Compensation action recognition system and method for elbow joint rehabilitation training of children

Abstract

This invention belongs to the field of medical and health informatics and computer-aided rehabilitation assessment technology, and discloses a system and method for recognizing compensatory movements in children's elbow joint rehabilitation training. The system includes: S1, acquiring basic information and initial movement data of the child being trained, and establishing an individualized movement baseline model based on the basic information and the initial movement data. This invention segments the movement based on kinematic parameters such as wrist displacement changes, elbow movement trajectory and angular velocity changes, trunk posture, and center of gravity shift. The system divides a single rehabilitation movement into sequence segments such as preparation phase, initiation phase, main driving phase, terminal approach phase, and withdrawal phase according to movement initiation characteristics, main driving characteristics, terminal approach characteristics, and withdrawal characteristics. It can extract features separately within each independent movement phase, avoiding interference between features from different phases, providing a time reference for locating local movement deviations, and improving the accuracy of compensatory movement recognition.

Description

Technical Field

[0001] This invention belongs to the fields of medical and health care informatics and computer-aided rehabilitation technology, specifically a system and method for recognizing compensatory movements in children's elbow joint rehabilitation training. Background Technology

[0002] Elbow fractures are extremely common in children, especially active children aged 5-10. They are often caused by falls where the elbow hits the ground or by external force pulling on the joint. Surgery is an effective treatment for elbow fractures, restoring the joint's anatomical structure and providing favorable conditions for fracture healing and functional recovery. Due to the thin joint capsule, the tight internal structures of the elbow joint, and the close relationship between the joint capsule and ligaments and muscles, elbow injuries are particularly prone to contractures and stiffness. Post-operative active and passive exercises can progressively improve joint range of motion and prevent adhesions or contractures of ligaments and joint capsules. However, clinical practice has shown that due to the complex anatomical structure of the elbow joint, postoperative pain is significant. School-aged children are in a critical period of cognitive development, have poor pain tolerance, and have low adherence to rehabilitation exercises. Furthermore, children's lack of mastery of the methods, intensity, and frequency of functional exercises greatly affects their rehabilitation outcomes.

[0003] In existing elbow joint rehabilitation training, motion tracking devices and sensors are needed to monitor children's movement postures in order to help them complete specific tasks. Current motion recognition systems typically collect data on joint angles and movement trajectories during task execution and compare this data with a pre-set normal kinematic model to determine whether the training goals have been achieved. This data-driven system replaces purely manual visual observation, improving the digitalization of rehabilitation assessment.

[0004] However, existing technologies have limitations in recognizing compensatory movements in children. First, most existing systems use a uniform static evaluation threshold, failing to fully consider individual differences among trainees in terms of age, body type, affected side, and rehabilitation stage. This uniform threshold baseline model easily leads to distorted recognition results and cannot adapt to children with different ability levels. Second, existing movement analysis often treats a complete training movement as a single unit for evaluation, lacking a phased division of the entire movement cycle. This prevents the system from accurately locating the specific stage where abnormal movements occur, blurring the time points between primary and compensatory movements, and making it difficult to guide subsequent movement correction based on the evaluation results.

[0005] Furthermore, existing technologies generally suffer from limitations such as delayed assessment and singular control, affecting the effectiveness of rehabilitation training. On the one hand, existing compensatory identification often only triggers an alarm after the compensatory movement is fully formed and the patient exhibits an abnormal posture, lacking the ability to capture the state in the early stages of the compensatory movement's evolution, thus missing the opportunity for early intervention. On the other hand, existing rehabilitation systems are mostly passive monitoring and recording systems, unable to adaptively adjust subsequent target distances, heights, and rhythms based on the child's current level of compensation. They also lack a closed-loop baseline iteration mechanism that can automatically update based on the child's rehabilitation progress, causing the system to be unable to adapt to the child's ability improvement. This easily leads to the risk of overcompensation due to mismatched training difficulty, thus limiting the efficacy of intelligent rehabilitation. Summary of the Invention

[0006] The purpose of this invention is to provide a system and method for recognizing compensatory movements in children's elbow joint rehabilitation training, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a system and method for recognizing compensatory movements in children's elbow joint rehabilitation training, comprising:

[0008] S1. Obtain the basic information and initial movement data of the trainee child, establish an individualized movement baseline model based on the basic information and the initial movement data, and determine the ability boundary parameters by the individualized movement baseline model. The basic information is information that characterizes the individual differences and current rehabilitation status of the trainee child, and the initial movement data is the movement data generated by the trainee child when performing preset movements in the initial testing phase.

[0009] S2. During the process of training children performing preset elbow joint rehabilitation training tasks, acquire training movement data, and combine the individualized motion baseline model to perform stage segmentation processing on the training movement data to obtain the movement stage label sequence.

[0010] S3. Based on the action phase label sequence and capability boundary parameters, extract the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters respectively, and generate the compensation precursor state based on the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters.

[0011] S4. Compare the elbow main motion characteristic parameters, shoulder-trunk compensation characteristic parameters and compensation precursor state with the individualized motion baseline model to obtain the elbow main motion contribution results, compensation type, compensation degree and compensation initiation point.

[0012] S5. Based on the elbow main motion contribution results, compensation type, compensation degree and compensation initiation point, generate training quality results and compensation risk results, and generate training control parameters based on the training quality results and compensation risk results.

[0013] S6. Control the preset elbow joint rehabilitation training task based on the training control parameters, obtain the corresponding updated movement data, update the individualized motion baseline model based on the updated movement data, and then determine the updated capability boundary parameters by the updated individualized motion baseline model for closed-loop control of the subsequent preset elbow joint rehabilitation training task.

[0014] As a further technical solution of the present invention, the basic information includes at least one of age, affected side, upper arm length, forearm length, joint range of motion, and current rehabilitation stage, and the initial movement data includes the initial test movement data generated when the trained child performs standard test movements;

[0015] The individualized motion baseline model is established, and the capability boundary parameters are determined by the individualized motion baseline model, including: normalizing the initial motion data based on the basic information to generate the elbow main motion baseline, the compensatory motion baseline and the task completion baseline;

[0016] According to the characteristic correspondence under the same preset elbow joint rehabilitation training task and the same movement stage, the elbow main movement feature in the elbow main movement baseline is mapped to the shoulder-trunk compensatory feature in the compensatory movement baseline; the center value of the main movement baseline and the width of the main movement boundary of each elbow main movement feature are determined, as well as the center value of the compensatory baseline and the compensation warning width of each shoulder-trunk compensatory feature; the center value of the main movement baseline and the width of the main movement boundary are used to form the current acceptable main movement deviation range, and the center value of the compensatory baseline and the compensation warning width are used to form the compensation warning range, and the current acceptable main movement deviation range and the compensation warning range are determined as the capability boundary parameters.

