A Gait Analysis System for Orthopedic Rehabilitation Based on Multi-Sensor Fusion

The orthopedic rehabilitation gait analysis system, which integrates multiple sensors, solves the problem of difficulty in distinguishing between the limited local physical structure on the affected side and the translateral compensatory behavior in existing technologies. It enables collaborative analysis of data from the affected and healthy sides and accurate determination of gait response types, thereby improving the reliability of gait assessment.

CN123058718APending Publication Date: 2026-11-10UNIV OF ELECTRONICS SCI & TECH OF CHINA +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611321517.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing orthopedic rehabilitation gait analysis systems are unable to effectively distinguish between local physical structural limitations on the affected side and translateral compensatory behavior. Furthermore, phase delays and fixed time windows in conventional data processing lead to errors in gait response type determination, making it impossible to accurately assess abnormal gait.

Method used

An orthopedic rehabilitation gait analysis system employing multi-sensor fusion reconstructs the kinematic state space through a data preconditioning module, adaptively sets the time window using a feedforward windowing module, and outputs jerk sequences and translateral transient momentum transfer indices by combining spatial constraint calculation modules and momentum transfer calculation modules, thereby achieving collaborative analysis and feature decoupling of data from the affected and healthy sides.

Benefits of technology

It eliminates the phase delay problem, ensures accurate determination of gait response type, improves the reliability and accuracy of gait assessment, and can quantify translateral momentum compensation transfer behavior.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123058718A_ABST
    Figure CN123058718A_ABST
Patent Text Reader

Abstract

This invention relates to the field of medical rehabilitation assessment technology and discloses an orthopedic rehabilitation gait analysis system based on multi-sensor fusion. The system includes: a data preconditioning module that reconstructs the kinematic state space using a state observer and outputs an acceleration sequence; a feedforward windowing module that calculates the transient pseudo-kinetic energy gradient based on the joint angular velocity before heel strike and adaptively sets an effective time window; a spatial constraint calculation module that calculates the composite hysteresis index and spatial escape norm within the window, combining plantar force characteristics; a momentum transfer calculation module that calculates the composite hysteresis index and the translateral transient momentum transfer index of the healthy side acceleration within the positive hysteresis range; and a feature decoupling output module that performs a state machine comparison of the spatial escape norm and momentum transfer index, outputting gait classification labels. This invention decouples and distinguishes between physically constrained and translateral compensatory gait, eliminates the phase delay introduced by conventional differential filtering, and improves the accuracy of rehabilitation gait assessment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical rehabilitation assessment technology, specifically to an orthopedic rehabilitation gait analysis system based on multi-sensor fusion. Background Technology

[0002] In orthopedic rehabilitation, gait analysis is an important tool for assessing the recovery of lower limb motor function. Current analysis systems typically utilize multidimensional sensors to collect lower limb kinematic and dynamic data to aid in evaluating the rehabilitation progress.

[0003] Existing orthopedic rehabilitation gait analysis methods largely rely on spatial kinematic parameters of a single limb side or simple ground-touching event parameters for assessment. Due to the limited dimensionality of the information obtained, existing systems struggle to effectively distinguish between local physical structural constraints on the affected side and translateral compensatory momentum transfer behaviors caused by pain avoidance when patients exhibit abnormal gait. This results in assessments that often only reflect surface-level movement abnormalities, making it difficult to decouple and determine the actual physical causes of the abnormal gait.

[0004] Meanwhile, in conventional sensor data processing, to obtain higher-order kinematic parameters, systems typically rely on differential operations combined with low-pass filtering. While this approach smooths the data, it easily introduces a fixed temporal phase delay. When performing joint analysis of bilateral lower limb data, this fixed delay disrupts the phase consistency of cross-lateral temporal correlation calculations, causing shifts in the calculated correlation parameters and leading to misjudgments of gait response types. Furthermore, existing analysis systems mostly use fixed-length time windows when extracting gait event data, failing to dynamically adjust based on the patient's actual motion buffering state before ground contact. When facing abnormal gait speeds or variable foot impact conditions at different rehabilitation stages, fixed windows easily truncate crucial dynamic mechanical response data, further increasing errors in subsequent feature extraction and classification.

[0005] Therefore, this invention proposes an orthopedic rehabilitation gait analysis system based on multi-sensor fusion to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an orthopedic rehabilitation gait analysis system based on multi-sensor fusion. This system solves the problems of existing gait analysis methods, which struggle to decouple and distinguish between the physical constraints of the affected side and translateral compensatory behavior, and the fact that phase delay and fixed time windows in conventional data processing can easily lead to errors in the determination of gait response types.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an orthopedic rehabilitation gait analysis system based on multi-sensor fusion, comprising: The data preconditioning module is used to receive the raw inertial data sequence, perform coordinate system transformation and gravity component subtraction, and reconstruct the kinematic state space through the state observer to output the jerk and angular acceleration sequences; The feedforward windowing module is used to extract the joint angular velocity sequence before heel contact with the ground from the original inertial data sequence, calculate the transient pseudo-kinetic energy gradient, perform forward processing on the transient pseudo-kinetic energy gradient to obtain the feedforward impact intensity parameter, and set an effective time window based on the feedforward impact intensity parameter. The spatial constraint calculation module is used to calculate the composite hysteresis index, which includes the sagittal load-bearing hysteresis index, based on the dynamic change of the total vertical force of the foot and the real-time effective ground contact area within the effective time window. It also constructs a three-dimensional escape vector by combining preset multi-dimensional motion parameters to output the spatial escape norm. The momentum transfer calculation module is used to extract the composite hysteresis index within the effective time window, and to extract the healthy side jerk sequence from the jerk sequence, and to calculate the translateral transient momentum transfer index using a normalized cross-correlation function within the positive hysteresis range; The feature decoupling output module is used to receive the spatial escape norm and the translateral transient momentum transfer index, and execute state machine comparison logic to output a classification label for the physical response type.

[0008] Furthermore, the process by which the data preconditioning module reconstructs the kinematic state space through the state observer specifically includes: Within the Kalman filter framework, kinematic state equations and observation equations are established. The original inertial data sequence, after coordinate system transformation and gravity component subtraction, is used as the observation variable. Higher-order derivative parameters are incorporated into the state variables for real-time recursive estimation. The jerk sequence and the angular acceleration sequence are extracted from the updated state variables.

[0009] Furthermore, the process by which the feedforward windowing module calculates the transient pseudo-kinetic energy gradient specifically includes: Based on the heel contact event trigger point, extract the sagittal angular velocity of the affected knee joint and the sagittal angular velocity of the affected ankle joint at the time corresponding to the heel contact event trigger point and the time corresponding to the preset discrete derivative step size backward from the joint angular velocity sequence. Based on the system's discrete sampling period, the rate of change of the sagittal angular velocity of the affected knee joint and the sagittal angular velocity of the affected ankle joint within the preset discrete derivative step size is calculated respectively. Based on the system's discrete sampling period, the rate of change of the sagittal angular velocity of the affected knee joint and the sagittal angular velocity of the affected ankle joint within the preset discrete derivative step size is calculated respectively. The two calculated rates of change are weighted and summed by setting normalized weighting coefficients to obtain the transient pseudo-kinetic energy gradient.

