced to estimate the time required for the current motion state to reach the stability boundary: Where ε is a very small positive number to prevent the denominator from being zero. When When the value continuously decreases or falls below a preset threshold, it is determined that the human body's movement state has a high risk of instability in a short period of time, which can trigger corresponding early warning or intervention mechanisms.
[0079] In some embodiments of this application, the construction of the risk state vector based on stability confidence and dynamic stability margin data in S104 can be implemented through S501-S504, as follows: S501. The dynamic stability margin in the dynamic stability margin data is processed by a normalization function to determine the instantaneous stability margin risk vector.
[0080] S502. The rate of change of stability margin in the dynamic stability margin data is processed by a normalization function to obtain the risk vector of stable trend change.
[0081] S503. Based on stability confidence, determine the risk vector of deviation from the individual steady-state pattern.
[0082] S504. Determine the risk state vector based on the immediate stability margin risk vector, the stable trend change risk vector, and the individual steady-state mode deviation risk vector.
[0083] For example, at any given time t, a dynamic stability margin characterizing the stability of the human body's current motion state can be obtained. and the rate of change of stability margin, which reflects the trend of stability change. or time margin Meanwhile, based on the individualized stability feature modeling method, an individualized stability score can be obtained, representing the degree of deviation of the current motion state from the individual's stable motion pattern. Considering that fall risk is usually not triggered by a single factor, but rather by the combined effects of factors such as a continuous decline in stability, an intensified trend of instability, and abnormal individual motor behavior, this application maps the aforementioned stability-related quantities to a risk assessment space to construct a joint risk state vector: Immediate stability margin risk vector: in, Represents the normalization function. This represents the safety margin risk of the current motion state within the stable region, used to characterize how close the human body is to the instability boundary at the current moment. As the motion state gradually approaches or crosses the stable boundary, this quantity increases monotonically, reflecting the increase in immediate instability risk.
[0084] Stable trend change risk vector: in, Represents the normalization function. This represents the risk of changing stability trends and is used to describe the speed at which a human body's motion state approaches the instability boundary. This quantity increases significantly when the stability margin decreases rapidly over time, and can reflect potential risks even if the instability boundary has not yet been reached.
[0085] Individual steady-state pattern deviates from risk vector: =1- in, This measure represents the risk of deviation from an individual's stable movement pattern, used to characterize the degree of abnormality of current movement behavior relative to an individual's historical stable movement patterns. It reflects the risk of instability at the behavioral pattern level and can capture potential fall risks caused by fatigue, compensatory movements, or abnormal postures.
[0086] In some embodiments of this application, the prediction of human fall risk status based on the risk state vector and preset weight coefficients in S104, and the determination of the fall risk judgment result, can be achieved through S601 and S602, as follows: S601. Determine the risk potential function value based on the risk state vector and preset weight coefficients.
[0087] S602. Using the risk potential function value and preset time weight, predict the risk status of human falls and determine the fall risk assessment result.
[0088] In some embodiments of this application, the risk state of a human body falling is predicted by using a risk potential function value and a preset time weight to obtain a fall risk function value; the fall risk function value is mapped to a preset risk function value corresponding to a preset fall risk threshold to obtain a fall risk probability; if the fall risk probability is greater than the preset fall risk threshold, the fall risk judgment result is determined to be that there is a fall risk; if the fall risk probability is less than or equal to the preset fall risk threshold, the fall risk judgment result is determined to be that there is no fall risk.
[0089] For example, to characterize the overall instability of the current motion state, a risk potential function based on the energy function form is introduced. A unified model is then constructed for the aforementioned risk state vector, and its expression is as follows: in, , , For preset weighting coefficients, · , · , · It is a monotonically increasing mapping function used to convert different stability characteristics into a uniform risk contribution.
[0090] In a preferred embodiment, the mapping function is in exponential or logarithmic form to enhance sensitivity to changes in the critical steady state, for example: By modeling with the risk potential function, when the stability margin continues to decrease, the rate of stability decline accelerates, or the individualized stability score decreases significantly, the corresponding risk potential value ε(t) will increase significantly, thus reflecting the process of the human body's motion state gradually approaching the unstable region.
[0091] Furthermore, considering that the risk of falling has a significant time-cumulative characteristic, this application integrates or weights the risk potential function over time, thereby increasing the risk potential function... With time weighting function · Perform convolution to obtain continuous fall risk function values: in, To preset the length of the risk assessment time window, · This is a time-weighting function used to assign higher weights to risk states that are closer to the current time.
[0092] Using the above method, the fall risk function value is... It can simultaneously reflect the current level of human stability, the trend of stability changes, and the degree of deviation of individual behavioral patterns.
[0093] To enhance the probabilistic interpretability of the risk assessment results, this application further maps the fall risk function value to a fall risk probability output, in the form of: When the fall risk function value exceeds the preset risk function value At that time, it is determined that the human body has a high risk of falling in the present or in the near future. Among them, This represents the critical threshold corresponding to a 50% probability of falling, as indicated by the risk function. The probability output can be used for risk grading and early warning, as well as triggering adaptive intervention strategies.