[0017] As a further technical solution of the present invention, the training action data is processed into stages to obtain the action stage label sequence, including: based on the changes in elbow movement trajectory, wrist displacement, trunk posture and center of gravity shift, the training action data is processed to detect the start of action and the end of action.

[0018] Between the action start detection result and the action end detection result, the training action data is divided into at least some of the stages of preparation period, start period, main drive period, terminal approach period and withdrawal period according to the action start characteristics, main drive characteristics, terminal approach characteristics and withdrawal characteristics, and the action stage label sequence corresponding to the start and end times of each stage is generated.

[0019] As a further technical solution of the present invention, based on the action stage label sequence and the capability boundary parameters, the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters are extracted respectively, and the compensation precursor state is generated, including: according to the action stage label sequence, extracting the elbow main motion feature parameters representing elbow flexion and extension, elbow angular velocity, insufficient terminal extension and forearm pronation and supination in each stage, and extracting the shoulder-trunk compensation feature parameters representing shoulder elevation, shoulder abduction, trunk forward tilt, trunk rotation, trunk lateral tilt and center of gravity shift;

[0020] The elbow primary motion characteristic parameters and the shoulder-trunk compensation characteristic parameters of each stage are matched with the capability boundary parameters. When the elbow primary motion characteristic parameters shift outward from the capability boundary parameters and the shoulder-trunk compensation characteristic parameters show a continuous increasing trend, the corresponding stage's pre-compensation state is generated.

[0021] As a further technical solution of the present invention, the elbow main motion characteristic parameters, the shoulder-trunk compensation characteristic parameters, and the compensation precursor state are compared with the individualized motion baseline model to obtain the elbow main motion contribution result, the compensation type, the compensation degree, and the compensation initiation point. This includes: according to the action stage label sequence, comparing the elbow main motion characteristic parameters of each stage with the elbow main motion baseline corresponding to the same stage to obtain the elbow main motion deviation degree, and generating the elbow main motion contribution result of the corresponding stage based on the elbow main motion deviation degree, wherein the smaller the elbow main motion deviation degree, the higher the elbow main motion contribution result of the corresponding stage.

[0022] The shoulder-trunk compensation characteristic parameters of each stage are compared with the compensation motion baseline of the same stage to obtain the shoulder-trunk compensation deviation degree. The compensation degree of the corresponding stage is generated according to the shoulder-trunk compensation deviation degree. The greater the shoulder-trunk compensation deviation degree, the higher the compensation degree of the corresponding stage.

[0023] The elbow active motion contribution and compensation degree at each stage are weighted and summarized to obtain the elbow active motion contribution and compensation degree for the entire training movement.

[0024] Based on the shoulder-trunk compensation feature category that exceeds the corresponding compensation warning range and has the largest deviation degree, the compensation type is determined; and the stage and corresponding time in the action stage label sequence that first meets the conditions for the establishment of the compensation precursor state are determined as the compensation initiation point.

[0025] As a further technical solution of the present invention, the training quality result and the compensation risk result are generated based on the elbow main motion contribution result, the compensation type, the compensation degree and the compensation initiation point, and the training control parameters are generated based on the training quality result and the compensation risk result, including: determining the current training action as one of effective training, compensatory completion or ineffective training according to the combination relationship between the elbow main motion contribution result and the compensation degree;

[0026] The compensation risk outcome is determined based on the compensation type, the compensation degree, and the compensation initiation point;

[0027] The training control parameters are generated based on the training quality results and the compensation risk results. The training control parameters include at least one of the following: target distance adjustment parameters, target height adjustment parameters, movement rhythm adjustment parameters, and training subtask switching parameters.

[0028] As a further technical solution of the present invention, the preset elbow joint rehabilitation training task is controlled based on the training control parameters to obtain the corresponding updated action data, and the individualized motion baseline model is updated based on the updated action data. Then, the updated ability boundary parameters are determined by the updated individualized motion baseline model. This includes: after controlling the preset elbow joint rehabilitation training task, obtaining the updated action data corresponding to the training control parameters.

[0029] Based on the updated motion data, the elbow primary motion contribution result, the degree of compensation, and the compensation initiation point are regenerated.

[0030] When the updated elbow primary motion contribution result increases and the updated compensation level decreases, the individualized motion baseline model is updated using the updated motion data, and the updated capability boundary parameters are determined by the updated individualized motion baseline model for closed-loop control of the subsequent preset elbow joint rehabilitation training task.

[0031] A pediatric elbow joint rehabilitation training compensatory movement recognition system, including:

[0032] The basic modeling module is used to acquire basic information and initial movement data of the children being trained, establish an individualized movement baseline model based on the basic information and initial movement data, and determine the ability boundary parameters by the individualized movement baseline model.

[0033] The phase segmentation module is used to acquire training movement data during the process of training children performing preset elbow joint rehabilitation training tasks, and to perform phase segmentation processing on the training movement data in combination with an individualized motion baseline model to obtain a sequence of movement phase labels.

[0034] The feature generation module is used to extract elbow main motion feature parameters and shoulder-trunk compensation feature parameters based on the action phase label sequence and capability boundary parameters, and to generate compensation precursor states based on the elbow main motion feature parameters and shoulder-trunk compensation feature parameters.

[0035] The comprehensive judgment module is used to compare the elbow main motion characteristic parameters, shoulder-trunk compensation characteristic parameters and compensation precursor state with the individualized motion baseline model to obtain the elbow main motion contribution result, compensation type, compensation degree and compensation initiation point.

[0036] The parameter generation module is used to generate training quality results and compensation risk results based on the elbow main motion contribution results, compensation type, compensation degree and compensation initiation point, and to generate training control parameters based on the training quality results and compensation risk results.

