An integrated rehabilitation monitoring device and gait evaluation method

Abstract

This invention discloses an integrated rehabilitation monitoring device and gait evaluation method, comprising a standardized fixed-length mat with embedded matrix pressure sensors, a three-dimensional visual perception module with edge privacy desensitization function, and a central evaluation module with built-in statistical gait norms. This method extracts anonymized skeletal vectors in real time at the front end using visual sensors, ensuring patient privacy. Under natural conditions, it simultaneously collects features of the patient traversing the mat. Utilizing a height- and weight-based normalization algorithm and a dynamic time warping algorithm, the system automatically eliminates the influence of individual differences and uneven patient walking speeds (such as pauses or slow walking) on ​​the data, mapping real-time features to a large data norm envelope for quantitative comparison. This invention achieves low-invasiveness and highly robust rehabilitation effect evaluation, solving the problems of poor privacy compliance and insufficient adaptability to abnormal gait in existing devices, and providing an objective and accurate clinical assessment method for hemiplegic rehabilitation.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation equipment technology, and in particular to an integrated rehabilitation monitoring device and gait evaluation method. Background Technology

[0002] Currently, in the field of rehabilitation medicine, gait analysis is the "gold standard" for evaluating the degree of motor function recovery in patients after stroke, spinal cord injury, and orthopedic surgery. However, existing gait assessment technologies face multiple bottlenecks in clinical application. First, assessment methods are limited and highly invasive: traditional assessments rely on scales, which are highly subjective and coarse-grained. Sophisticated devices such as wearable sensors need to be strapped to the patient's limbs, which is difficult for hemiplegic patients to put on and take off, and the psychological burden of these devices often leads to significant deviations between the collected gait data and the natural state. Infrared light capture systems, while highly accurate, have stringent space requirements, making them inconvenient for deployment in clinics. Second, there are privacy protection deficiencies and data security risks: while visual perception technology can provide rich features, directly collecting patient video streams poses a serious risk of medical privacy breaches. During data transmission and storage, the exposure of facial and physiological features often makes compliance audits difficult, limiting the widespread adoption of intelligent assessment devices. Finally, the algorithms lack robustness under complex gaits: hemiplegic gait is highly uncertain, manifesting as slow walking, non-periodic pauses, dragging of the feet, or circling movements. Existing algorithms are mostly based on the assumption of "uniform periodicity," which makes them prone to failure when faced with patient pauses or abnormal tremors. In addition, the pressure interference from walking aids such as canes and accompanying persons also significantly reduces the signal-to-noise ratio.

[0003] Traditional wearable devices (IMU sensors, foot pressure pads) require patients to wear limb-binding modules, which is not only inconvenient to put on and take off, but also easily makes patients feel "monitored", causing the collected gait to become a "contrived gait" controlled by consciousness.

[0004] Traditional visual assessment equipment often collects raw video streams, which involves serious issues of patient facial and physiological privacy, and the processes for data anonymization, transmission, and ethical approval in hospitals are extremely cumbersome.

[0005] Most existing gait algorithms are based on the assumption of "uniform speed and periodic cycles". When hemiplegic patients experience pauses, tremors, or irregular walking due to dystonia, traditional algorithms often lose data or report errors because they cannot identify the cycle.

[0006] Traditional pressure mats cannot distinguish pressure sources. When patients use crutches or are assisted by caregivers, the mats collect a large number of interfering pressure points, leading to errors in gait parameter calculations.

[0007] Existing equipment can often only provide basic parameters such as stride length and stride speed, making it difficult to quantify the subtle movements unique to hemiplegic individuals, such as "dragging".

[0008] This invention features a specially designed algorithm for extracting the drag friction vector of the foot, which can quantitatively analyze the subtle interaction between the toes and the ground during the swing phase of the subject's foot. This indicator can directly reflect the specific effects of rehabilitation training on tibialis anterior muscle control and foot drop, providing a scientific basis for precise rehabilitation guidance.

[0009] To address the problems of existing rehabilitation gait evaluation technologies, such as highly invasive assessment methods, reliance on wearable devices leading to gait deviations from natural states, poor compliance with medical privacy regulations related to visual perception, and insufficient algorithm robustness due to abnormal gait patterns in hemiplegic patients (e.g., pauses, dragging, interference from assistive devices), this invention provides an integrated rehabilitation monitoring device and gait evaluation method. Summary of the Invention

[0010] The purpose of this invention is to provide an integrated rehabilitation monitoring device and gait evaluation method that can solve the problems of highly invasive rehabilitation assessment, privacy leakage risks, and poor adaptability to abnormal gaits in existing technologies. Privacy is protected through edge-side desensitization technology, and the DTW algorithm is used to solve the problem of non-uniform walking assessment.