[0010] Furthermore, the process by which the feedforward windowing module sets the effective time window based on the feedforward impact intensity parameter specifically includes: Obtain the preset baseline window length, minimum window physical limit, maximum window physical limit, and gradient adjustment coefficient; The basic adjustment length is obtained by subtracting the product of the gradient adjustment coefficient and the feedforward impact strength parameter from the reference window length. The basic adjustment length is truncated within the range of the minimum window physical limit and the maximum window physical limit, and then determined as the final length of the effective time window after integerization.

[0011] Furthermore, the process by which the spatial constraint calculation module calculates the composite hysteresis index, which includes the sagittal load-bearing hysteresis index, specifically includes: Calculate the absolute value of the first-order rate of ascent of the total vertical force on the sole of the foot and the sagittal deceleration of the pressure center; A monotonically transitioning area weighting function is constructed using the real-time effective ground contact area as the independent variable; The sagittal load-bearing hysteresis index is obtained by combining the reciprocal of the absolute value of the first-order rate of ascent of the total vertical force on the sole of the foot with the sagittal deceleration of the pressure center using the area weighting function and its complementary weights, and the sagittal load-bearing hysteresis index is used as the composite hysteresis index.

[0012] Furthermore, the preset multidimensional motion parameters include the yaw angle acceleration on the affected side horizontal plane and the coronal pressure gradient. The process by which the spatial constraint calculation module constructs a three-dimensional escape vector in conjunction with the multidimensional motion parameters to output the spatial escape norm specifically includes: Extract the yaw angle acceleration on the affected side in the horizontal plane; The coronal pressure gradient is obtained by calculating the difference between the pressure sum of the medial region and the pressure sum of the lateral region of the plantar pressure array on the affected side. The sagittal load-bearing hysteresis index, the yaw angle acceleration of the affected side horizontal plane, and the coronal pressure gradient are combined according to a preset weight to form the three-dimensional escape vector, and the maximum value of the weighted L2 norm is obtained within the effective time window as the spatial escape norm.

[0013] Furthermore, the process by which the momentum transfer calculation module calculates the translateral transient momentum transfer exponent using a normalized cross-correlation function within the positive lag range specifically includes: The preset maximum physical lag limit is compared with the relative boundary limit obtained by subtracting one from the length of the effective time window, and the minimum value of the two is selected as the upper limit of the search. The search range for the sliding hysteresis step is limited to an interval that is greater than zero and does not exceed the upper limit of the search, which constitutes the actual optimization range; Within the actual optimization range, the cross-correlation coefficient is calculated after performing mean-reduction and variance-normalization processing on the acceleration sequence of the extracted composite hysteresis index and the corresponding hysteresis step number of the translation. The maximum value of the cross-correlation coefficient is searched within the actual optimization range, and the maximum value is extracted as the translateral transient momentum transfer index.

[0014] Furthermore, the process by which the feature decoupling output module executes the state machine comparison logic specifically includes: Spatial escape norm thresholds and translateral momentum transfer temporal correlation thresholds were pre-calibrated using a subject sample library; The received spatial escape norm is compared with the spatial escape norm threshold to generate a Boolean variable for spatial escape characteristics; The received translateral transient momentum transfer index is compared with the translateral momentum transfer time-series correlation threshold to generate a time-series momentum transfer feature Boolean variable.

[0015] Furthermore, the process by which the feature decoupling output module outputs a classification label for the physical response type based on the combination of the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable specifically includes: When both the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable indicate that the corresponding feature exceeds the corresponding threshold, the evasive high-frequency momentum transfer feature label is output. When the spatial escape feature Boolean variable indicates that the corresponding feature exceeds the corresponding threshold and the temporal momentum transfer feature Boolean variable indicates that the corresponding feature does not exceed the corresponding threshold, the structurally restricted feature label is output.

[0016] Furthermore, the process by which the feature decoupling output module outputs a classification label for the physical response type based on the combination of the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable also includes: When both the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable indicate that the corresponding feature has not exceeded the corresponding threshold, output a gait label without abnormal compensation. When the spatial escape feature Boolean variable indicates that the corresponding feature does not exceed the corresponding threshold and the temporal momentum transfer feature Boolean variable indicates that the corresponding feature exceeds the corresponding threshold, the label of invalid environmental interference sample is output and invalidation processing is performed on the current window data.

[0017] This invention provides an orthopedic rehabilitation gait analysis system based on multi-sensor fusion. It has the following beneficial effects: 1. This invention reconstructs the kinematic state space using a state observer to output higher-order components such as jerk sequences, replacing conventional differential low-pass filtering. Combined with a mechanism for calculating the translateral transient momentum transfer exponent within the positive lag range, it eliminates the phase delay problem introduced by conventional filtering and establishes a unified time reference for the collaborative analysis of data from the affected and healthy sides. This ensures the accuracy of the temporal correlation calculation between the weight-bearing hysteresis of the affected side and the transient acceleration of the healthy side, thereby objectively quantifying the translateral momentum compensation transfer behavior in gait.

[0018] 2. This invention utilizes a feedforward windowing module to extract the joint angular velocity sequence before heel strike and calculate the transient pseudo-kinetic energy gradient, then adaptively sets the effective time window based on this parameter. This mechanism enables the system to dynamically adjust the length of the post-strike analysis interval according to the actual motion buffering state of the lower limb before strike, solving the problem of time-domain truncation errors that are easily caused by traditional fixed time windows under different walking speeds or abnormal gaits, ensuring complete coverage and accurate extraction of the dynamic mechanical response process in the initial stage of strike.

[0019] 3. This invention combines the spatial escape norm constructed by the spatial constraint calculation module with the translateral transient momentum transfer index, and performs state machine comparison through the feature decoupling output module. This structure changes the judgment mode that relies solely on single-limb side parameters, and jointly compares the three-dimensional weight-bearing escape characteristics of the affected side with the associated compensatory characteristics of the translateral body. This achieves decoupling and differentiation between physically constrained gait and evasive high-frequency momentum transfer gait, and can directly filter out invalid samples caused by environmental interference, thus improving the reliability of gait assessment. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the system framework of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a schematic diagram of the data preconditioning signal processing principle of the present invention; Figure 4 This is a flowchart of the logic for feedforward windowing at the end of the swing phase in this invention. Figure 5 This is a diagram of the spatial constraint calculation logic structure of the present invention; Figure 6 This is a timing correlation diagram for the translateral transient momentum transfer calculation of the present invention; Figure 7 This is the logic diagram of the decoupled output state machine of the present invention; Figure 8 The waveform diagrams of the time series data for the translateral feature calculation of the test subjects in this invention are shown in Figure (a), which is a waveform diagram of feedforward window extraction and angular velocity deceleration feature, and Figure (b) is a waveform diagram of the time series correlation calculation of translateral transient momentum transfer. Figure 9 This is a two-dimensional scatter clustering distribution diagram for gait feature decoupling in this invention.