[0037] The closed-loop update module is used to control the preset elbow joint rehabilitation training task based on the training control parameters, obtain the corresponding updated movement data, update the individualized motion baseline model based on the updated movement data, and then determine the updated capability boundary parameters by the updated individualized motion baseline model for closed-loop control of subsequent preset elbow joint rehabilitation training tasks.

[0038] As a further technical solution of the present invention, the basic modeling module is used to normalize the initial movement data based on at least one of the following: the age of the child being trained, the affected side, the upper arm length, the forearm length, the range of motion of the joint, and the current rehabilitation stage, to generate an elbow main movement baseline, a compensatory movement baseline, and a task completion baseline, and to determine the capability boundary parameters based on the correspondence between the elbow main movement baseline and the compensatory movement baseline; the stage segmentation module is used to perform movement initiation detection and movement termination detection on the training movement data based on changes in elbow movement trajectory, wrist displacement, trunk posture, and center of gravity shift, and to generate the movement stage label sequence between the movement initiation detection result and the movement termination detection result according to movement initiation characteristics, main driving characteristics, terminal approach characteristics, and withdrawal characteristics;

[0039] The feature generation module is used to generate the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters based on the action phase label sequence and the capability boundary parameters, and to generate the compensation precursor state when the elbow main motion feature parameters shift outward from the capability boundary parameters and the shoulder-trunk compensation feature parameters show a continuous increasing trend.

[0040] As a further technical solution of the present invention, the comprehensive judgment module is used to calculate the degree of deviation of the elbow main motion characteristic parameter relative to the elbow main motion baseline and the degree of deviation of the shoulder-trunk compensation characteristic parameter relative to the compensation motion baseline, so as to generate the elbow main motion contribution result and the degree of compensation, and determine the compensation type according to the feature category corresponding to the degree of deviation, and determine the compensation initiation point according to the stage in the action stage label sequence that first meets the condition for the establishment of the compensation precursor state and the corresponding time.

[0041] The parameter generation module is used to generate the training quality result based on the combination relationship between the elbow main motion contribution result and the compensation degree, generate the compensation risk result based on the compensation type, the compensation degree and the compensation initiation point, and generate the training control parameters based on the compensation risk result.

[0042] The closed-loop update module is used to obtain the updated action data corresponding to the training control parameters after controlling the preset elbow joint rehabilitation training task based on the training control parameters, and regenerate the elbow main movement contribution result, the compensation degree and the compensation initiation point based on the updated action data. When the updated elbow main movement contribution result increases and the updated compensation degree decreases, the individualized motion baseline model is updated, and the updated ability boundary parameters are determined by the updated individualized motion baseline model.

[0043] The beneficial effects of this invention are as follows:

[0044] 1. This invention solves the recognition limitations of traditional rehabilitation systems caused by uniform thresholds and overall assessments by working in tandem with a basic modeling module and a stage segmentation module. In the basic modeling stage, the system acquires basic information such as the age, affected side, upper arm length, forearm length, joint range of motion, and current rehabilitation stage of the individual children being trained. The system combines the data reflecting individual developmental differences and rehabilitation status with the initial movement data and normalizes the initial movement data through scale reference. The system can independently generate the elbow main movement baseline, compensatory movement baseline, and task completion baseline, and determine the ability boundary parameters accordingly. This mechanism ensures that the movement assessment is based on a reference that fits the individual characteristics of the child, thus improving the objectivity of the assessment. Furthermore, this invention introduces a segmented processing of movement stages based on kinematic parameters such as wrist displacement changes, elbow movement trajectory and angular velocity changes, trunk posture and center of gravity shift. The system divides a single rehabilitation movement into sequence segments such as preparation period, initiation period, main driving period, terminal approach period and withdrawal period according to movement initiation characteristics, main driving characteristics, terminal approach period and withdrawal period. It can extract features in each independent movement stage separately, avoiding mutual interference between features of different stages, providing a time reference for locating local movement deviations, and improving the accuracy of compensatory movement recognition.

[0045] 2. This invention, through a feature generation module and a comprehensive judgment module, changes the traditional method of alarming after compensatory movements are formed. It establishes an early identification mechanism. Based on stage segmentation, the system independently extracts elbow main motion feature parameters such as elbow flexion and extension and angular velocity, as well as shoulder and trunk compensatory feature parameters such as shoulder elevation and trunk forward tilt, within each stage. The system introduces a dynamic monitoring mechanism. When the comparison finds that the elbow main motion feature is shifting to the outside of the individualized ability boundary and the shoulder and trunk compensatory feature shows a continuous increasing trend, the system generates the corresponding stage's pre-compensation state. Since compensation is a gradual process of non-target segments being gradually intervened due to insufficient main force muscle groups, the capture of the pre-compensation state allows the rehabilitation intervention point to be moved forward. Subsequently, by calculating the deviation of the feature baseline, the system can output a comprehensive evaluation conclusion, determine the current training movement category, and output the elbow main motion contribution result, compensation type, compensation degree, and compensation initiation point according to the combination relationship and feature category. The above quantitative results enable the system to identify the source of movement drive and the time of compensation intervention, providing data support for the formulation of subsequent intervention strategies.

[0046] 3. This invention addresses the lack of interactive control and baseline fixation in existing systems through a parameter generation module and a closed-loop update module. After acquiring the elbow primary motor contribution result, compensation degree, and compensation initiation point, the system logically generates training quality results and determines the compensation risk result classification. Based on this, the system generates training control parameters to adaptively control the preset elbow joint rehabilitation training task. According to different compensation risk situations, the system outputs various instructions in the target distance adjustment parameter, target height adjustment parameter, movement rhythm adjustment parameter, and training sub-task switching parameter. The adaptive parameter adjustment mechanism controls the training difficulty within a reasonable range, avoiding the problem of ineffective training due to difficulty mismatch. At the same time, this invention designs a controlled closed-loop update mechanism. After controlling the task based on the training control parameters and acquiring updated action data, the system regenerates the judgment result and verifies whether the updated elbow primary motor contribution result has improved and whether the compensation degree has decreased. Only when the above conditions are met, the system uses the updated action data to update the individualized movement baseline model, thereby determining the updated ability boundary parameters. The above closed-loop update logic prevents abnormal data from affecting the benchmark model, enabling the identification system to adjust parameters and update the benchmark according to changes in the child's ability. Attached Figure Description

[0047] Figure 1 This is a block diagram of the method of the present invention;

[0048] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0049] 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.