[0011] The technical solution adopted in this invention is as follows:

[0012] An integrated rehabilitation monitoring device includes,

[0013] A standardized pressure-sensing mat has a fixed-length base and an internally integrated matrix-distributed array of pressure sensors for real-time capture of the plantar pressure distribution image stream P(x, y, t), the ground contact phase sequence, and the drag friction vector, where x and y represent the positions of the pressure sensors in the mat's planar coordinate system, and t represents the sampling time.

[0014] A three-dimensional visual perception module, installed above one end of the mat, is used for non-contact acquisition of the subject's three-dimensional skeletal node motion trajectory B. i (t), where i is the node number and B represents the coordinate vector of the node in three-dimensional space;

[0015] The spatiotemporal alignment processing unit is used to align the two-dimensional coordinate system S of the floor mat. pad (x,y) and the three-dimensional spatial coordinate system S of the three-dimensional visual perception module vision (X,Y,Z) are mapped and registered, and millisecond-level time synchronization is achieved;

[0016] The central evaluation engine has pre-stored statistical gait norms built based on big data samples, which are used to execute evaluation algorithms and output rehabilitation progress reports; the outputs of the standardized pressure sensing mat and the three-dimensional visual perception module are respectively connected to the input of the central evaluation module through the spatiotemporal alignment processing unit.

[0017] The standardized pressure-sensing floor mat is composed of multiple standardized modules cascaded together. Each module has a built-in matrix of resistive thin-film pressure sensors. The floor mat base uses a hard aluminum alloy shell combined with a polymer anti-slip skin.

[0018] The three-dimensional visual perception module uses an edge computing terminal with an integrated depth camera, which is fixed to the top of the end of the mat via a bracket, with a top angle set to 30° to 45°. The three-dimensional visual perception module integrates an edge-side anonymization processing module, which is used to identify and extract the coordinates of key points of the human skeleton in real time at the image acquisition front end, and simultaneously perform Gaussian blur or black frame masking on the face and biometric information in the original image, and only output a vectorized human frame model without identity recognition features to the central evaluation module.

[0019] The edge-side anonymization processing module includes:

[0020] Infrared depth imaging unit, used to acquire depth maps rather than color RGB images;

[0021] The skeletal point extraction unit has a built-in embedded processor and runs a lightweight model to extract 21 core skeletal points of the human body in real time. The core skeletal points include the bilateral hip joints, knee joints, ankle joints, heels, and toes.

[0022] The instantaneous destruction unit is used to erase the original depth map and point cloud data from the cache after extracting the skeletal vector sequence;

[0023] The output unit is used to output a lightweight text stream, which includes a frame number, a timestamp, a sequence of skeletal joint coordinates, and an anonymization status identifier.

[0024] The central evaluation module includes a non-uniform gait robustness algorithm module, the specific execution steps of which include: extracting the subject's original walking time sequence Q={q1,q2, ..., q n}, where q i The instantaneous center of gravity displacement characteristic value of the subject; calling the pre-stored standard gait sequence R={r1,r2, ..., r m Construct the distance matrix D m×n The formula for calculating the elements is d(i, j) = (q i -r j ) 2 , where q i r represents the feature value of the i-th frame in the original time series Q of the subject. j Let d(i, j) represent the feature value of the j-th frame in the standard gait sequence R, and let d(i, j) represent the squared Euclidean distance between two frames. The Dynamic Time Warping (DTW) algorithm is used to calculate the minimum warping path to eliminate the nonlinear time offset caused by pauses and slowing down.

[0025] The spatiotemporal alignment processing unit includes:

[0026] The spatial mapping subunit is used to establish the homography matrix H between the two-dimensional coordinate system and the three-dimensional spatial coordinate system by using infrared reflective markers set at the corner points of the ground mat.

[0027] The time synchronization subunit is used to send synchronization pulses to the ground mat and the 3D vision perception module simultaneously, aligning the ground mat sampling frames with the camera depth frames.

[0028] The interpolation completion sub-unit is used to complete the visual node data by using a third-order spline interpolation algorithm to fill the gaps on the time axis.

[0029] The spatiotemporal alignment processing unit uses a homography transformation matrix H to achieve coordinate mapping:

[0030]

[0031] Where (X, Y) are the horizontal and vertical coordinates of the skeletal nodes projected onto the ground by the 3D vision perception module, (x, y) are the corresponding physical coordinates of the pressure mat, and H is a 3×3 transformation matrix used to characterize the rotation, translation and scaling relationship between the two coordinate systems; by capturing the instantaneous overlap between the foot contact point and the visual projection point in the early stage of detection, the spatial transformation parameters are calculated and corrected.