[0094] Example: Suppose that after a person is modeled, the system collects and calculates the following three risk indicator values at a certain moment: Construct a risk potential function based on a preset fusion strategy: Among them, setting As a weighting coefficient, then: Next, the system within the time window The time-weighted product is performed within the time frame, using an exponentially decaying weight function: Get the fall risk function value at the current moment: The integral calculation result is: Finally, the preset risk function value is taken. (The corresponding preset fall risk threshold is a critical point with a 50% probability of falling) is mapped to the fall risk probability through a function: Output conclusion: At time t, the probability of falling obtained by the system assessment is 60.3%, which exceeds the warning threshold. Therefore, the system can trigger a high-risk warning and recommends to start corresponding preventive measures (such as issuing an alarm, adjusting the support strength of auxiliary equipment, etc.).
[0095] The embodiments of this application have the following beneficial effects: This application acquires multi-source motion state data reflecting human posture, balance, and movement changes during rehabilitation training; performs time-synchronized processing and feature fusion on the multi-source motion state data to construct a unified multimodal temporal motion state representation; models individual stability characteristics based on historical training data and extracts individualized stability feature parameters that reflect differences in movement among individuals; constructs a human dynamic stability margin analysis model to quantitatively assess stability changes during human movement; then predicts and judges the risk of fall within a preset time window and outputs the corresponding fall risk judgment results; it can identify and warn of risk states in advance before a fall occurs, improving the safety, reliability, and intelligence level of the rehabilitation training process.
[0096] Based on the fall prevention prediction method for rehabilitation training scenarios described in the above embodiments, this application also provides a fall prevention prediction system for rehabilitation training scenarios, such as... Figure 4 As shown, Figure 4 This is a schematic diagram of a fall prevention prediction system for rehabilitation training scenarios provided in an embodiment of this application. The fall prevention prediction system 4 for rehabilitation training scenarios includes: an acquisition module 401, a preprocessing module 402, a determination module 403, and a prediction module 404, wherein... The acquisition module 401 is used to acquire multimodal motion state data during rehabilitation training; the multimodal motion state data includes multi-source sensor motion state data and visual motion state data. The preprocessing module 402 is used to preprocess the multi-source sensor motion state data and visual motion state data to obtain a multimodal motion sequence. The determining module 403 is used to encode the multimodal motion sequence to obtain individual stable feature centers and stability features; and to determine stability confidence and dynamic stability margin data based on the individual stable feature centers and stability features. The prediction module 404 is used to construct a risk state vector based on the stability confidence and the dynamic stability margin data; and to predict the human fall risk state based on the risk state vector and the preset weight coefficients, and determine the fall risk judgment result.
[0097] Based on the above embodiments of the fall prevention prediction method for rehabilitation training scenarios, this application also provides a fall prevention prediction device for rehabilitation training scenarios, such as... Figure 5 As shown, Figure 5 This is a schematic diagram of a fall prevention prediction device for a rehabilitation training scenario provided in an embodiment of this application. The fall prevention prediction device 5 for a rehabilitation training scenario includes: a processor 501 and a memory 502. The memory 502 is used to store computer programs; the processor 501 is used to call and run the computer programs from the memory to execute a fall prevention prediction method for a rehabilitation training scenario as described in the above embodiment.
[0098] In the embodiments of this application, the processor 501 described above can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that for different devices, the electronic device used to implement the above processor function can also be other types, and the embodiments of this application do not specifically limit it.
[0099] This application provides a computer-readable storage medium storing a computer program for implementing, when executed by a processor, a fall prevention prediction method for rehabilitation training scenarios as described in any of the above embodiments.
[0100] For example, the program instructions corresponding to a fall prevention prediction method for a rehabilitation training scenario in this embodiment can be stored on a storage medium such as an optical disc, hard disk, or USB flash drive. When the program instructions corresponding to a fall prevention prediction method for a rehabilitation training scenario in the storage medium are read or executed by an electronic device, a fall prevention prediction method for a rehabilitation training scenario as described in any of the above embodiments can be implemented.
[0101] Furthermore, in the embodiments of this application, the functional modules can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.
[0102] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method of this embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0103] It should be understood that the phrases "one embodiment," "an embodiment," or "some embodiments" mentioned throughout the specification mean that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment," "in one embodiment," or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the embodiments in this application are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The descriptions of the various embodiments above tend to emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, these will not be repeated here.
[0104] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.
[0105] In addition, each functional module in the various embodiments of this application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the integrated modules can be implemented in hardware or in the form of hardware plus software functional units.
[0106] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.
[0107] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.
[0108] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.
[0109] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.
[0110] The above description is merely an embodiment of this application, but the protection scope of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.
