Intelligent rehabilitation rod-based action standard trajectory comparison and scoring method and system and storage medium

Abstract

The invention discloses an action standard trajectory comparison and scoring method and system based on an intelligent rehabilitation rod and a storage medium, and belongs to the technical field of medical rehabilitation and data processing. The method comprises the following steps: acquiring an actually measured joint angle track sequence corresponding to one rehabilitation training action completed by a user; providing a standard joint angle track template corresponding to the rehabilitation training action; adopting a dynamic time warping algorithm to calculate an optimal matching path between the actually measured joint angle track sequence and the standard joint angle track template, and accumulating along the path to obtain an integral track difference value; and converting the overall track difference value into a quantitative score representing the motion completion quality through a preset score mapping function. According to the method, the technical problem that a traditional evaluation method can only perform binary judgment and lacks quantitative evaluation is solved, and the interference of the action rate difference of the patient is overcome by introducing dynamic time warping.

Description

Technical Field

[0001] This invention relates to the field of medical rehabilitation and information processing technology, and in particular to a method, system and storage medium for comparing and scoring standard movement trajectories based on intelligent rehabilitation bars. Background Technology

[0002] In the field of physical therapy and rehabilitation medicine, upper limb dysfunction is a common sequela for patients with stroke, nerve injury, or orthopedic surgery. Rehabilitation training is the core means of restoring or improving upper limb function, and its effectiveness largely depends on the standardization, accuracy, and repeatability of the training movements. In traditional rehabilitation processes, the assessment of movement quality mainly relies on the therapist's visual observation and experience judgment. This method is highly subjective, difficult to quantify, and cannot achieve continuous, real-time monitoring of patients, especially in home rehabilitation settings.

[0003] To achieve objective and quantitative assessment of rehabilitation training, existing technologies have proposed several solutions. One approach employs wearable sensor-based motion capture systems, directly measuring joint angles by having the patient wear an inertial measurement unit (IMU) on their shoulder, elbow, or wrist. However, these solutions typically require patients to wear complex devices, making configuration cumbersome. This not only increases equipment costs but also significantly impacts patient comfort and compliance, making them unsuitable for frequent, routine rehabilitation training.

[0004] Another set of solutions attempts to use non-contact visual capture technology, such as depth cameras. However, this approach is highly sensitive to ambient lighting, background complexity, and occlusion, and its deployment cost is high, making it difficult to use in the ever-changing home environment. Some solutions utilize rehabilitation devices with integrated sensors (such as dumbbells and resistance bands), but their analytical capabilities are usually limited to the movement trajectory of the device itself, such as determining whether the device follows a preset spatial path. These methods only perform a simple "device-trajectory" comparison, which has significant technical limitations: First, it cannot accurately map the movement of the device to the body's own biomechanical parameters (such as joint angles), thus failing to determine whether the patient is "tricking" the system to complete the task through incorrect compensatory movements such as shrugging or twisting the torso; second, its evaluation dimensions are singular, typically only providing a binary "pass / fail" judgment, unable to continuously and quantitatively score quality indicators such as the smoothness and coordination of the movement, and even less able to address the problem of varying movement speeds due to individual patient differences, easily leading to misjudgments.

[0005] Therefore, there is an urgent need in this field for a technical solution that does not require patients to wear additional sensors, is simple to deploy, can accurately deduce human posture from the motion data of rehabilitation equipment, detect non-standard compensatory movements in real time, and provide an objective quantitative score for the overall quality of the movement. This solution would address the technical problems of existing assessment methods, such as inconvenience in use, limited analytical dimensions, inability to effectively identify compensation, and insufficient quantitative assessment capabilities. Summary of the Invention

[0006] The first aspect of this application provides a method, system, and storage medium for comparing and scoring standard movement trajectories based on a smart rehabilitation bar, aiming to solve the technical problem that traditional assessment methods can only make binary judgments and lack continuous, multi-dimensional quantitative assessment of movement quality.

[0007] According to a third aspect of this application, a method for comparing and scoring standard movement trajectories based on a smart rehabilitation bar is provided, comprising: acquiring a measured joint angle trajectory sequence corresponding to a user completing a rehabilitation training movement; providing a standard joint angle trajectory template corresponding to the rehabilitation training movement, consisting of multiple joint angle dimensions; employing a dynamic time warping algorithm to calculate an optimal matching path between the measured joint angle trajectory sequence and the standard joint angle trajectory template, and accumulating along the path to obtain an overall trajectory difference value; and converting the overall trajectory difference value into a quantitative score characterizing the quality of movement completion through a preset scoring mapping function.