[0032] Includes the following steps:

[0033] Multi-source noise removal: Establish a pressure connectivity model, and set the dynamic search radius R based on the centroid projection position of visual tracking. mask Filter out interference signals from assistive devices or caregivers outside the filter radius;

[0034] Hemiplegic-specific feature extraction: Define the plantar drag index TDI, monitor the force on the mat in real time during the swing phase, if a continuous weak pressure vector along the direction of travel is detected and the vector coincides with the visually observed toe trajectory, it is determined to be plantar drag; specifically, the lower limb lifting ability is quantitatively assessed by calculating the low-pressure friction vector monitored by the mat during the swing phase.

[0035] Wherein, TDI stands for Toe Drag Index, used to quantify abnormal friction in the oscillating phase; T swing P represents the time set of the subject's lower limbs in the swing phase as identified by the 3D vision module. low (t) represents the time T swing During this period, the floor mat detected a weak pressure vector below a preset pressure threshold; Area sweepThe total physical area swept by the sole of the foot during the entire gait cycle is used to quantitatively assess the insufficient foot lift height and proximal compensation of the affected lower limb.

[0036] Data normalization mapping: Based on the subject's height and weight, the raw data is converted into percentage support P. norm and normalized step size L norm Obtain the subject's height H sub and weight W sub Convert the original pressure value P into a percentage support force P equal to body weight. norm =(P / W sub )×100%; Convert the stride length L to a normalized stride length L relative to the height. norm =(L / H sub )×100%;

[0037] Norm comparison and evaluation: The subject data are mapped to the confidence interval envelope of the statistical norm, the overlap coefficient and symmetry assessment index are calculated, and a rehabilitation progress report is generated.

[0038] In the multi-source noise removal step, a dynamic search mask is established with the projection points of the left and right ankle nodes in the skeletal model onto the mat as the center. Only the pressure signal inside the mask is retained, and the pressure signal outside the mask is marked as environmental noise and removed.

[0039] The amplitude range of the continuous weak pressure vector is 1N to 5N.

[0040] In the norm comparison and evaluation step, the generated rehabilitation progress report includes a comparison view of the subject's gait curve and the norm envelope.

[0041] An integrated rehabilitation monitoring system includes the integrated rehabilitation monitoring device described in any one of the claims, wherein the central evaluation engine is configured to perform the gait evaluation method described herein.

[0042] This invention integrates non-contact 3D vision with an embedded floor mat, allowing subjects to pass through naturally without wearing any devices. This "non-invasive" monitoring captures the most accurate pathological conditions, significantly improving the objectivity of clinical assessments.

[0043] Furthermore, this invention integrates an edge-side anonymization processing module, completing the "image-to-vector" transformation instantly upon image acquisition. The system neither outputs nor stores the original image, eliminating the risk of privacy leaks at the source and allowing the device to be easily deployed in open clinics or public rehabilitation corridors. Furthermore, this invention introduces a Dynamic Time Warping (DTW) algorithm, achieving nonlinear time alignment between real-time signals and standard norms. This technology can automatically "compress" or "stretch" the time axis, accurately identifying gait characteristics in non-uniform states and effectively quantifying "walking hesitation" caused by cognitive or motor planning impairments. Furthermore, through the spatiotemporal coupling of visual skeletal trajectories and pressure contact points, this invention allows the system to automatically identify and filter out pressure noise not belonging to the patient's foot (such as cane support points and caregiver footprints). This enables the device to be applied to critically ill patients in the early stages of rehabilitation, filling the gap in the compatibility of existing devices with walking aids.

[0044] This invention features a specially designed algorithm for extracting the drag friction vector of the foot, which can quantitatively analyze the subtle interaction between the toes and the ground during the swing phase of the subject's foot. This indicator can directly reflect the specific effects of rehabilitation training on tibialis anterior muscle control and foot drop, providing a scientific basis for precise rehabilitation guidance. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A schematic diagram of the overall system architecture and installation layout of an integrated rehabilitation monitoring device provided in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram illustrating the principle of spatiotemporal coordinate system mapping and dynamic mask search in an embodiment of the present invention;

[0048] Figure 3 This is a logic mapping diagram for nonlinear time axis alignment using the Dynamic Time Warping (DTW) algorithm in an embodiment of the present invention;

[0049] Figure 4 This is a partial schematic diagram of the identification of drag marks on the soles of hemiplegic patients and the extraction of friction vectors in an embodiment of the present invention;

[0050] Figure 5 This invention provides an overall process flow diagram of a gait evaluation method.