[0050] like Figures 1 to 2As shown, the specific implementation process of the present invention will be further described in detail. To facilitate implementation by those skilled in the art, the following will elaborate on the input data organization method, processing logic, result generation method, and preferred calculation process without changing the existing core technical concept; the values ​​such as sampling frequency, repetition count, threshold interval, window length, weight coefficient, and update step size are all exemplary parameters in the preferred embodiments, used to illustrate how to implement, and do not constitute a limitation on the scope of protection. The overall main line of the present invention is still: first, an individualized motion baseline model is established based on the age, affected side, body segment length, joint range of motion, and rehabilitation stage of the trained child; then, during the preset elbow joint rehabilitation training task, data on changes in elbow, shoulder, trunk, and center of gravity are collected; single movements are segmented into stages; elbow main motion parameters and shoulder-trunk compensation parameters are separated and extracted; and then the elbow main motion contribution result, compensation type, compensation degree, and compensation initiation point are obtained; and based on the above results, training quality is judged, training parameters are adaptively adjusted, and closed-loop updates are performed.

[0051] In a preferred embodiment, the basic modeling module first acquires the basic information and initial movement data of the trained child and establishes an individualized movement baseline model. The basic information may include at least age, affected side information, upper arm length, forearm length, joint range of motion, and current rehabilitation stage. The initial movement data can be taken from the movement sequences generated when the trained child completes several standard test movements during the initial testing phase. To ensure subsequent statistical stability, it is preferable to repeat each standard test movement 3 to 5 times, each lasting 3 to 8 seconds. The visual sampling frequency can be 30Hz to 60Hz, the pressure sampling frequency can be 50Hz to 100Hz, and the optional posture sampling frequency can be 50Hz to 100Hz. The basic modeling module first aligns the coordinates of the key movement points, the coordinates of the pressure center, and the optional posture angle sequences with unified timestamps, and normalizes the raw data using the length of the corresponding body segment on the healthy side, the sum of the upper arm and forearm lengths on the trained side, or the functional length from the shoulder point to the wrist point as a scale reference. For example, when the shoulder point, elbow point, and wrist point are respectively denoted as… , , First, construct the upper arm vector and the forearm vector:

[0052]

[0053] The elbow flexion-extension angle can then be written as:

[0054]

[0055] in, To prevent extremely small positive numbers with a denominator of zero, it is preferable to select... If the line connecting the midpoint of the shoulder and the midpoint of the hip represents the principal axis of the torso, then this vector is denoted as... The vertical reference vector is denoted as Then the forward tilt angle of the torso can be written as:

[0056]

[0057] For shoulder elevation measurement, the vertical displacement difference between the shoulder point and the ipsilateral hip point can be preferred, and the measurement should be based on trunk length. Normalization:

[0058]

[0059] in, for The amount of shoulder elevation at any given moment. for The coordinates of the ipsilateral hip point in the vertical direction at any given time. for The coordinates of the shoulder point on the same side at any given time in the vertical direction. The length of the torso. This represents the current sampling time.

[0060] For the pressure center offset, the pressure center coordinates output by the pressure sensing component can be used. , ) relative to the seated reference coordinates ( , Determination of deviation:

[0061]

[0062] Subsequently, the median, mean, and standard deviation of the elbow active motion characteristics and compensatory characteristics in multiple repetitions of each standard test action can be calculated to form the elbow active motion baseline, compensatory motion baseline, and task completion baseline.

[0063] For example, if a certain feature is in The values ​​in each of the valid repetitions are respectively , ..., Then the baseline mean and fluctuation range of this feature can be taken as:

[0064]

[0065] In a preferred implementation, the capability boundary parameter can be jointly constructed by the upper bound of the acceptable fluctuation of the main motion and the lower bound of the compensation warning, wherein the upper bound of the acceptable fluctuation of the main motion can be taken as... 1.5 Or the lower bound of the compensation warning can be taken in the 10% to 90% quantile range. 1.2 to 2.0 The system outputs an individualized motion baseline model and capability boundary parameters through the above processing. This ensures that all subsequent identification and judgment are based on an individualized reference that matches the child's age, body type, affected side, and rehabilitation status, thus solving the problem that a uniform threshold is difficult to adapt to different children.

[0066] In a preferred embodiment, the stage segmentation module acquires training motion data during the training of children performing preset elbow joint rehabilitation training tasks, and performs stage segmentation processing on the training motion data to generate a sequence of motion stage labels. The training motion data includes at least elbow motion data, shoulder motion data, trunk motion data, and center of gravity change data, preferably sourced from a visual acquisition component, a pressure sensing component, and an optional posture acquisition component. Training tasks can be selected from a task library such as reaching forward to pick up an object from a table, hand to mouth, hand to head, hand to opposite shoulder, lifting a cup, and forearm rotation for picking up and placing objects, and can preset target positions, heights, distances, movement rhythms, repetition counts, and completion time limits. To improve segmentation stability, low-pass smoothing is preferably performed on the key point coordinate sequence before stage segmentation, such as using a sliding average window of 5 to 9 frames, or using first-order exponential smoothing.

[0067]

[0068] in, for Keypoint coordinates after time-smoothing for The original keypoint coordinates collected at all times. These are the smoothed coordinates of the key points from the previous sampling time. For smoothing coefficients, This is the sampling sequence number.

[0069] in, The optimal value is between 0.3 and 0.6. Subsequently, the system uses wrist displacement velocity, elbow angular velocity, trunk forward tilt rate of change, and pressure center offset rate of change as the basis for motion initiation and termination detection. Let the two-dimensional position of the wrist point be... Then the wrist velocity modulus can be written as:

[0070]

[0071] in, for The wrist speed modulus at all times, for Wrist position at all times For interval Previous wrist point position, The time interval corresponding to two consecutive velocity calculations. This represents the magnitude of the vector.