[0008] Optionally, the local distance metric in the dynamic time warping algorithm is defined as the weighted squared Euclidean distance between the measured joint angle trajectory sequence and the standard joint angle trajectory template across all joint angle dimensions at the same matching time.

[0009] Optionally, in the weighted squared Euclidean distance, the weight coefficient assigned to the angle difference of the core training joint is higher than the weight coefficient assigned to the angle difference of the auxiliary joint.

[0010] Optionally, the dynamic time warping algorithm further includes a phase matching supplementation step to handle the problem of inconsistent starting phases of periodic actions.

[0011] Optionally, the scoring mapping function is an exponentially decaying function.

[0012] Optionally, the method further includes: during the rehabilitation training exercise, detecting the user's postural compensatory movements in real time; and using the number or severity of the detected compensatory movements as a penalty item to correct the quantified score.

[0013] Optionally, obtaining the measured joint angle trajectory sequence includes: obtaining the measured joint angle trajectory sequence by performing inverse kinematics solution on the pose data of a smart rehabilitation bar.

[0014] Secondly, this application also provides a motion standard trajectory comparison and scoring system based on a smart rehabilitation bar, comprising: a trajectory acquisition module configured to acquire a measured joint angle trajectory sequence corresponding to a user completing a rehabilitation training movement; a template storage module configured to provide a standard joint angle trajectory template corresponding to the rehabilitation training movement, consisting of multiple joint angle dimensions; a difference calculation module configured to use a dynamic time warping algorithm to calculate an optimal matching path between the measured joint angle trajectory sequence and the standard joint angle trajectory template, and accumulate along the path to obtain an overall trajectory difference value; and a scoring generation module configured to convert the overall trajectory difference value into a quantitative score characterizing the quality of movement completion through a preset scoring mapping function.

[0015] Optionally, the system further includes: a compensation detection module configured to detect the user's postural compensation movements in real time during the rehabilitation training movements; and the scoring generation module is further configured to use the number or severity of the detected compensation movements as a penalty item to correct the quantified score.

[0016] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of the first aspects.

[0017] This application effectively addresses the issue of varying movement rates due to individual patient differences by introducing a Dynamic Time Warping (DTW) algorithm, achieving precise comparison of movement trajectory patterns. This method integrates multi-dimensional trajectory differences into an objective, continuous (e.g., 0-100 points) quantitative score, comprehensively reflecting the accuracy, coordination, and stability of the movement. This not only provides patients with more refined training feedback but also offers therapists reliable and traceable data support for tracking rehabilitation progress and adjusting treatment plans. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of a human posture inversion method provided according to an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of the coordinate system and vector relationship for human posture inversion according to an embodiment of this application.

[0020] Figure 3 This is a structural block diagram of a human posture inversion system provided according to an embodiment of this application.

[0021] Figure 4 This is an overall flowchart of a human posture compensation detection method provided according to an embodiment of this application.

[0022] Figure 5 This is a structural block diagram of a human posture compensation detection system provided according to an embodiment of this application.

[0023] Figure 6 This is an overall flowchart of a motion standard trajectory comparison and scoring method provided according to an embodiment of this application.

[0024] Figure 7 This is a structural block diagram of a motion standard trajectory comparison and scoring system provided according to an embodiment of this application.

[0025] Figure 8 This is a schematic diagram of a two-dimensional planar chain kinematic model provided according to an embodiment of this application, used to illustrate the geometric principle of inverse kinematics solution.

[0026] Figure 9 This is a schematic diagram of a compensation strength function model provided according to an embodiment of this application, used to illustrate the weighted fusion process of multi-dimensional errors.

[0027] Figure 10 This is a schematic diagram of the Dynamic Time Warping (DTW) algorithm provided according to an embodiment of this application, used to illustrate the alignment method of unequal time length sequences. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0030] Example 1 This embodiment provides a technical solution for human motion analysis based on a smart rehabilitation bar. The solution includes human posture inversion, real-time compensation detection, and motion quantification scoring. The steps of this solution will be described below.