[0051] Figure 6 This is a schematic diagram of the rehabilitation progress report interface output by an embodiment of the present invention, showing the comparison between the subject's gait curve and the norm envelope. Detailed Implementation

[0052] The technical solutions of 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.

[0053] like Figure 1 , 2 As shown in Figures 3 and 4, the present invention includes...

[0054] The standardized pressure-sensing mat has a fixed-length base and an integrated matrix-distributed array of pressure sensors for real-time capture of the plantar pressure distribution image stream P(x, y, t), ground contact phase sequence, and drag friction vector. Here, x and y represent the positions of the pressure sensors in the mat's planar coordinate system, and t represents the sampling time. The standardized pressure-sensing mat is composed of multiple cascaded standardized modules. Each module contains a matrix of resistive thin-film pressure sensors. The mat base uses a hard aluminum alloy shell combined with a high-polymer anti-slip skin. The standardized pressure-sensing mat is assembled from multiple 1.5m × 0.8m module units, with a sensor density of no less than 4 sensors / cm² and a sampling frequency of no less than 100Hz within each module unit.

[0055] A three-dimensional visual perception module, installed above one end of the mat, is used for non-contact acquisition of the subject's three-dimensional skeletal node motion trajectory B. i (t), where i is the node number and B represents the coordinate vector of the node in three-dimensional space; the three-dimensional visual perception module adopts an edge computing terminal with an integrated depth camera, which is fixed above the end of the mat by a bracket, and the downward angle is set to 30° to 45°; the three-dimensional visual perception module integrates an edge-side anonymization processing module, which is used to identify and extract the coordinates of key points of human skeleton in real time at the image acquisition front end, and simultaneously perform Gaussian blur or black frame masking on the face and biometric information in the original image, and only outputs a vectorized human frame model without identity recognition features to the central evaluation module;

[0056] The edge-side anonymization processing module includes:

[0057] Infrared depth imaging unit, used to acquire depth maps rather than color RGB images;

[0058] The skeletal point extraction unit has a built-in embedded processor and runs a lightweight model to extract 21 core skeletal points of the human body in real time. The core skeletal points include the bilateral hip joints, knee joints, ankle joints, heels, and toes.

[0059] The instantaneous destruction unit is used to erase the original depth map and point cloud data from the cache after extracting the skeletal vector sequence;

[0060] The output unit is used to output a lightweight text stream, which includes a frame number, a timestamp, a sequence of skeletal joint coordinates, and an anonymization status identifier.

[0061] The spatiotemporal alignment processing unit is used to align the two-dimensional coordinate system S of the floor mat. pad (x,y) and the three-dimensional spatial coordinate system S of the three-dimensional visual perception module vision (X,Y,Z) are mapped and registered, and millisecond-level time synchronization is achieved; the spatiotemporal alignment processing unit includes:

[0062] The spatial mapping subunit is used to establish the homography matrix H between the two-dimensional coordinate system and the three-dimensional spatial coordinate system by using infrared reflective markers set at the corner points of the ground mat.

[0063] The time synchronization subunit is used to send synchronization pulses to the ground mat and the 3D vision perception module simultaneously, aligning the ground mat sampling frames with the camera depth frames.

[0064] The interpolation completion sub-unit is used to complete the visual node data by using a third-order spline interpolation algorithm to fill the gaps on the time axis.

[0065] The spatiotemporal alignment processing unit uses a homography transformation matrix H to achieve coordinate mapping:

[0066]

[0067] Where (X, Y) are the horizontal and vertical coordinates of the skeletal nodes projected onto the ground by the 3D vision perception module, (x, y) are the corresponding physical coordinates of the pressure mat, and H is a 3×3 transformation matrix used to characterize the rotation, translation and scaling relationship between the two coordinate systems; by capturing the instantaneous overlap between the foot contact point and the visual projection point in the early stage of detection, the spatial transformation parameters are calculated and corrected.

[0068] The central evaluation engine has pre-stored statistical gait norms built based on big data samples, which are used to execute evaluation algorithms and output rehabilitation progress reports; the outputs of the standardized pressure sensing mat and the three-dimensional visual perception module are respectively connected to the input of the central evaluation module through the spatiotemporal alignment processing unit.