[0021] Among them, 10 is the data preconditioning module; 20 is the feedforward windowing module; 30 is the spatial constraint calculation module; 40 is the momentum transfer calculation module; and 50 is the feature decoupling output module. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see the appendix Figure 1 This invention provides an orthopedic rehabilitation gait analysis system based on multi-sensor fusion, comprising: The data preconditioning module 10 receives the raw inertial data sequence acquired by the underlying sensor, performs a transformation from the sensor coordinate system to the human anatomical coordinate system or the ground reference coordinate system on the raw inertial data sequence, and after deducting the gravity component from the acceleration sequence, reconstructs the kinematic state space through the state observer, and outputs the acceleration sequence and angular acceleration sequence with low delay synchronous estimation. The feedforward windowing module 20 extracts the joint angular velocity sequence of the affected lower limb before the heel strike event to calculate the transient pseudo-kinetic energy gradient, and performs positive processing on the transient pseudo-kinetic energy gradient to obtain the feedforward impact intensity parameter, and adaptively sets the time window length of the post-strike stage. The spatial constraint calculation module 30 calculates the composite hysteresis index based on the dynamic change of the total vertical force of the plantar pressure and the real-time effective ground contact area distribution within the time window set by the feedforward window module 20. The composite hysteresis index includes the sagittal load-bearing hysteresis index, and combines multi-dimensional parameters to construct a three-dimensional escape vector to output the spatial escape norm. Within the time window determined by the feedforward windowing module 20, the momentum transfer calculation module 40 extracts the composite hysteresis index generated by the spatial constraint calculation module 30 and the healthy side vertical jerk sequence output by the data preconditioning module 10, and calculates the transverse transient momentum transfer index using the normalized cross-correlation function within the restricted positive hysteresis range. The feature decoupling output module 50 receives the spatial escape norm output by the spatial constraint calculation module 30 and the translateral transient momentum transfer index output by the momentum transfer calculation module 40, executes the state machine comparison logic, and outputs the classification label of the lower limb physical response type.

[0024] See attached document Figure 2This invention provides a gait analysis method for orthopedic rehabilitation based on multi-sensor fusion. This method is executed by the aforementioned orthopedic rehabilitation gait analysis system based on multi-sensor fusion and includes the following steps: S10, the data preconditioning module 10 receives the raw inertial data sequence synchronously acquired by the underlying sensor, performs coordinate system transformation and gravity component subtraction processing, reconstructs the kinematic state space through the state observer, and outputs the jerk sequence and angular acceleration sequence; S20, the feedforward windowing module 20 extracts the joint angular velocity sequence of the affected lower limb before the heel touches the ground to solve the transient pseudo-kinetic energy gradient, performs positive processing on the transient pseudo-kinetic energy gradient to obtain the feedforward impact intensity parameter, and sets the effective time window for the post-ground contact stage based on the feedforward impact intensity parameter. S30, the spatial constraint calculation module 30 calculates the composite hysteresis index based on the vertical force change of the plantar pressure array and the real-time effective ground contact area within the set time window, constructs the three-dimensional escape vector and calculates the spatial escape norm. S40, the momentum transfer calculation module 40 extracts the calculated composite hysteresis index and the healthy side vertical jerk sequence within the time window, establishes a time-series correlation model using the discrete cross-correlation function, and optimizes within the positive hysteresis range that satisfies the physical causal relationship to calculate the trans-side transient momentum transfer index. S50, the feature decoupling output module 50 compares the spatial escape norm with the translateral transient momentum transfer index, determines whether the parameter reaches the feature threshold, and outputs the classification label of the physical response type.

[0025] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0026] See attached document Figure 3 The data preconditioning module 10 is used to synchronously receive, unify coordinates, subtract gravity components, and perform low-latency state estimation of multi-source inertial data. Its processing includes the following sub-steps: S101, establish the system reference sampling frequency and discrete time index rules, and extract the original inertial data sequence.

[0027] In one embodiment of the present invention, the data preconditioning module 10 operates according to a reference sampling frequency. The data received from the underlying sensor nodes has a corresponding discrete sampling period of 1. ,and The system is based on As a discrete-time index.

[0028] The data preconditioning module 10 synchronously acquires data from sensor nodes distributed across different limbs of the subject according to a discrete sampling period, and extracts the raw inertial data sequence from the synchronously acquired dataset. The raw inertial data sequence includes the acceleration sequence of the affected side. Angular velocity sequence of the affected side healthy side acceleration sequence And the healthy side angular velocity sequence.

[0029] In one specific embodiment, inertial sensor nodes are respectively disposed on the affected thigh, affected calf, affected foot, and healthy calf or healthy foot, and the plantar pressure array is disposed in the insole of the affected side. The data preconditioning module 10 acquires the installation attitude calibration parameters of each inertial sensor node in a static standing state or a preset calibration posture of the subject, and converts the inertial data in the sensor coordinate system to the human anatomical coordinate system or the ground reference coordinate system according to the calibration parameters. For acceleration sequences, the data preconditioning module 10 subtracts the gravity component after completing the coordinate transformation to obtain the motion components in the sagittal, coronal, horizontal, and vertical directions.

[0030] Synchronization triggering, bus reading, and data alignment between underlying sensor nodes can be achieved using hardware trigger signals, timestamp alignment, or microcontroller synchronous acquisition mechanisms; this embodiment does not limit the specific communication method.

[0031] S102, construct the kinematic state observer and reconstruct the kinematic state space.

[0032] In gait dynamics analysis, directly performing differential operations on the original discrete inertial sequence to obtain higher-order differential components can amplify sensor measurement noise and easily generate high-frequency jitter components. Conventional methods typically involve applying a low-pass filter after the differential operation; however, while suppressing high-frequency noise, the low-pass filter introduces group delay, causing a time offset in the signal. This time offset weakens phase consistency in subsequent cross-side timing correlation calculations.

[0033] To maintain consistency in sampling timing while suppressing noise, the data preconditioning module 10 replaces the differential cascaded low-pass filter with a discrete linear noise reduction observer. This observer incorporates higher-order derivative parameters into the state variables for estimation during real-time recursion, thereby reducing the differential noise amplification effect and avoiding the fixed group delay introduced by conventional low-pass filters.

[0034] Taking the acceleration channel as an example, the data preconditioning module 10 establishes the kinematic state equation and observation equation within the Kalman filter framework. Among them, the two-dimensional state vector... Defined as: ; In the formula, This represents the system acceleration state estimated by the observer. This represents the estimated jerk state of the system.

[0035] The state equation established by the data preconditioning module 10 is as follows: ; In the formula, The state transition matrix can be set according to the discrete-time recursive relationship between acceleration states and acceleration states. This is a system process noise sequence.

[0036] The data preconditioning module 10 treats the acceleration component sequence after coordinate system transformation and gravity component subtraction as the observed variables of the system. Establish the observation equation: ; In the formula, The observation matrix is ​​set as follows: , To measure the noise sequence.

[0037] To ensure stable convergence of the observer, the data preconditioning module 10 predetermines the measurement noise covariance matrix. process noise covariance matrix Measurement noise covariance matrix The process noise covariance matrix can be determined by the continuous sampling variance of the sensor in a static, motionless state. It can be initialized based on the empirical boundary of acceleration changes during daily walking and can be further corrected during the system calibration phase. The angular velocity channel adopts the same state-space modeling method, estimating angular velocity and its first derivative angular acceleration as state variables.

[0038] S103 extracts high-order kinematic parameters for low-latency synchronous estimation through state recursion update.

[0039] The data preconditioning module 10 performs prediction, gain calculation, and state correction point by point based on the state equation and observation equation. Specifically, the data preconditioning module 10 performs prior prediction based on the state estimation result and estimation error covariance of the previous sampling time, and then combines it with the observed variables at the current sampling time. Calculate the Kalman gain and apply it to the state vector. Perform a correction and update.