Claims
1. A fall prevention prediction method for rehabilitation training scenarios, characterized in that, The method includes: Acquire multimodal motion state data during rehabilitation training; wherein, the multimodal motion state data includes multi-source sensor motion state data and visual motion state data; The multi-source sensor motion state data and visual motion state data are preprocessed to obtain a multimodal motion sequence; Based on the multimodal motion sequence, encoding is performed to obtain individual stable feature centers and stability features; and based on the individual stable feature centers and stability features, stability confidence and dynamic stability margin data are determined. Based on the stability confidence level and the dynamic stability margin data, a risk state vector is constructed; and based on the risk state vector and preset weight coefficients, the risk state of human fall is predicted, and the fall risk judgment result is determined.
2. The method according to claim 1, characterized in that, The preprocessing of the multi-source sensor motion state data and visual motion state data to obtain a multimodal motion sequence includes: The multi-source sensor motion state data and visual motion state data are aligned and resampled to obtain aligned multi-source sensor motion state data and aligned visual motion state data. Feature extraction and feature fusion are performed on the aligned multi-source sensor motion state data and the aligned visual motion state data to obtain a comprehensive motion state vector; Based on a predetermined individual scale factor, the comprehensive motion state vector is normalized to obtain the current normalized state vector. Anomaly analysis is performed based on the normalized state vector and the pre-acquired historical moment state vector to determine a reliable state; and based on the reliable state, the multimodal motion sequence is determined.
3. The method according to claim 1, characterized in that, The encoding based on the multimodal motion sequence yields individual stable feature centers and stability features, including: Multiple different temporal enhancement operators are applied to the multimodal motion sequence to generate multiple view sequences; The multiple view sequences are encoded using a deep temporal coding network to obtain latent features; Based on the latent features, the individual stable feature center is determined; The stability features are obtained by encoding the multimodal motion sequence using a pre-determined feature encoding model.
4. The method according to claim 1, characterized in that, The determination of stability confidence and dynamic stability margin data based on the individual stable feature center and the stability feature includes: Stability is calculated based on the individual stable feature center and the stability feature to determine the stability confidence level; Based on the individual stability feature center, the stability feature, and the preset stability domain radius, the dynamic stability margin is determined; Based on the dynamic stability margin, calculate the rate of change of stability margin; The dynamic stability margin and the rate of change of the stability margin are defined as the dynamic stability margin data.
5. The method according to claim 1, characterized in that, The construction of a risk state vector based on the stability confidence level and the dynamic stability margin data includes: The dynamic stability margin in the dynamic stability margin data is processed by a normalization function to determine the instantaneous stability margin risk vector. The stability margin change rate in the dynamic stability margin data is processed by a normalization function to obtain a stability trend change risk vector. Based on the stability confidence level, the individual steady-state pattern deviation risk vector is determined; The risk state vector is determined based on the instantaneous stability margin risk vector, the stable trend change risk vector, and the individual steady-state mode deviation risk vector.
6. The method according to claim 1, characterized in that, The step of predicting the risk status of a person falling based on the risk status vector and preset weight coefficients, and determining the fall risk assessment result, includes: The risk potential function value is determined based on the risk state vector and the preset weight coefficients; The risk state of a human body falling is predicted by using the risk potential function value and a preset time weight, and the result of the fall risk assessment is determined.
7. The method according to claim 6, characterized in that, The step of predicting the risk state of a human body from a fall using the risk potential function value and a preset time weight, and determining the fall risk assessment result, includes: By using the risk potential function value and the preset time weight, the risk state of human fall is predicted, and the fall risk function value is obtained. The fall risk probability is obtained by mapping the fall risk function value to the preset risk function value corresponding to the preset fall risk threshold; If the probability of falling is greater than the preset risk threshold, then the risk assessment result is determined to be that there is a risk of falling. If the probability of falling is less than or equal to the preset fall risk threshold, then the fall risk judgment result is determined to be that there is no fall risk.
8. A fall prevention prediction system for rehabilitation training scenarios, characterized in that, The fall prevention prediction system for rehabilitation training scenarios includes: an acquisition module, a preprocessing module, a determination module, and a prediction module, wherein, The acquisition module is used to acquire multimodal motion state data during rehabilitation training; wherein, the multimodal motion state data includes multi-source sensor motion state data and visual motion state data; The preprocessing module is used to preprocess the multi-source sensor motion state data and visual motion state data to obtain a multimodal motion sequence; The determining module is used to encode the multimodal motion sequence to obtain individual stable feature centers and stability features; and to determine stability confidence and dynamic stability margin data based on the individual stable feature centers and stability features. The prediction module is used to construct a risk state vector based on the stability confidence level and the dynamic stability margin data; and to predict the human fall risk state based on the risk state vector and preset weight coefficients, and determine the fall risk judgment result.
9. A fall prevention prediction device for rehabilitation training scenarios, characterized in that, include: Processor and memory, of which, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It stores executable instructions for causing a processor to execute, thereby implementing the method of any one of claims 1 to 7.