[0069] The central evaluation module includes a non-uniform gait robustness algorithm module, the specific execution steps of which include: extracting the subject's original walking time sequence Q={q1,q2, ..., q n}, where q i The instantaneous center of gravity displacement characteristic value of the subject; calling the pre-stored standard gait sequence R={r1,r2, ..., r m Construct the distance matrix D m×n The formula for calculating the elements is d(i, j) = (q i -r j ) 2 , where q i r represents the feature value of the i-th frame in the original time series Q of the subject. j Let d(i, j) represent the feature value of the j-th frame in the standard gait sequence R, and let d(i, j) represent the squared Euclidean distance between the two frames. The Dynamic Time Warping (DTW) algorithm is used to calculate the minimum warping path to eliminate the nonlinear time offset caused by pauses and slowing down.

[0070] A gait evaluation method specifically includes the following steps:

[0071] Multi-source noise removal: Establish an 8-model of pressure connectivity domain, and set a dynamic search radius R based on the centroid projection position of visual tracking. mask The process filters out interference signals from assistive devices or caregivers outside the filter radius. In the multi-source noise removal step, a dynamic search mask is established with the projection points of the left and right ankle nodes in the skeletal model onto the mat as the center. Only the pressure signals within the mask are retained, and the pressure signals outside the mask are marked as environmental noise and removed.

[0072] Hemiplegic-specific feature extraction: Define the plantar drag index TDI, monitor the force on the mat in real time during the swing phase, if a continuous weak pressure vector along the direction of travel is detected and the vector coincides with the visually observed toe trajectory, it is determined to be plantar drag; specifically, the lower limb lifting ability is quantitatively assessed by calculating the low-pressure friction vector monitored by the mat during the swing phase.

[0073] TDI stands for Toe Drag Index, used to quantify abnormal friction in the oscillating phase; T swing P represents the time set of the subject's lower limbs in the swing phase as identified by the 3D vision module. low (t) represents the time T swing During this period, the floor mat detected a weak pressure vector below a preset pressure threshold (e.g., 5N); Area sweepThe total physical area swept by the sole of the foot during the entire gait cycle is used to quantitatively assess the insufficient foot lift and proximal compensation of the affected lower limb; the amplitude range of the continuous weak pressure vector is 1N to 5N.

[0074] Data normalization mapping: Based on the subject's height and weight, the raw data is converted into percentage support P. norm and normalized step size L norm Obtain the subject's height H sub and weight W sub The original pressure value P is converted into a percentage support force P of body weight. norm =(P / W sub )×100%. Convert the stride length L to a normalized stride length L relative to height. norm =(L / H sub )×100%.

[0075] Norm Comparison and Evaluation: Subject data are mapped to the confidence interval envelope of statistical norms, and the overlap coefficient and symmetry assessment index are calculated to generate a rehabilitation progress report. The central evaluation module performs confidence interval envelope comparison: the subject's P... norm The curve is superimposed on the norm curve of healthy individuals of the same age and sex to calculate the overlap coefficient S and the deviation vector V. The rehabilitation progress report includes a symmetry assessment indicator: calculating the ratio T of the ground contact time between the affected and healthy sides. ratio Maximum load ratio F ratio And the symmetry of the center of gravity swing trajectory (COP). In the norm comparison and evaluation step, the generated rehabilitation progress report includes a comparison view of the subject's gait curve and the norm envelope.

[0076] This application also discloses an integrated rehabilitation monitoring system, including the aforementioned integrated rehabilitation monitoring device, wherein the central evaluation engine is configured to execute the aforementioned gait evaluation method.

[0077] This invention achieves "precision, non-intrusiveness, and anonymity" in rehabilitation assessment through multimodal fusion and edge computing technologies. It employs a standardized pressure-sensing mat of fixed length, combined with a depth camera for remote monitoring. Subjects do not need to wear any components; data collection is completed naturally by walking through the mat without any instructions. This approach minimizes psychological interference and ensures the authenticity of rehabilitation evaluations through non-contact, non-intrusive monitoring. The visual module incorporates an anonymization module, performing "skeleton vectorization" processing at the hardware front end. The system only outputs the coordinate stream of human joint points, without retaining the original images. This strategy of blocking privacy data at the physical source completely solves compliance challenges in medical assessments. A Dynamic Time Warping (DTW) algorithm is introduced to perform non-linear time alignment on non-uniform walking sequences, eliminating the impact of pauses on periodic analysis. The system possesses a specific recognition algorithm that can accurately distinguish and filter pressure noise from crutches and caregivers, and specifically extract the "foot dragging marks" unique to hemiplegic patients, achieving accurate identification of pathological characteristics. Statistical norms are constructed based on big data, and normalized mapping is performed using covariates such as height and weight. The system can eliminate individual physiological differences and map the patient's gait to a standard confidence interval envelope for quantitative comparison. This approach can not only assess the rehabilitation stage but also provide refined improvement guidance for indicators such as center of gravity shift and symmetry.