[0072] when continuous Frames greater than the start threshold And the absolute value of the rate of change of the elbow angle is continuous. Frames greater than the threshold When this happens, it can be determined that the action has entered the initiation-related phase;

[0073] Among them, the starting threshold and threshold The maximum value of the resting fluctuation state extracted from the individualized motion baseline model is preferably determined by adding a safety margin, or a preset fixed empirical threshold is used.

[0074] when continuous Frames below the termination threshold Furthermore, when the target pose has reached the task completion range, the action can be determined to have entered the termination phase. and Preferably, the frame count is 3 to 8 frames. For phase division, the interval before the start of the movement and before the major joints have shown significant movement is preferably marked as the preparation period;

[0075] The interval where wrist speed increases but has not yet entered a sustained main drive is marked as the initiation period, and the interval where the rate of change of elbow angle is consistently higher than the individualized main drive threshold and the task is mainly progressing is marked as the main drive period.

[0076] The interval in which the end approaches the target, the wrist speed decreases, and the stability of the terminal position begins to affect the result is marked as the terminal approach period;

[0077] The recovery trajectory interval after the objective is completed is marked as the pullback period. If a training task does not have a complete pullback action, the action phase label sequence is allowed to include only a portion of the preparation phase, initiation phase, main drive phase, and terminal approach phase. Through this processing, the phase segmentation module outputs an action phase label sequence with start and end times and phase names. This sequence directly serves as the time reference for subsequent feature extraction, compensation precursor state identification, and compensation initiation point determination.

[0078] In a preferred embodiment, the feature generation module extracts elbow primary motion feature parameters and shoulder-trunk compensation feature parameters based on the action phase label sequence and capability boundary parameters, and generates a compensation precursor state accordingly. For the elbow primary motion feature parameters, preferably, at least the elbow flexion-extension angle, elbow angular velocity, elbow angular acceleration, terminal extension insufficiency, and forearm pronation and supination parameters are included. Elbow angular velocity and angular acceleration can be discretized using the following formula:

[0079]

[0080] Insufficient terminal extension can be defined as the target terminal extension angle. With actual terminal angle The difference:

[0081]

[0082] If the system is equipped with a forearm posture acquisition component, the forearm pronation and supination angles can be directly calculated from the posture angles; if not, they can be indirectly estimated from the projection angle of the forearm contour direction relative to the reference plane. For shoulder-trunk compensation characteristic parameters, preferably at least the following are included: shoulder elevation, shoulder abduction, trunk forward tilt, trunk rotation, trunk lateral deviation, and pressure center offset. The shoulder abduction angle can be determined by the angle between the shoulder point, elbow point, and the trunk lateral reference axis; the trunk rotation angle can be determined by the angle between the line connecting the two shoulders and the line connecting the pelvis or the forward axis of the world coordinate system; the trunk lateral deviation can be obtained by the lateral displacement of the midpoint of the shoulder or the suprasternal notch relative to the midpoint of the pelvis and normalized according to the trunk length. To generate a pre-compensation state, the system preferably does not require compensation to have reached the formal establishment threshold, but instead first calculates the main movement deviation factor and the compensation enhancement factor. Let:

[0083]

[0084] in, This refers to a specific main elbow movement characteristic at the current stage. For the corresponding center value of the main motion baseline, To correspond to the capability boundary width; at the same time, let:

[0085]

[0086] in, For a certain compensatory characteristic, This corresponds to the center value of the compensation baseline. To correspond to the compensation warning width, a phased precursor score is then constructed:

[0087]

[0088] in, , , The weighting coefficients are non-negative, and preferably satisfy the following conditions: .when In continuous Intra-frame greater than the precursor threshold When this occurs, it can be determined that a pre-compensation state is generated at this stage. In practice, 3 to 5 frames can be taken. A value of 0.8 to 1.2 is acceptable. The reason for this approach is that compensation is often not an instantaneous change, but rather the result of the co-evolution of insufficient main movement and the gradual intervention of non-target segments. Therefore, by conducting joint analysis of the elbow main movement characteristic parameters and the shoulder-trunk compensation characteristic parameters within a stage, we can identify the precursors before the compensation is fully formed, which helps to advance the timing of control.

[0089] In a preferred embodiment, the comprehensive judgment module compares the elbow active motion characteristic parameters, shoulder-trunk compensation characteristic parameters, and pre-compensation state with the individualized motion baseline model to generate the elbow active motion contribution result, compensation type, compensation degree, and compensation initiation point. Preferably, the system calculates the elbow active motion characteristic deviation degree and compensation characteristic deviation degree for each stage, and then generates a stage-specific elbow active motion contribution index and a stage-specific compensation contribution index. Let the first... The set of elbow main motion characteristics for each stage is as follows: The shoulder-trunk compensatory feature set is Then they can be defined separately:

[0090]

[0091]

[0092] in, This refers to the sequence number of the action phase. For the first Elbow primary motion contribution index for each phase of the movement. For the first The compensatory contribution index for each action stage; This refers to the sequence number of the main elbow movement characteristic. The sequence number of the shoulder-trunk compensatory feature; For the first Normalized weights of the main motion features of the neck and elbow. For the first Normalized weights of neck-shoulder-trunk compensatory features; For the first Within the first action phase Characteristic values ​​of the main motion of the neck and elbow. For the first Within the first action phase Neck-shoulder-trunk compensatory characteristic values; For the first The center value of the baseline of the main motion characteristic of the neck and elbow. For the first The width of the main motion boundary of the main motion characteristics of the elbow; For the first The central value of the compensatory baseline for the neck-shoulder-trunk compensatory features. For the first The compensatory warning width of the neck-shoulder-trunk compensatory characteristics; For characterizing the first A function of the degree of consistency between the main elbow movement characteristics and the corresponding elbow main movement baseline. For characterizing the first The function of the degree to which the compensatory features of the neck, shoulders, and trunk exceed the corresponding compensation warning range.

[0093] in, and To normalize the weights, Used to characterize the degree of consistency between a certain main motion feature and the baseline. This is used to characterize the degree to which a certain compensatory feature exceeds the warning interval. A preferred implementation is to let:

[0094]

[0095] but The larger the value, the closer that this stage is to being completed by the elbow joint itself. The larger the value, the more significant the compensatory involvement in that stage. The total contribution of the elbow to the main movement of the entire motion can be obtained by weighted summation of each stage:

[0096]

[0097] in, For stage weighting, it is preferable to assign a larger weight to the main driving period, for example, the main driving option weight could be 0.35 to 0.50, the initiation period and the terminal approach period could each be 0.15 to 0.25, and the preparation period and the pullback period could have relatively lower weights. The compensation type can be determined by... The feature category that contributes the most to the calculation is determined. If two or more compensatory features exceed their respective thresholds at the same time, it can be marked as a concurrent compensatory mode.