[0040] The updated state vector directly contains higher-order kinematic states, therefore the data preconditioning module 10 does not need to perform post-stage differential filtering on the original sequence to extract the estimated jerk sequence. Accordingly, in the angular velocity channel, the data preconditioning module 10 extracts the estimated angular acceleration sequence from the updated angular velocity state vector. .

[0041] The above processing method, while suppressing the amplification of original measurement noise and differential noise, also improves the acceleration sequence. With angular acceleration sequence It can be updated synchronously with the current sampling period, thus providing a unified time reference for subsequent spatial constraint calculations and cross-side temporal correlation calculations.

[0042] The data preconditioning module 10 uses the jerk sequence obtained from low-latency synchronous estimation. and angular acceleration sequence The output is sent to the system data flow bus for use by the feedforward windowing module 20, the spatial constraint calculation module 30, and the momentum transfer calculation module 40.

[0043] See attached document Figure 4 The feedforward windowing module 20 is used to extract the pre-movement state of the lower limb before the heel of the affected side touches the ground, and to determine the adaptive time window used for feature analysis after ground contact. The processing procedure specifically includes the following sub-steps: S201, determine the critical event of the gait cycle and extract the joint angular velocity parameters before ground contact.

[0044] The feedforward window module 20 receives the total vertical force data output by the plantar pressure insole on the affected side in real time. When the total vertical force exceeds the preset initial contact load threshold from zero baseline, the feedforward window module 20 determines the discrete sampling point corresponding to that moment as the heel contact event trigger point, denoted as... .

[0045] In one embodiment, the initial contact weight threshold can be determined as a fixed percentage of the total calibrated weight of the subject when standing statically with both feet, for example, set to 3% to 5% of body weight. By setting this threshold, the influence of insole base capacitance or resistance noise, suspension deformation, and slight contact disturbances on heel contact determination can be reduced.

[0046] Heel strike event trigger point The previous stage corresponds to the end of the swing phase of the affected lower limb. The feedforward fenestration module 20 extracts the trigger point from the system data cache. Previous sagittal angular velocity sequence of the affected knee joint Sagittal angular velocity sequence of the affected ankle joint .

[0047] Among them, the sagittal plane angular velocity sequence of the affected knee joint The relative angular velocities of the affected thigh inertial sensor nodes and the affected lower leg inertial sensor nodes in the sagittal plane of the human body were calculated; the sagittal angular velocity sequence of the affected ankle joint was obtained. It is calculated from the relative angular velocity of the inertial sensor node on the affected lower leg and the inertial sensor node on the affected foot in the sagittal plane of the human body.

[0048] Heel strike events can also be determined by combining plantar pressure thresholds, peak characteristics from inertial sensors, or a fusion of both. This embodiment does not limit the specific event detection algorithm, as long as the ground contact sampling point can be determined under the same discrete time reference. That's all.

[0049] S202, establish a transient pseudo-kinetic energy gradient model and generate feedforward impact strength parameters.

[0050] In a normal gait, there is usually a brief deceleration process at the end of the lower limb swing before heel strike to reduce the impact at the moment of contact. When the joint mobility on the affected side is limited, or when the subject compensates defensively due to pain, the swing velocity and extension velocity of the lower limb before contact will change. Therefore, the changes in the angular velocity of the knee and ankle joints in the short period before contact can be used as a feedforward parameter reflecting the pre-contact preparation state.

[0051] Feedforward window module 20 uses heel touch event trigger point Based on this, extract the time point and work backwards. The joint angular velocity parameters at each sampling point are calculated, and the transient pseudo-kinetic energy gradient is determined. Its mathematical expression is: ; In the formula, The discrete sampling period of the underlying hardware; The discrete derivative step size constant is set, and its value is determined based on the system's reference sampling frequency. The setting makes This corresponds to a physical time span of 30 to 50 milliseconds before ground contact, thus covering the actual action preparation period and smoothing out data jumps caused by single-point measurements. and These are the normalized weighted coefficients for the knee and ankle joints, respectively. .

[0052] In actual configuration, and It can be determined based on the differences in mass distribution and moment of inertia among the lower limb segments. Since the thigh segment typically has a larger mass and moment of inertia, It can be set to greater than This demonstrates the significant contribution of knee joint movement changes to the overall impact trend of the lower limbs before ground contact.

[0053] To avoid inconsistent window mapping directions due to different directions of angular velocity change, the feedforward windowing module 20 pairs transient pseudo-kinetic energy gradients. Perform forward processing to obtain feedforward impact strength parameters. In one embodiment, Can be taken as In another embodiment, It can also be taken as . It is used to characterize the intensity of changes in the motion state of the affected lower limb at the end of the swing phase before ground contact.

[0054] S203 is an adaptive time window after ground contact based on feedforward impact strength parameter mapping.

[0055] The feedforward window module 20 is based on the feedforward impact strength parameters. Adjust the length of the time window used for feature extraction and translateral correlation analysis after heel strike. A larger value indicates a more dramatic change in the lower limb's motion state before ground contact, with the impact response and momentum transfer process typically completed within a short period. In this case, shortening the observation window is beneficial for improving the accuracy of capturing local transient features. Conversely, when... When the load is relatively small, the response process after ground contact is relatively smooth, and the feedforward window module 20 extends the observation window accordingly to retain complete load-bearing response information.

[0056] The adaptive time window length set by the feedforward windowing module 20 satisfy: ; In the formula, Indicates the length of the adaptive time window; This represents the physical limit of the minimum window; Indicates the maximum physical limit of the window; Indicates the length of the reference window; Indicates the gradient adjustment coefficient; Represents the transient pseudo-kinetic energy gradient Feedforward impact strength parameters obtained after forward processing.

[0057] in, The number of sampling points can be determined based on the time span of the standard weight-bearing reaction period when a normal subject walks in a straight line on flat ground, which usually corresponds to 10% to 12% of the gait cycle time. Used to set the adjustment range of the feedforward impact intensity on the window length.

[0058] When obtained according to the above formula When the integer is not an integer, the feedforward windowing module has 20 pairs. Perform integer conversion, and use the resulting integer number of sample points as the final window length. Integer conversion can be performed using rounding, rounding up, or rounding down.

[0059] To prevent extreme gait or transient outliers from causing the window to be too short, too long, or unable to participate in subsequent related calculations, the feedforward windowing module 20 uses... and right Apply boundary constraints. It is set to be no less than the shortest number of sampling points required for a single initial heel impact response, and satisfies that the subsequent cross-correlation optimization process contains at least one valid positive hysteresis point; The number of sampling points shall not exceed the maximum number corresponding to the dual support period.

[0060] By using the aforementioned feedforward windowing method, the system can adaptively select the analysis window according to the pre-ground motion state after the ground contact event occurs, reducing the time-domain truncation error caused by the fixed window in asymmetric gait or abnormal impact gait.

[0061] See attached document Figure 5 The spatial constraint calculation module 30 is used to analyze the plantar pressure distribution, weight-bearing hysteresis state, and non-propulsion plane offset characteristics of the affected side within an adaptive time window. Its processing includes the following sub-steps: S301, extract the pressure parameters and effective contact area after ground contact.