[0078] To further explain and illustrate this invention, the following detailed discussion uses specific examples to illustrate the system hardware architecture and environment deployment of this invention. The physical deployment of the device is as follows: Figure 1 As shown. Standardized pressure-sensing mat: Composed of four cascaded standardized modules, each 1.5 meters long, totaling 6 meters in length and 0.8 meters in width. Each module contains a matrix of 300×160 resistive thin-film pressure sensors. The mat base uses a rigid aluminum alloy shell combined with a high-polymer anti-slip surface to ensure no wrinkles form when the patient drags along. 3D visual perception module: Employs an edge computing terminal with an integrated depth camera, fixed 2.0 meters directly above the end of the mat via a hanger, with a top-down viewing angle set to 30°–45° to ensure coverage of the entire mat area's gait trajectory. Computing hardware: The mat connects to the central evaluation module via an industrial Ethernet network, and the visual module transmits vector data via a dedicated encrypted link.

[0079] Automatic registration and synchronization of spatial coordinate systems. For example... Figure 2As shown, to achieve the fusion of visual trajectory and pressure distribution, this invention performs the following registration steps: Spatial mapping: During system initialization, infrared reflective markers are placed at the four corners of the mat. The 3D vision module identifies the coordinates (X,Y,Z) of the markers in the camera coordinate system and establishes a homography matrix H with the physical coordinate system (x,y) of the mat. Time synchronization: The central evaluation module simultaneously sends synchronization pulses to the mat and the camera, aligning the 100Hz sampling frame of the mat with the 30fps depth frame of the camera. For gaps on the time axis, a third-order spline interpolation algorithm is used to complete the visual node data, ensuring that at any given millisecond, there is a correspondence between pressure points and skeletal joint points.

[0080] Edge-side privacy anonymization logic. Its core lies in the preprocessing of the vision module: Infrared depth imaging: The camera does not acquire color RGB images, only depth maps. Joint extraction: The built-in embedded processor (such as the NVIDIA Jetson series) runs a lightweight model to extract 21 core skeletal points of the human body in real time (including both hips, knees, ankles, heels, and toes). Instantaneous destruction: After extracting the skeletal vector sequence, the original depth map and point cloud data that may contain facial features are immediately erased from the cache. The final output to the evaluation module is only a lightweight text stream in the following format: {Frame_ID:001, Timestamp: 164000, Skeleton_Joints: [[x1, y1, z1], ...], Status:Anonymized}.

[0081] Multimodal evaluation algorithm for complex gait. 1. Interference noise identification and removal. In clinical settings, when patients use canes or are assisted by caregivers, the floor mat generates multiple pressure clusters. This algorithm filters through the following logic: calculating the left and right ankle nodes (x... ankle ,y ankle The projection point on the mat. Establish a radius R around the projection point. mask =15cm dynamic search mask. Only pressure signals within the mask are retained; pressure signals outside the mask (such as the dotted pressure of the cane, the caregiver's gait) are marked as environmental noise and removed. 2. DTW normalization processing for non-uniform gait. For example... Figure 3 As shown, for the common "paroxysmal pauses" in hemiplegic patients, the system performs the following mathematical processing: Let the collected sequence of the subject's center of gravity fluctuations be P. test The pre-stored standard gait norm sequence is P. std Because of P test The length is affected by pauses much more than P std The system constructs a cumulative distance matrix C(i,j): C(i,j)=dist(q i ,rj )+min{C(i-1,j),C(i,j),C(i-1,j-1)}. By backtracking to find the optimal path, the sampling frames corresponding to the pause points are "folded". In this way, even if the patient pauses midway, the dynamic characteristics of their gait (such as the peak force at the moment of ground contact) can still be compared with the norm at the same phase point. 3. Hemiplegic characteristics: Foot drag mark recognition. For example Figure 4 As shown, the system monitors the force on the mat in real time during the swing phase. In a normal person, the pressure on the sole of the foot should be zero during the swing phase. If the mat detects a continuous, weak pressure vector (1N–5N) along the direction of travel, and this vector coincides with the visually observed toe trajectory, it is determined as "foot dragging." The system records the dragging duration and friction area as key indicators for assessing the subject's lower limb lifting ability.

[0082] The key technical parameters of the integrated rehabilitation monitoring device are shown in Table 1.

[0083] Table 1 Key Technical Parameters of Integrated Rehabilitation Monitoring Device

[0084]

[0085] In practical applications, the specific execution process of the gait evaluation method described in this invention can be divided into the following stages:

[0086] System calibration phase: Perform coordinate mapping and time synchronization.