[0098] The degree of compensation can be preferentially selected according to The range is divided into mild, moderate, and severe, for example, when A value of 0.30 is defined as mild compensation. A value of 0.60 is defined as moderate compensation. A value of 0.60 is defined as severe compensation. The compensation initiation point can be defined as the first time the pre-compensation state condition is met or the first time the stage compensation contribution index is met in the action phase label sequence. The stage that exceeds the corresponding threshold and the corresponding time. Through the above implementation, the comprehensive judgment module not only outputs whether compensation has been made, but also outputs who drove the completion of the compensation, what type of compensation is dominant, how many compensations there are, and when the compensation started.

[0099] In a preferred embodiment, the parameter generation module generates training quality results, compensation risk results, and training control parameters based on the elbow's primary motion contribution, compensation type, compensation degree, and compensation initiation point. The training quality results are preferably obtained through a coupled analysis of the elbow's primary motion contribution and compensation degree, rather than solely determined by whether the end-point task is completed. Therefore, a training quality score can be defined:

[0100]

[0101] in, It contributes to the main movement of the elbow in the entire motion. Contributing to the overall compensation outcome of the action. The penalty coefficient for the compensation initiation point. , , These are the weighting coefficients for the elbow's primary motion, compensatory interference, and the timing of compensatory initiation. To ensure the reasonableness of the scoring, preferably, the above weighting coefficients satisfy... + + =1, and the preferred value ranges are: [0.4, 0.6], [0.2, 0.4], and [0.1, 0.3]. Taking the larger value means that the compensation initiation point only appears near the terminal stage. Take the smaller value. Preferably, let:

[0102]

[0103] when and When the training is effective, it can be considered valid; when... However, if the task has been completed at the end, it can be considered a compensatory completion; if the task has not been completed at the end and Below the invalid threshold If this occurs, it can be considered invalid training. The outcome of compensation risk can be determined jointly by the type of compensation, the degree of compensation, and the initiation point of compensation. For example, a risk score can be constructed:

[0104]

[0105] in, To score the training quality, To effectively train the scoring threshold, The upper limit threshold for compensation contribution, The threshold for scoring invalid training; For the risk assessment of compensation, , , These are the weighting coefficients of the compensation contribution result, the compensation initiation point penalty coefficient, and the concurrent compensation mode coefficient in the risk score, respectively, preferably satisfying the following conditions. ; This is the concurrent compensation mode coefficient, used to characterize whether two or more compensation types occur simultaneously.

[0106] The higher value is used when two or more forms of compensation occur simultaneously; otherwise, the lower value is used. This is based on a risk score. The system can categorize compensation risk results into three levels: low risk, medium risk, and high risk. The optimal training control parameters are mapped from these compensation risk results: when the compensation risk result is low risk, only visual, audio, or rhythmic cues are output; when the compensation risk result is medium risk, the target distance can be shortened by 5% to 15%, the target height variation range reduced by 5% to 20%, or the required movement rhythm reduced by 10% to 20% while providing cues; when the compensation risk result is high risk, the system can further switch to a corrective training subtask, or pause scoring and reset the current task. Because the training control parameters are based on a coupled analysis of the elbow's primary movement contribution, the degree of compensation, and the compensation initiation point, this control method enables the system to move from simple detection to targeted correction and dynamic parameter tuning.

[0107] In a preferred embodiment, the closed-loop update module acquires the corresponding update action data after executing the training control parameters, and uses this update action data to update the individualized motion baseline model. The updated individualized motion baseline model then determines the updated capability boundary parameters. To ensure the reliability of the update process, the system preferably does not immediately and completely replace the existing baseline after each task, but rather employs a controlled update mechanism. Specifically, after adjusting the target distance, target height, movement rhythm, or training sub-task switching for subsequent tasks based on the training control parameters, the system re-acquires update action data and recalculates the updated elbow primary motion contribution result, the updated compensation degree, and the updated compensation initiation point in the same manner as described above. The system can only update the individualized motion baseline model when the updated elbow primary motion contribution result increases and the updated compensation degree decreases. Preferably, an exponential moving average method can be used to update each feature baseline.

[0108]

[0109] in, To optimize the update step size, a value between 0.05 and 0.20 is preferred. To update the feature representation values ​​corresponding to the action data, the corresponding capability boundary parameters can also be updated synchronously:

[0110]

[0111] in, The sequence number of the motion feature to be updated. For the updated number Baseline center value of the feature, For the previous version Baseline center value of the feature; For the updated number Item feature boundary width, For the previous version Item feature boundary width; This indicates the first action data to be updated. The representative value of the feature item and the updated value of the first item The absolute deviation between the center values ​​of the baselines of the featured items. (The superscript above...) Indicates the updated parameter, superscript This indicates the parameters before the update.

[0112] Alternatively, the mean and standard deviation can be recalculated using a new sample window. If several consecutive updates of movement data meet the update conditions, the system can appropriately widen the acceptable range for primary movement and appropriately tighten the compensation warning range to reflect the increased training requirements after the trainee's ability improves. If the updated movement data shows a decrease in the contribution of elbow primary movement or an increase in the degree of compensation, the system can maintain the existing baseline to avoid abnormal data or occasional errors contaminating the individualized reference model.