[0062] The adaptive time window determined by the feedforward windowing module 20 Inside, the spatial constraint calculation module 30 reads the array pressure data of the insole on the affected side of the foot. The spatial constraint calculation module 30 sums the pressure values ​​of each sensing unit in the array to obtain the current discrete time point. Total vertical force and to Perform a first-order forward difference to obtain the vertical force rise rate. .

[0063] Simultaneously, the spatial constraint calculation module 30 identifies sensing units with pressure values ​​exceeding a preset noise floor threshold as valid activated units and counts the number of valid activated units. The product of the number of valid activated units and the physical area of ​​a single sensing unit is the real-time valid ground contact activation area. .

[0064] In one embodiment, the noise floor threshold is determined by the baseline output amplitude of each pressure sensing unit under no-load conditions, for example, by taking the maximum baseline output value during continuous sampling or by adding a safety margin to the maximum baseline output value. This reduces the impact of zero drift, resistance drift, capacitance residue, and slight deformation of the insole on the determination of the effective contact area.

[0065] The data addressing, pressure conversion, and sensor unit traversal method of the plantar pressure array can be determined according to the hardware specifications of the selected pressure insole. This embodiment does not limit its specific circuit interface and reading method.

[0066] S302, calculate the sagittal hysteresis index based on the effective ground contact area.

[0067] The spatial constraint calculation module 30 calculates the sagittal load-bearing hysteresis index. As one of the specific parameters of the composite hysteresis index, this parameter is used to describe the speed of weight-bearing establishment and the degree of hysteresis in the sagittal plane after the heel of the affected side contacts the ground, and serves as an input for subsequent cross-lateral temporal correlation calculations.

[0068] During the initial heel contact phase, the pressure-bearing area on the sole of the foot has not yet fully expanded, resulting in a relatively small total pressure used for normalization in the pressure center coordinate calculation. This can easily lead to numerical fluctuations in the pressure center position and its second-order variation. To reduce the impact of analytical instability of the pressure center on hysteresis determination during this phase, the spatial constraint calculation module 30 uses the real-time effective ground contact activation area. As a transitional control variable, a composite weighted model is established between the reciprocal of the rate of rise of the total vertical force and the sagittal deceleration of the pressure center.

[0069] Before performing composite weighting, the spatial constraint calculation module 30 can perform amplitude normalization or baseline normalization on the vertical force hysteresis term and the sagittal plane deceleration term of the pressure center. The normalized parameters will still be represented by their original symbols in the following formulas.

[0070] Spatial constraint calculation module 30 calculates the sagittal plane load-bearing hysteresis index. The formula is: ; In the formula, Represents the current discrete time point Sagittal load-bearing hysteresis index; This represents the area weighting function, whose value varies with... It increases and transitions monotonically from 0 to 1; in one implementation... ,in It is half the total physical area of ​​the sole of the foot. The slope constant; Represents the complementary weights of the vertical force hysteresis term; It represents the absolute value of the first-order rate of rise of the total vertical force; This represents a minimum positive constant to prevent zero, and its value can be set within the range of 10⁻⁶ to 10⁻⁴ to avoid... Division overflow occurs when the value approaches zero; This indicates the sagittal deceleration at the center of pressure.

[0071] in, Using absolute values ​​avoids the impact of local jitter on the stability of the hysteresis term when the difference value is negative. If the spatial constraint calculation module 30 performs normalization on the vertical force hysteresis term and the sagittal deceleration term at the pressure center, then the formula... and All of these represent the corresponding parameters after normalization.

[0072] Specifically, the spatial constraint calculation module 30 calculates the pressure center sagittal plane coordinates using a pressure-weighted method based on the pressure values ​​of each pressure sensing unit and their sagittal plane position coordinates in the insole coordinate system; then, it performs second-order difference on the pressure center sagittal plane coordinate sequence, or uses the same state observer estimation method as the data preconditioning module 10, to obtain the pressure center sagittal plane deceleration. .

[0073] Through the aforementioned area-weighted mechanism, when the effective contact area is small in the early stages of contact, the model primarily uses the reciprocal of the vertical force rise rate to characterize the hysteresis of weight-bearing establishment. As the plantar contact area expands, the model gradually increases the weight of the sagittal deceleration of the pressure center to reflect changes in the movement of the plantar pressure center. Thus, the sagittal weight-bearing inhibition state of the affected side can be continuously characterized from the initial heel contact to the weight-bearing development.

[0074] S303, construct the three-dimensional escape vector and calculate the spatial escape norm.

[0075] Abnormal weight-bearing during lower limb orthopedic rehabilitation is not necessarily limited to the sagittal plane. When there is pain avoidance, limited joint movement, or unstable support on the affected side, the compensatory response may extend towards horizontal rotation and coronal lateral loading. Therefore, the spatial constraint calculation module 30, based on the sagittal weight-bearing hysteresis index, introduces horizontal yaw angle acceleration and coronal pressure gradient to construct a three-dimensional escape vector.

[0076] The spatial constraint calculation module 30 extracts the data preconditioning module 10, which synchronously outputs the yaw angle acceleration on the affected side horizontal plane. In one embodiment, The angular velocity data from the inertial sensor node of the affected foot or the inertial sensor node of the affected lower leg are estimated by coordinate transformation and state observer. The corresponding angular acceleration component in the vertical axis direction in the human anatomical coordinate system is used to characterize the horizontal yaw response of the affected lower limb during the ground contact and weight-bearing phase.

[0077] The spatial constraint calculation module 30 further divides the plantar pressure array on the affected side into medial and lateral regions along the longitudinal geometric center axis of the foot. The module calculates the pressure in the medial region and... and outer region pressure and The difference between the two is used as the coronal pressure gradient. ,Right now: ; In obtaining , and Subsequently, the spatial constraint calculation module 30 combines the three factors according to sensitivity weights into a three-dimensional escape vector, and calculates the maximum value of its weighted L2 norm within the adaptive time window, which is used as the spatial escape norm. The calculation formula is as follows: ; In the formula, This represents the weighted spatial escape norm output by the spatial constraint calculation module 30; Indicates in Take the maximum value within the range; This indicates taking the L2 norm of the internal column vectors; The sensitivity weighting coefficient represents the sagittal hysteresis parameter. The sensitivity weighting coefficient representing the horizontal yaw parameter; The sensitivity weighting coefficient representing the coronal pressure gradient; Indicates the sagittal plane load-bearing hysteresis index; This represents the yaw acceleration on the horizontal plane. This represents the coronal pressure gradient.

[0078] To reduce the impact of different dimensions and numerical scales on norm calculations, , and It can be initialized based on the reciprocal of the variance of the corresponding parameter in the healthy subject sample database, or normalized based on the system calibration data.

[0079] The output of the spatial constraint calculation module 30 This is used to quantify the comprehensive spatial escape degree of the affected side during the ground contact and load-bearing phase, spreading towards sagittal hysteresis, horizontal yaw, and coronal eccentric loading, providing spatial dimensional characteristics for subsequent physical response type decoupling.

[0080] See attached document Figure 6 The momentum transfer calculation module 40 is used to analyze the positive temporal coupling relationship between the weight-bearing hysteresis state of the affected side and the vertical acceleration response of the healthy side within an adaptive time window after heel strike. The processing procedure specifically includes the following sub-steps: S401, extracts cross-side timing data and aligns it to the time base.