[0087] Data acquisition phase: The floor mat and vision module synchronously acquire pressure image streams and skeletal point vectors.

[0088] Privacy desensitization phase: Perform edge-side vectorization processing.

[0089] Noise filtering stage: Dynamic masking technology is used to eliminate pressure noise from canes and accompanying persons.

[0090] Feature normalization stage: The DTW algorithm is called to normalize the time axis of the original sequence of non-uniform walking.

[0091] Report generation phase: Data is mapped to norms, and the final output of the central evaluation module is as follows: Figure 6 The rehabilitation progress report shown is illustrated above. The specific process of each stage has already been discussed and will not be repeated here.

[0092] Scenario: Early rehabilitation assessment for stroke patients. 1. Patient Zhang, left hemiplegic, requires crutches for walking. 2. Zhang, without wearing any sensors, is assisted by family members through this device. 3. System processing: Automatically removes signals from the three support points of the crutches and the family member's footprints, retaining only the pressure from Zhang's feet. 4. Algorithm recognition: DTW identifies a 1.5-second pause by Zhang at a point 3 meters due to support; the algorithm automatically removes this time redundancy. 5. Output results: The system calculates that Zhang's left foot contact time accounts for 75% of the step cycle (normal is 60%), the center of gravity trajectory shifts significantly to the right, and obvious toe drag is detected. 6. Rehabilitation recommendations: It is recommended to strengthen the left tibialis anterior muscle and adjust the height of the walking aid to improve balance.

[0093] In the description of this invention, it should be noted that for directional terms, such as "center," "lateral," and "vertical," the appropriate terms may be used.

[0094] The directions and positional relationships indicated by symbols such as "direction", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", and "counterclockwise" are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and should not be construed as limiting the specific protection scope of the present invention.

[0095] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0096] Note that the above description is merely a preferred embodiment and application of the technical principles of the present invention. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the specific embodiments described herein, and may include many other effective embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. An integrated rehabilitation monitoring device, characterized in that: include, A standardized pressure-sensing mat has a fixed-length base and an internally integrated matrix-distributed array of pressure sensors for real-time capture of the plantar pressure distribution image stream P(x, y, t), the ground contact phase sequence, and the drag friction vector, where x and y represent the positions of the pressure sensors in the mat's planar coordinate system, and t represents the sampling time. A three-dimensional visual perception module, installed above one end of the mat, is used for non-contact acquisition of the subject's three-dimensional skeletal node motion trajectory B. i (t), where i is the node number and B represents the coordinate vector of the node in three-dimensional space; The spatiotemporal alignment processing unit is used to align the two-dimensional coordinate system S of the floor mat. pad (x,y) and the three-dimensional spatial coordinate system S of the three-dimensional visual perception module vision (X,Y,Z) are mapped and registered, and millisecond-level time synchronization is achieved; The central evaluation engine has pre-stored statistical gait norms built based on big data samples, which are used to execute evaluation algorithms and output rehabilitation progress reports; the outputs of the standardized pressure sensing mat and the three-dimensional visual perception module are respectively connected to the input of the central evaluation module through the spatiotemporal alignment processing unit.

2. The integrated rehabilitation monitoring device according to claim 1, characterized in that, The standardized pressure-sensing floor mat is composed of multiple standardized modules cascaded together. Each module has a built-in matrix of resistive thin-film pressure sensors. The floor mat base uses a hard aluminum alloy shell combined with a polymer anti-slip skin.

3. The integrated rehabilitation monitoring device according to claim 1, characterized in that, The three-dimensional visual perception module uses an edge computing terminal with an integrated depth camera, which is fixed to the top of the end of the mat via a bracket, with a top angle set to 30° to 45°. The three-dimensional visual perception module integrates an edge-side anonymization processing module, which is used to identify and extract the coordinates of key points of the human skeleton in real time at the image acquisition front end, and simultaneously perform Gaussian blur or black frame masking on the face and biometric information in the original image, and only output a vectorized human frame model without identity recognition features to the central evaluation module. The edge-side anonymization processing module includes: Infrared depth imaging unit, used to acquire depth maps rather than color RGB images; The skeletal point extraction unit has a built-in embedded processor and runs a lightweight model to extract 21 core skeletal points of the human body in real time. The core skeletal points include the bilateral hip joints, knee joints, ankle joints, heels, and toes. The instantaneous destruction unit is used to erase the original depth map and point cloud data from the cache after extracting the skeletal vector sequence; The output unit is used to output a lightweight text stream, which includes a frame number, a timestamp, a sequence of skeletal joint coordinates, and an anonymization status identifier.