[0113] In a specific use case, assuming the trainee is a child with insufficient elbow extension and forearm rotation control, the system first records their age, affected side, upper arm length, forearm length, current rehabilitation stage, and basic range of motion test results. It also collects initial test movement data generated when the child performs standard test movements, thereby establishing an individualized motor baseline model and determining capability boundary parameters. Subsequently, the system presents the child with a tabletop object-reaching task. The visual acquisition component continuously collects key point information from the head, shoulders, elbows, wrists, and trunk, while the pressure sensing component continuously collects the center of gravity shift on the seat or support surface. The motion segmentation module divides a single object-reaching action into a preparation phase, initiation phase, main drive phase, terminal approach phase, and withdrawal phase. The feature generation module further extracts parameters such as elbow flexion / extension angle, angular velocity, forearm pronation / supination changes, as well as shoulder height changes, shoulder abduction changes, trunk forward tilt changes, trunk rotation changes, and pressure center shift. If the system detects that although the end target has been completed, the elbow's primary motor contribution is low, while the compensatory features corresponding to trunk forward leaning and shoulder elevation increase simultaneously, and both types of compensation begin to appear during the primary driving phase, the comprehensive judgment module can determine that the current action is a compensatory completion and output a high compensation risk level. Based on this, the parameter generation module generates target distance adjustment parameters and movement rhythm adjustment parameters. The closed-loop update module controls subsequent tasks to shorten the target distance and appropriately reduce the movement rhythm. Then, it re-collects and updates the action data. After confirming that the updated elbow primary motor contribution has increased and the updated compensation level has decreased, it updates the individualized movement baseline model and redetermines the updated capability boundary parameters.

Claims

1. A method for recognizing compensatory movements in children's elbow joint rehabilitation training, characterized by: include: S1. Obtain the basic information and initial movement data of the trainee child, establish an individualized movement baseline model based on the basic information and the initial movement data, and determine the ability boundary parameters by the individualized movement baseline model. The basic information is information that characterizes the individual differences and current rehabilitation status of the trainee child, and the initial movement data is the movement data generated by the trainee child when performing preset movements in the initial testing phase. S2. During the process of training children performing preset elbow joint rehabilitation training tasks, acquire training movement data, and combine the individualized motion baseline model to perform stage segmentation processing on the training movement data to obtain the movement stage label sequence. S3. Based on the action phase label sequence and capability boundary parameters, extract the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters respectively, and generate the compensation precursor state based on the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters. S4. Compare the elbow main motion characteristic parameters, shoulder-trunk compensation characteristic parameters and compensation precursor state with the individualized motion baseline model to obtain the elbow main motion contribution results, compensation type, compensation degree and compensation initiation point. S5. Based on the elbow main motion contribution results, compensation type, compensation degree and compensation initiation point, generate training quality results and compensation risk results, and generate training control parameters based on the training quality results and compensation risk results. S6. Control the preset elbow joint rehabilitation training task based on the training control parameters, obtain the corresponding updated movement data, update the individualized motion baseline model based on the updated movement data, and then determine the updated capability boundary parameters by the updated individualized motion baseline model for closed-loop control of the subsequent preset elbow joint rehabilitation training task.

2. The method for recognizing compensatory movements in children's elbow joint rehabilitation training according to claim 1, characterized in that: The basic information includes at least one of the following: age, affected side, upper arm length, forearm length, joint range of motion, and current rehabilitation stage; the initial movement data includes the initial test movement data generated when the trained child performs standard test movements. The individualized motion baseline model is established, and the capability boundary parameters are determined by the individualized motion baseline model, including: normalizing the initial motion data based on the basic information to generate the elbow main motion baseline, the compensatory motion baseline and the task completion baseline; According to the characteristic correspondence under the same preset elbow joint rehabilitation training task and the same movement stage, the elbow main movement feature in the elbow main movement baseline is mapped to the shoulder-trunk compensatory feature in the compensatory movement baseline; the center value of the main movement baseline and the width of the main movement boundary of each elbow main movement feature are determined, as well as the center value of the compensatory baseline and the compensation warning width of each shoulder-trunk compensatory feature; the center value of the main movement baseline and the width of the main movement boundary are used to form the current acceptable main movement deviation range, and the center value of the compensatory baseline and the compensation warning width are used to form the compensation warning range, and the current acceptable main movement deviation range and the compensation warning range are determined as the capability boundary parameters.

3. The method for recognizing compensatory movements in children's elbow joint rehabilitation training according to claim 2, characterized in that: The training motion data is segmented into stages to obtain the motion stage label sequence, including: motion initiation detection and motion termination detection based on changes in elbow movement trajectory, wrist displacement, trunk posture, and center of gravity shift. Between the action start detection result and the action end detection result, the training action data is divided into at least some of the stages of preparation period, start period, main drive period, terminal approach period and withdrawal period according to the action start characteristics, main drive characteristics, terminal approach characteristics and withdrawal characteristics, and the action stage label sequence corresponding to the start and end times of each stage is generated.

4. The method for recognizing compensatory movements in children's elbow joint rehabilitation training according to claim 3, characterized in that: Based on the action phase label sequence and the capability boundary parameters, the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters are extracted respectively, and the compensation precursor state is generated, including: according to the action phase label sequence, extracting the elbow main motion feature parameters representing elbow flexion and extension, elbow angular velocity, insufficient terminal extension and forearm pronation and supination in each phase, and extracting the shoulder-trunk compensation feature parameters representing shoulder elevation, shoulder abduction, trunk forward tilt, trunk rotation, trunk lateral tilt and center of gravity shift; The elbow primary motion characteristic parameters and the shoulder-trunk compensation characteristic parameters of each stage are matched with the capability boundary parameters. When the elbow primary motion characteristic parameters shift outward from the capability boundary parameters and the shoulder-trunk compensation characteristic parameters show a continuous increasing trend, the corresponding stage's pre-compensation state is generated.

5. The method for recognizing compensatory movements in children's elbow joint rehabilitation training according to claim 4, characterized in that: The elbow primary motion feature parameters, the shoulder-trunk compensation feature parameters, and the pre-compensation state are compared with the individualized motion baseline model to obtain the elbow primary motion contribution result, the compensation type, the compensation degree, and the compensation initiation point. This includes: according to the action stage label sequence, comparing the elbow primary motion feature parameters of each stage with the elbow primary motion baseline corresponding to the same stage to obtain the elbow primary motion deviation degree, and generating the elbow primary motion contribution result of the corresponding stage based on the elbow primary motion deviation degree, wherein the smaller the elbow primary motion deviation degree, the higher the elbow primary motion contribution result of the corresponding stage; The shoulder-trunk compensation characteristic parameters of each stage are compared with the compensation motion baseline of the same stage to obtain the shoulder-trunk compensation deviation degree. The compensation degree of the corresponding stage is generated according to the shoulder-trunk compensation deviation degree. The greater the shoulder-trunk compensation deviation degree, the higher the compensation degree of the corresponding stage. The elbow active motion contribution and compensation degree at each stage are weighted and summarized to obtain the elbow active motion contribution and compensation degree for the entire training movement. Based on the shoulder-trunk compensation feature category that exceeds the corresponding compensation warning range and has the largest deviation degree, the compensation type is determined; and the stage and corresponding time in the action stage label sequence that first meets the conditions for the establishment of the compensation precursor state are determined as the compensation initiation point.