[0081] After the feedforward windowing module 20 determines the effective analysis time window after heel contact, the momentum transfer calculation module 40 uses this window as the cross-side time-series correlation analysis interval. The discrete-time index range of the time window is: ; Within this time window, the momentum transfer calculation module 40 reads the affected side sagittal plane load-bearing hysteresis index sequence output by the spatial constraint calculation module 30 from the system data flow bus. In this embodiment, As a specific form of the composite hysteresis index sequence, it is used to characterize the hysteresis state during the weight-bearing establishment process after the heel of the affected side touches the ground.

[0082] Simultaneously, the momentum transfer calculation module 40 reads the healthy side vertical jerk sequence output by the data preconditioning module 10. . The acceleration data collected by the healthy side inertial sensor node is obtained after coordinate system transformation, gravity component subtraction, and state observer estimation, corresponding to the vertical acceleration component in the human anatomical coordinate system or the ground reference coordinate system.

[0083] because and Both are based on the same discrete sampling period and the same ground contact event time reference, and the healthy side's jerk sequence is obtained through low-delay state estimation. Therefore, the two can be compared at the sampling time level. As a result, the momentum transfer calculation module 40 can reduce the phase shift caused by traditional post-stage filtering delay or buffer misalignment, providing a consistent time basis for subsequent cross-correlation calculations.

[0084] The buffering, array addressing, and data retrieval within the window for discrete sequences can be implemented using pointer offset, circular buffer, or fixed-length array slicing. This embodiment does not limit the specific storage structure.

[0085] S402, establish a mean-free normalized discrete cross-correlation model.

[0086] The sagittal weight-bearing lag index of the affected side Vertical acceleration sequence of the healthy side These are different physical quantities, with varying amplitude ranges, dimensions, and DC biases. Direct inner product calculations can easily lead to results affected by amplitude scales and baseline shifts. Therefore, the momentum transfer calculation module 40 employs a mean-normalized sliding cross-correlation method to evaluate the degree of waveform synchronization between the two sequences.

[0087] With sliding hysteresis steps of At that time, the momentum transfer calculation module 40 calculates the cross-side cross-correlation coefficient according to the following formula. : ; In the formula, Indicates the number of sliding hysteresis steps. The cross-correlation coefficient obtained at that time; This represents the number of sliding hysteresis points on the discrete time axis. This indicates the sagittal weight-bearing hysteresis index of the affected side within the time window; Relative to Postponed Vertical accelerometer of the healthy side at each sampling point; express In the current sliding calculation area The arithmetic mean within; express In the corresponding shift interval The arithmetic mean within; To prevent zero from being a very small positive number, for example, 10 can be used. -6 This is used to avoid division by zero when the variance of a sequence is close to zero.

[0088] By removing the mean and normalizing... It primarily reflects the temporal consistency of the two sets of sequences in terms of trend fluctuations, rather than simply reflecting the magnitude of the signal amplitude.

[0089] S403, extracting the translateral transient momentum transfer index based on positive physical hysteresis constraints.

[0090] During weight-bearing in the human gait, obstructed ground contact or delayed weight-bearing on the affected side can be transmitted to the healthy side via the pelvis, trunk, and stabilizing chain, resulting in a transient acceleration response in the vertical direction on the healthy side. This transmission process typically has a positive time delay. The number of lag steps in the cross-correlation calculation... A value less than or equal to zero means that the healthy side response occurs before the affected side ground contact delay, which does not conform to the translateral compensatory transmission sequence described in this embodiment.

[0091] Therefore, the momentum transfer calculation module 40 first limits the physical candidate range of the sliding hysteresis step number m to the positive hysteresis interval: ; In the formula, The maximum physical hysteresis limit can be determined based on the empirical delay range of the translateral biomechanical response of the human body. In one embodiment, The latency can be set from 100 milliseconds to 150 milliseconds, and the system reference sampling frequency can be adjusted accordingly. The result is obtained through conversion.

[0092] To avoid When the value is greater than or equal to the current window length, the sliding calculation interval becomes empty. The momentum transfer calculation module 40 further sets the actual optimization upper limit. : ; Therefore, the sliding hysteresis steps The actual optimization range is limited to: ; The momentum transfer calculation module 40 searches within the aforementioned actual optimization range. The maximum value is taken as the translateral transient momentum transfer exponent. The extraction rules are as follows: ; In the formula, Indicates the translateral transient momentum transfer index; Indicates in Take the maximum value within the range; This represents the upper limit of the actual optimization, and .

[0093] It is used to characterize the positive hysteresis coupling strength between the weight-bearing hysteresis sequence on the affected side and the vertical acceleration sequence on the healthy side. It does not represent the absolute value of mechanical momentum, but rather a dimensionless time-series correlation index obtained based on a normalized cross-correlation function. When A higher value indicates a strong transient coupling between the weight-bearing hysteresis change on the affected side and the subsequent vertical acceleration response on the healthy side, which can be used as the temporal feature input for the subsequent feature decoupling output module 50 to determine the translateral compensatory response.

[0094] See attached document Figure 7 The feature decoupling output module 50 is used to jointly determine the spatial escape features and cross-side temporal coupling features, and output the corresponding gait mechanical response type label. Its processing process specifically includes the following sub-steps: S501, set the spatial feature threshold and the temporal feature threshold.

[0095] The feature decoupling output module 50 receives the weighted spatial escape norm Esp output by the spatial constraint calculation module 30, and the translateral transient momentum transfer index output by the momentum transfer calculation module 40. To ensure a consistent reference for judgment results from different test subjects and at different time periods, the feature decoupling output module 50 is pre-configured with a spatial escape norm threshold. Threshold for correlation between translateral momentum transfer time series .

[0096] In one embodiment, and Calibration was performed using baseline data from a healthy subject sample bank. Baseline data were collected in a standard flat, straight, steady-state walking environment, with acquisition conditions including the same or calculable baseline sampling frequency, preset walking distance, flat test surface, and stable walking speed range.

[0097] When there are differences in weight, leg length, natural gait frequency, or gait cycle length among different test subjects, the feature decoupling output module 50 can first... and Normalization is performed, and then statistical threshold calculation is performed.

[0098] Specifically, the feature decoupling output module 50 can calculate the health sample database separately. and The statistical mean and standard deviation are calculated, and the mean plus three times the standard deviation is used as the corresponding threshold. For the translateral transient momentum transfer index... Since it is obtained from the normalized cross-correlation function, The value can be limited to no more than 1; in another embodiment, a preset percentile of the distribution of healthy samples can also be used as the threshold. .

[0099] Using the threshold calibration method described above, the feature decoupling output module 50 can distinguish normal physiological fluctuations from abnormal weight-bearing or translateral compensatory responses by taking the healthy gait baseline as a reference.

[0100] S502, construct the state machine decision logic for spatial and temporal features.

[0101] Feature decoupling output module 50 will output continuous numerical forms and The variables are converted into discrete state variables for state machine comparison. Specifically, the feature decoupling output module 50 will... With spatial escape norm threshold Compare and generate Boolean variables for spatial escape characteristics. At the same time, Threshold associated with translateral momentum transfer time series Comparisons are made to generate time-series momentum transfer characteristic Boolean variables. .