[0031] S100: Obtain the joint angle sequence representing the posture of the human upper limb. like Figure 1 As shown, in this embodiment, the system first executes S100, the purpose of which is to accurately inversely calculate the real-time joint angles of the user's upper limbs (mainly including the shoulder, elbow, and wrist joints) from the physical motion data of the smart rehabilitation stick held by the user. This step is the data foundation for all subsequent analyses (such as compensation detection and movement scoring), and its core lies in solving how to achieve a precise mapping from the device posture to the human posture without directly contacting or observing the human limbs, thereby overcoming the inconvenience brought by traditional wearable devices. The execution of this step makes the rehabilitation assessment process for the user merely a simple handheld device training session, while the system completes the reconstruction of complex human biomechanical parameters in the background. To achieve this goal, S100 includes the fusion processing of sensor data, the unified construction of the coordinate system, the precise determination of the hand end pose, and the final inverse kinematics solution.

[0032] To ensure the accuracy of subsequent posture inversion, this approach performs an initial calibration step of personalized parameters before formal rehabilitation training begins. This step addresses the issue of unclear sources for key parameters in the human model (such as limb length and shoulder joint reference position), enabling the model to adapt to individual differences among different users.

[0033] S105: Personalized parameter calibration In S105, when a new user uses the system for the first time, or when the system detects a user change, a guided interactive calibration process is initiated. This process requires no assistance from external professionals; the user simply needs to complete a few simple specified actions according to on-screen or voice prompts.

[0034] In a specific implementation, the calibration process includes the following sub-steps: First, the system will prompt the user to enter their height value. (For example, 1.75 meters), this value will serve as a benchmark for subsequent human proportion model estimations.

[0035] Secondly, perform shoulder joint baseline positioning. The system will guide the user to stand in a standard, natural posture via screen animation or voice commands, with feet shoulder-width apart, body upright, arms hanging naturally at the sides, holding one end of the smart rehabilitation stick in each hand or holding it in the middle with one hand. Once the user confirms that they are ready, the system will trigger the S110's data acquisition module to record the world coordinate system position of the center point of the rehabilitation stick at that moment. The system integrates a proportional model based on anthropometry data. For example, this model presets the height of the shoulder joint center. Approximately 82% of the user's height, that is The horizontal position (Y-axis coordinate) of the center of the shoulder joint on one side can be estimated based on the distance between the hands when both arms are hanging naturally, or an empirical value related to height can be used directly, such as assuming that the shoulder joint's offset relative to the body's centerline in the Y-axis direction is... . The X-axis coordinate (front-back direction) is usually the same as the X-axis coordinate of the hand in a natural standing posture. Maintaining consistency. Thus, the system estimates the reference position of the shoulder joint in the world coordinate system. .

[0036] Next, the length of the upper limb segments was measured. (upper arm length) and (Forearm length) calibration. The system guides the user to perform an "L" shaped posture: raise the upper arm of the training side horizontally to the side (shoulder joint abduction 90 degrees), keeping the upper arm parallel to the ground; simultaneously, bend the elbow joint 90 degrees, so that the forearm points vertically upward. The user holds this posture for several seconds, and the system records the world coordinates of the hand's grip center at this moment. In this specific posture, there are clear geometric constraints: the elbow joint height It must be equal to the height of the shoulder joint. Upper arm length This is the projected distance between the shoulder joint and the elbow joint on the XY plane. Forearm length That is, the vertical distance between the hand and the elbow joint. Hand position and elbow position The projections onto the XY plane coincide, that is... and Therefore, the system can directly calculate the limb length: as well as .

[0037] For example, let's assume the user's height Meters. During shoulder point calibration, the system recorded... The system calculates the shoulder height. Meters, shoulder Y coordinate Meters. Therefore, the shoulder joint reference point is defined as: Next, during limb length calibration, the user assumed an "L" shaped posture, and the system recorded the hand position. The system calculates based on this: upper arm length rice.

[0038] Forearm length rice.

[0039] At this point, personalized parameters All parameters have been calibrated. The system associates and stores these parameters with the user ID, providing personalized model input for the subsequent inverse kinematics solution of S130.