4. The integrated rehabilitation monitoring device according to claim 1, characterized in that, The central evaluation module includes a robust algorithm module for non-uniform gait, and the specific execution steps include: extracting the subject's original walking time sequence Q={q1,q2,...,q...} n }, where q i The instantaneous center of gravity displacement characteristic value of the subject; calling the pre-stored standard gait sequence R={r1,r2,..., r m Construct the distance matrix D m×n The formula for calculating the elements is d(i, j) = (q i -r j ) 2 , where q i r represents the feature value of the i-th frame in the original time series Q of the subject. j Let d(i, j) represent the feature value of the j-th frame in the standard gait sequence R, and let d(i, j) represent the squared Euclidean distance between two frames. The Dynamic Time Warping (DTW) algorithm is used to calculate the minimum warping path to eliminate the nonlinear time offset caused by pauses and slowing down.

5. The integrated rehabilitation monitoring device according to claim 1, characterized in that, The spatiotemporal alignment processing unit includes: The spatial mapping subunit is used to establish the homography matrix H between the two-dimensional coordinate system and the three-dimensional spatial coordinate system by using infrared reflective markers set at the corner points of the ground mat. The time synchronization subunit is used to send synchronization pulses to the ground mat and the 3D vision perception module simultaneously, aligning the ground mat sampling frames with the camera depth frames. The interpolation completion sub-unit is used to complete the visual node data by using a third-order spline interpolation algorithm to fill the gaps on the time axis. The spatiotemporal alignment processing unit uses a homography transformation matrix H to achieve coordinate mapping: ； Where (X, Y) are the horizontal and vertical coordinates of the skeletal nodes projected onto the ground by the 3D vision perception module, (x, y) are the physical coordinates of the corresponding pressure mat, and H is a 3×3 transformation matrix used to characterize the rotation, translation and scaling relationship between the two coordinate systems; by capturing the instantaneous overlap between the foot contact point and the visual projection point in the early stage of detection, the spatial transformation parameters are calculated and corrected.

6. A gait evaluation method based on the integrated rehabilitation monitoring device according to any one of claims 1-5, characterized in that, Includes the following steps: Multi-source noise removal: Establish a pressure connectivity model, and set the dynamic search radius R based on the centroid projection position of visual tracking. mask Filter out interference signals from assistive devices or caregivers outside the filter radius; Hemiplegic-specific feature extraction: Define the plantar drag index TDI, monitor the force on the mat in real time during the swing phase, if a continuous weak pressure vector along the direction of travel is detected and the vector coincides with the visually observed toe trajectory, it is determined to be plantar drag; specifically, the lower limb lifting ability is quantitatively assessed by calculating the low-pressure friction vector monitored by the mat during the swing phase. ; Wherein, TDI stands for Toe Drag Index, used to quantify abnormal friction in the oscillating phase; T swing P represents the time set of the subject's lower limbs in the swing phase as identified by the 3D vision module. low (t) represents the time T swing During this period, the floor mat detected a weak pressure vector below a preset pressure threshold; Area sweep The total physical area swept by the sole of the foot during the entire gait cycle is used to quantitatively assess the insufficient foot lift height and proximal compensation of the affected lower limb. Data normalization mapping: Based on the subject's height and weight, the raw data is converted into percentage support P. norm and normalized step size L norm Obtain the subject's height H sub and weight W sub Convert the original pressure value P into a percentage support force P equal to body weight. norm =(P / W sub )×100%; Convert the stride length L to a normalized stride length L relative to the height. norm =(L / H sub )×100%; Norm comparison and evaluation: The subject data are mapped to the confidence interval envelope of the statistical norm, the overlap coefficient and symmetry assessment index are calculated, and a rehabilitation progress report is generated.

7. The gait evaluation method according to claim 6, characterized in that, In the multi-source noise removal step, a dynamic search mask is established with the projection points of the left and right ankle nodes in the skeletal model onto the mat as the center. Only the pressure signal inside the mask is retained, and the pressure signal outside the mask is marked as environmental noise and removed.

8. The gait evaluation method according to claim 6, characterized in that, The amplitude range of the continuous weak pressure vector is 1N to 5N.

9. The gait evaluation method according to claim 9, characterized in that, In the norm comparison and evaluation step, the generated rehabilitation progress report includes a comparison view of the subject's gait curve and the norm envelope.

10. An integrated rehabilitation monitoring system, characterized in that, The integrated rehabilitation monitoring device includes any one of claims 1 to 8, wherein the central evaluation engine is configured to perform the gait evaluation method according to any one of claims 9 to 13.

Images