[0102] The mathematical expression for its state determination is as follows: ; ; In the formula, Boolean variable representing spatial escape characteristics. This indicates that the spatial escape characteristic exceeds a set threshold. This indicates that the space escape characteristics are within the baseline fluctuation range; This represents the weighted spatial escape norm calculated in real time. Indicates the spatial escape norm threshold; Boolean variables representing the temporal momentum transfer characteristics. This indicates that the cross-side temporal coupling characteristics exceed a set threshold. This indicates that no significant cross-side temporal coupling was detected; This represents the translateral transient momentum transfer index calculated in real time. This represents the threshold for temporal correlation of cross-side momentum transfer.

[0103] pass and The combination of features and the decoupling output module 50 separates and judges spatial anomalies and temporal anomalies, avoiding classification based on a single image feature or a single correlation feature.

[0104] S503 outputs a classification label for the physical response type based on the state machine combination output.

[0105] Feature decoupling output module 50 based on and The combination of values ​​will output the corresponding gait mechanical response analysis labels.

[0106] when and This indicates that the subject simultaneously exhibits significant spatial escape and substantial translateral temporal coupling during the weight-bearing phase on the affected side. This state suggests that after local weight-bearing shift on the affected side, a corresponding transient acceleration response occurs on the healthy side, which can be used to characterize defensive compensatory tendencies induced by pain or discomfort avoidance. Under this condition, the feature decoupling output module 50 generates and outputs avoidance-related high-frequency momentum transfer feature labels.

[0107] when and When this condition is met, it indicates that there is localized spatial load-bearing escape on the affected side of the subject, but no obvious transient coupling response on the healthy side is detected. This state can be used to characterize situations such as limited local range of motion, limited load-bearing path, or rigid supporting structure on the affected side. Its mechanical performance is more biased towards localized structural constraints than rapid cross-side compensatory transmission. Under this condition, the feature decoupling output module 50 generates and outputs structurally constrained feature labels.

[0108] when and At this time, it indicates that neither the spatial escape feature nor the cross-side temporal coupling feature exceeds the corresponding threshold. The feature decoupling output module 50 determines that the gait mechanical response within the current window is within the normal fluctuation range and outputs a gait label without abnormal compensation or maintains the normal observation state.

[0109] when and When this condition occurs, it indicates that the system did not detect spatial escape on the affected side, but the translateral temporal coupling characteristic exceeded the threshold. This state may be caused by external tripping, slippage, sudden posture adjustment, or instantaneous sensor disturbances during the test process, and its data should not be directly used to determine the type of weight-bearing compensation on the affected side. Under this condition, the feature decoupling output module 50 generates invalid environmental interference sample labels and performs discarding or alarm processing on the data in this window.

[0110] The above four state combinations cover and All possible values ​​enable the feature decoupling output module 50 to form a clear classification boundary between spatial features and temporal features.

[0111] It should be noted that the physical response type classification label output by the feature decoupling output module 50 is used to help characterize the gait mechanical response pattern of the test subject during the ground contact and weight-bearing phase, and is not directly used as a disease diagnosis conclusion.

[0112] After the classification labels are generated, the system can encapsulate them according to a preset data format and send them to a display terminal, local storage, or rehabilitation assessment software. Label rendering, communication transmission, and data writing can be implemented using conventional host computer communication protocols or database writing methods; this embodiment does not limit the specific implementation form.

[0113] To further aid in understanding the technical details and practical application effects of the present invention, a typical clinical application example is provided below based on the specific embodiments described above.

[0114] Specific application examples: To more intuitively illustrate the execution logic and calculation process of the technical solution of this invention, the following provides a specific numerical calculation example in conjunction with the gait rehabilitation assessment process of a patient with right knee osteoarthritis (OA).

[0115] See attached document Figure 8 , Figure 8 This is a waveform diagram of time-series data for calculating translateral features of a subject according to an embodiment of the present invention. In this application scenario, the system reference sampling frequency is set. =100Hz, i.e., discrete sampling period =0.01 seconds. The system defines the patient's right lower limb as the affected side and the left lower limb as the healthy side.

[0116] During the data preconditioning phase, the system detected the trigger point of the right heel strike event on the affected side. =100 (corresponding to the first second after test start). Feedforward window module 20 extracts data before ground contact. =4 sampling points (i.e.) =96 (40 milliseconds before ground contact). The measured parameters are: At 96, the angular velocity of the affected knee joint =2.0 rad / s, ankle joint angular velocity =1.0 rad / s; When =100, =0.5rad / s, =0.2 rad / s. Set the weighting coefficient. =0.7, =0.3. The feedforward windowing module 20 calculates the transient pseudo-kinetic energy gradient as follows: ; After performing the forwarding process, the feedforward impact strength parameters are obtained. =∣−32.25∣=32.25. Set the baseline window. =15 (150 milliseconds), adjustment coefficient =0.2, window physical limit =8, =25.

[0117] The adaptive time window length is calculated as follows: ; After rounding to the nearest integer, the final effective analysis window length after ground contact was determined to be 9 sampling points, i.e. The shorter window reflects that the system accurately captured the sharp deceleration and avoidance maneuvers before ground contact.

[0118] During the spatial constraint calculation phase, the spatial constraint calculation module 30 calculates the maximum sagittal hysteresis parameter, horizontal yaw parameter, and coronal pressure gradient within the aforementioned window, and then weights them (assuming...). =0.6, =0.2, After combining (=0.2), the weighted spatial escape norm is calculated. =1.42.

[0119] During the momentum transfer calculation phase, the momentum transfer calculation module 40 sets a maximum physical hysteresis limit. =15 (corresponding to a maximum latency of 150 milliseconds). To avoid window overflow, the system calculates the actual optimization upper limit. .exist Within the range, the cross-correlation calculation model has a lag number of steps. The extreme value was found at =4 (corresponding to a 40 ms delay), and the translateral transient momentum transfer exponent was extracted. .

[0120] During the feature decoupling output phase, the system calls the pre-calibrated baseline threshold. and After comparing with the state machine, a determination is made. (Right now =1) and (Right now =1). Based on this, the system outputs a label indicating the avoidance of high-frequency momentum transfer, suggesting to clinicians that the patient's current gait abnormality is mainly due to defensive compensation caused by acute pain, rather than physical locking of the joint structure.

[0121] To verify the actual classification effect and robustness of the above technical solution, this embodiment further discloses relevant comparative experimental verification data. The experiment used a sample library of 40 subjects, including 15 patients with knee / hip osteoarthritis exhibiting acute avoidance pain (avoidance group), 15 patients with severe lower limb joint contractures or sclerosis (structural limitation group), and 10 healthy controls (healthy group). The experimental control group used the traditional fixed time window (150 milliseconds) and single acceleration peak threshold determination method for gait abnormality classification; the experimental test group used the dynamic windowing and feature decoupling algorithm of this invention.