6. The method for recognizing compensatory movements in children's elbow joint rehabilitation training according to claim 5, characterized in that: Based on the elbow active motion contribution result, the compensation type, the compensation degree, and the compensation initiation point, the training quality result and the compensation risk result are generated, and the training control parameters are generated based on the training quality result and the compensation risk result, including: judging the current training action as one of effective training, compensatory completion, or ineffective training according to the combination relationship between the elbow active motion contribution result and the compensation degree. The compensation risk outcome is determined based on the compensation type, the compensation degree, and the compensation initiation point; The training control parameters are generated based on the training quality results and the compensation risk results. The training control parameters include at least one of the following: target distance adjustment parameters, target height adjustment parameters, movement rhythm adjustment parameters, and training subtask switching parameters.

7. The method for recognizing compensatory movements in children's elbow joint rehabilitation training according to claim 6, characterized in that: The preset elbow joint rehabilitation training task is controlled based on the training control parameters, the corresponding updated movement data is obtained, the individualized motion baseline model is updated based on the updated movement data, and the updated capability boundary parameters are determined by the updated individualized motion baseline model. This includes: after controlling the preset elbow joint rehabilitation training task, obtaining the updated movement data corresponding to the training control parameters. Based on the updated motion data, the elbow primary motion contribution result, the degree of compensation, and the compensation initiation point are regenerated. When the updated elbow primary motion contribution result increases and the updated compensation level decreases, the individualized motion baseline model is updated using the updated motion data, and the updated capability boundary parameters are determined by the updated individualized motion baseline model for closed-loop control of the subsequent preset elbow joint rehabilitation training task.

8. A system for recognizing compensatory movements in pediatric elbow joint rehabilitation training, used to implement the method for recognizing compensatory movements in pediatric elbow joint rehabilitation training as described in any one of claims 1-7, characterized in that: include: The basic modeling module is used to acquire basic information and initial movement data of the children being trained, establish an individualized movement baseline model based on the basic information and initial movement data, and determine the ability boundary parameters by the individualized movement baseline model. The phase segmentation module is used to acquire training movement data during the process of training children performing preset elbow joint rehabilitation training tasks, and to perform phase segmentation processing on the training movement data in combination with an individualized motion baseline model to obtain a sequence of movement phase labels. The feature generation module is used to extract elbow main motion feature parameters and shoulder-trunk compensation feature parameters based on the action phase label sequence and capability boundary parameters, and to generate compensation precursor states based on the elbow main motion feature parameters and shoulder-trunk compensation feature parameters. The comprehensive judgment module is used to compare the elbow main motion characteristic parameters, shoulder-trunk compensation characteristic parameters and compensation precursor state with the individualized motion baseline model to obtain the elbow main motion contribution result, compensation type, compensation degree and compensation initiation point. The parameter generation module is used to generate training quality results and compensation risk results based on the elbow main motion contribution results, compensation type, compensation degree and compensation initiation point, and to generate training control parameters based on the training quality results and compensation risk results. The closed-loop update module is used to control the preset elbow joint rehabilitation training task based on the training control parameters, obtain the corresponding updated movement data, update the individualized motion baseline model based on the updated movement data, and then determine the updated capability boundary parameters by the updated individualized motion baseline model for closed-loop control of subsequent preset elbow joint rehabilitation training tasks.

9. The pediatric elbow joint rehabilitation training compensatory movement recognition system according to claim 8, characterized in that: The basic modeling module is used to normalize the initial movement data based on at least one of the following: the child's age, affected side, upper arm length, forearm length, joint range of motion, and current rehabilitation stage. This generates an elbow primary movement baseline, a compensatory movement baseline, and a task completion baseline. The module also determines the capability boundary parameters based on the correspondence between the elbow primary movement baseline and the compensatory movement baseline. The stage segmentation module is used to perform movement initiation and termination detection on the training movement data based on changes in elbow movement trajectory, wrist displacement, trunk posture, and center of gravity shift. Between the movement initiation detection results and the movement termination detection results, a sequence of movement stage labels is generated according to movement initiation features, primary driving features, terminal approach features, and withdrawal features. The feature generation module is used to generate the elbow main motion feature parameters and the shoulder-trunk compensation feature parameters based on the action phase label sequence and the capability boundary parameters, and to generate the compensation precursor state when the elbow main motion feature parameters shift outward from the capability boundary parameters and the shoulder-trunk compensation feature parameters show a continuous increasing trend.

10. The pediatric elbow joint rehabilitation training compensatory movement recognition system according to claim 9, characterized in that: The comprehensive judgment module is used to calculate the deviation of the elbow main motion feature parameters from the elbow main motion baseline and the deviation of the shoulder-trunk compensation feature parameters from the compensation motion baseline, respectively, to generate the elbow main motion contribution result and the compensation degree, and to determine the compensation type according to the feature category corresponding to the deviation degree, and to determine the compensation initiation point according to the stage in the action stage label sequence that first meets the conditions for the establishment of the compensation precursor state and the corresponding time. The parameter generation module is used to generate the training quality result based on the combination relationship between the elbow main motion contribution result and the compensation degree, generate the compensation risk result based on the compensation type, the compensation degree and the compensation initiation point, and generate the training control parameters based on the compensation risk result. The closed-loop update module is used to obtain the updated action data corresponding to the training control parameters after controlling the preset elbow joint rehabilitation training task based on the training control parameters, and regenerate the elbow main movement contribution result, the compensation degree and the compensation initiation point based on the updated action data. When the updated elbow main movement contribution result increases and the updated compensation degree decreases, the individualized motion baseline model is updated, and the updated ability boundary parameters are determined by the updated individualized motion baseline model.

Images