[0122] See attached document Figure 9 , Figure 9 This is a two-dimensional scatter plot showing the decoupled gait features according to an embodiment of the present invention. The horizontal axis in the figure represents the transverse transient momentum transfer index. The vertical axis represents the spatial escape norm. Experimental data shows that the healthy group samples are densely distributed in the low threshold range in the lower left corner of the chart (…). , The structurally confined group of samples is distributed upward along the longitudinal axis, exhibiting characteristics of high spatial escape but low momentum transfer. , The avoidance group samples were significantly concentrated in the high-value quadrant in the upper right of the chart (); , The state machine threshold line of this invention (shown as a dashed line in the figure) forms a clear two-dimensional decision boundary between the above three types of sample clusters, effectively avoiding cluster overlap that is easily caused by the traditional single-variable amplitude determination method.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A gait analysis system for orthopedic rehabilitation based on multi-sensor fusion, characterized in that, include: The data preconditioning module is used to receive the raw inertial data sequence, perform coordinate system transformation and gravity component subtraction, and reconstruct the kinematic state space through the state observer to output the jerk and angular acceleration sequences; The feedforward windowing module is used to extract the joint angular velocity sequence before heel contact with the ground from the original inertial data sequence, calculate the transient pseudo-kinetic energy gradient, perform forward processing on the transient pseudo-kinetic energy gradient to obtain the feedforward impact intensity parameter, and set an effective time window based on the feedforward impact intensity parameter. The spatial constraint calculation module is used to calculate the composite hysteresis index, which includes the sagittal load-bearing hysteresis index, based on the dynamic change of the total vertical force of the foot and the real-time effective ground contact area within the effective time window. It also constructs a three-dimensional escape vector by combining preset multi-dimensional motion parameters to output the spatial escape norm. The momentum transfer calculation module is used to extract the composite hysteresis index within the effective time window, and to extract the healthy side jerk sequence from the jerk sequence, and to calculate the translateral transient momentum transfer index using a normalized cross-correlation function within the positive hysteresis range; The feature decoupling output module is used to receive the spatial escape norm and the translateral transient momentum transfer index, and execute state machine comparison logic to output a classification label for the physical response type.

2. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 1, characterized in that, The process by which the data preconditioning module reconstructs the kinematic state space through a state observer specifically includes: Within the Kalman filter framework, kinematic state equations and observation equations are established. The original inertial data sequence, after coordinate system transformation and gravity component subtraction, is used as the observation variable. Higher-order derivative parameters are incorporated into the state variables for real-time recursive estimation. The jerk sequence and the angular acceleration sequence are extracted from the updated state variables.

3. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 1, characterized in that, The process by which the feedforward windowing module calculates the transient pseudo-kinetic energy gradient specifically includes: Based on the heel contact event trigger point, extract the sagittal angular velocity of the affected knee joint and the sagittal angular velocity of the affected ankle joint at the time corresponding to the heel contact event trigger point and the time corresponding to the preset discrete derivative step size backward from the joint angular velocity sequence. Based on the system's discrete sampling period, the rate of change of the sagittal angular velocity of the affected knee joint and the sagittal angular velocity of the affected ankle joint within the preset discrete derivative step size is calculated respectively. Based on the system's discrete sampling period, the rate of change of the sagittal angular velocity of the affected knee joint and the sagittal angular velocity of the affected ankle joint within the preset discrete derivative step size is calculated respectively. The two calculated rates of change are weighted and summed by setting normalized weighting coefficients to obtain the transient pseudo-kinetic energy gradient.

4. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 3, characterized in that, The process by which the feedforward windowing module sets the effective time window based on the feedforward impact intensity parameter specifically includes: Obtain the preset baseline window length, minimum window physical limit, maximum window physical limit, and gradient adjustment coefficient; The basic adjustment length is obtained by subtracting the product of the gradient adjustment coefficient and the feedforward impact strength parameter from the reference window length. The basic adjustment length is truncated within the range of the minimum window physical limit and the maximum window physical limit, and then determined as the final length of the effective time window after integerization.

5. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 1, characterized in that, The process by which the spatial constraint calculation module calculates the composite hysteresis index, which includes the sagittal load-bearing hysteresis index, specifically includes: Calculate the absolute value of the first-order rate of ascent of the total vertical force on the sole of the foot and the sagittal deceleration of the pressure center; A monotonically transitioning area weighting function is constructed using the real-time effective ground contact area as the independent variable; The sagittal load-bearing hysteresis index is obtained by combining the reciprocal of the absolute value of the first-order rate of ascent of the total vertical force on the sole of the foot with the sagittal deceleration of the pressure center using the area weighting function and its complementary weights, and the sagittal load-bearing hysteresis index is used as the composite hysteresis index.

6. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 5, characterized in that, The preset multidimensional motion parameters include the yaw angle acceleration on the affected side horizontal plane and the coronal pressure gradient. The process by which the spatial constraint calculation module combines the multidimensional motion parameters to construct a three-dimensional escape vector and output the spatial escape norm specifically includes: Extract the yaw angle acceleration on the affected side in the horizontal plane; The coronal pressure gradient is obtained by calculating the difference between the pressure sum of the medial region and the pressure sum of the lateral region of the plantar pressure array on the affected side. The sagittal load-bearing hysteresis index, the yaw angle acceleration of the affected side horizontal plane, and the coronal pressure gradient are combined according to a preset weight to form the three-dimensional escape vector, and the maximum value of the weighted L2 norm is obtained within the effective time window as the spatial escape norm.

7. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 1, characterized in that, The process by which the momentum transfer calculation module calculates the translateral transient momentum transfer index using a normalized cross-correlation function within the positive lag range specifically includes: The preset maximum physical lag limit is compared with the relative boundary limit obtained by subtracting one from the length of the effective time window, and the minimum value of the two is selected as the upper limit of the search. The search range for the sliding hysteresis step is limited to an interval that is greater than zero and does not exceed the upper limit of the search, which constitutes the actual optimization range; Within the actual optimization range, the cross-correlation coefficient is calculated after performing mean-reduction and variance-normalization processing on the acceleration sequence of the extracted composite hysteresis index and the corresponding hysteresis step number of the translation. The maximum value of the cross-correlation coefficient is searched within the actual optimization range, and the maximum value is extracted as the translateral transient momentum transfer index.

8. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 1, characterized in that, The process by which the feature decoupling output module executes the state machine comparison logic specifically includes: Spatial escape norm thresholds and translateral momentum transfer temporal correlation thresholds were pre-calibrated using a subject sample library; The received spatial escape norm is compared with the spatial escape norm threshold to generate a Boolean variable for spatial escape characteristics; The received translateral transient momentum transfer index is compared with the translateral momentum transfer time-series correlation threshold to generate a time-series momentum transfer feature Boolean variable.

9. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 8, characterized in that, The process by which the feature decoupling output module outputs a classification label for the physical response type based on the combination of the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable specifically includes: When both the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable indicate that the corresponding feature exceeds the corresponding threshold, the evasive high-frequency momentum transfer feature label is output. When the spatial escape feature Boolean variable indicates that the corresponding feature exceeds the corresponding threshold and the temporal momentum transfer feature Boolean variable indicates that the corresponding feature does not exceed the corresponding threshold, the structurally restricted feature label is output.

10. The orthopedic rehabilitation gait analysis system based on multi-sensor fusion according to claim 9, characterized in that, The process by which the feature decoupling output module outputs a classification label for the physical response type based on the combination of the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable further includes: When both the spatial escape feature Boolean variable and the temporal momentum transfer feature Boolean variable indicate that the corresponding feature has not exceeded the corresponding threshold, output a gait label without abnormal compensation. When the spatial escape feature Boolean variable indicates that the corresponding feature does not exceed the corresponding threshold and the temporal momentum transfer feature Boolean variable indicates that the corresponding feature exceeds the corresponding threshold, the label of invalid environmental interference sample is output and invalidation processing is performed on the current window data.