[0040] S110: Acquire real-time pose data of the intelligent rehabilitation stick in the world coordinate system. The pose data includes position information and attitude information.

[0041] In S110, the system's data acquisition module is responsible for collecting and processing raw sensor data from the smart rehabilitation stick to generate its real-time pose (position and attitude) data in three-dimensional space. The smart rehabilitation stick integrates a nine-axis inertial measurement unit (IMU) and one or more optical or infrared markers for external positioning at the hardware level.

[0042] An inertial measurement unit (IMU) can output readings from a triaxial accelerometer and a triaxial gyroscope at high frequencies (e.g., 100 Hz). The gyroscope measures the angular velocity of the rehabilitation bar in its local coordinate system; by integrating this velocity over time, relatively accurate attitude change information can be obtained in the short term. However, due to the bias and noise inherent in the sensors, long-term integration inevitably leads to cumulative attitude drift. The accelerometer measures the gravitational acceleration vector under static or slow motion conditions. This vector is constant in the world coordinate system and can therefore be used to correct attitude drift caused by gyroscope integration, particularly regarding roll and pitch angles on the horizontal plane. However, when the rehabilitation bar undergoes vigorous movement, the accelerometer measures the resultant vector of gravitational acceleration and motion acceleration. In this case, the gravitational information cannot be accurately separated, leading to errors in attitude correction.

[0043] To address the issues of long-term attitude drift and uncertain position inherent in pure IMU solutions, this embodiment introduces an external positioning system. This system can be one or more infrared or optical cameras fixed in the environment, used to track actively emitting or passively reflecting markers on the rehabilitation bar. This system can directly measure the three-dimensional coordinates of the markers in a world coordinate system (typically constructed with the positioning system base station as the origin) at a relatively low frequency (e.g., 60Hz), thus providing position information without accumulated errors.

[0044] After acquiring high-frequency IMU data and relatively low-frequency external location data, the system employs a sensor fusion algorithm to optimally estimate and fuse the data from these two sources. In this embodiment, an extended Kalman filter (EKF) is preferably used to achieve this goal. The state vector of this EKF... Defined as: ,in It is the three-dimensional position vector of the rehabilitation stick in the world coordinate system. It is its velocity vector. It is a quaternion representing the rotation from the local coordinate system of the rehabilitation rod to the world coordinate system. and These are the bias vectors of the accelerometer and gyroscope, respectively. The EKF prediction step utilizes high-frequency data from the IMU for state prediction. Specifically, it predicts the state based on the previous moment's state and the current IMU measurement (angular velocity). and acceleration The IMU predicts the current position, velocity, attitude, and bias by integrating the kinematic equations. When new external position measurement data arrives, the EKF performs an update step. This update step calculates the residual (information) between the predicted position and the externally measured position, and uses this residual, based on the Kalman gain, to correct all components in the state vector, including position, velocity, attitude, and sensor bias. In this way, the high-frequency dynamic information from the IMU is organically combined with the low-frequency absolute position information from the external positioning system. The resulting pose data has both high dynamic response and no long-term drift, thus providing a stable and reliable input for subsequent calculations.

[0045] For example, suppose at time... The EKF prediction step is complete, and the predicted state is obtained. At this moment, the external positioning system transmitted a new location measurement. The EKF update process is as follows: First, calculate the measurement residuals. ,in This is the measurement matrix, used to extract the position components from the state vector. Then, the Kalman gain is calculated. ,in It is the covariance matrix of the predicted state. It is the covariance matrix of the measurement noise. A typical example is... The matrix may be set as a diagonal matrix. ,in , , This represents the measurement variance of the positioning system on each axis. Finally, the state vector is updated. and state covariance matrix After the update, the quaternions in the state vector... This means that the rehabilitation stick is in The precise attitude after fusion correction at any given moment, and the position vector This represents its precise location. This process is repeated each time a positioning system data point arrives, thus continuously outputting a high-precision pose data stream.

[0046] S120: Determine the end-effector pose of the user's hand in the world coordinate system based on the posture data of the rehabilitation bar and a preset grip point model.

[0047] In S120, the system uses the precise pose data of the rehabilitation bar obtained in S110 to further calculate the end-effector pose of the user's hand (specifically, the center of the wrist joint or the equivalent center of hand grip) in the world coordinate system. The pose data of the rehabilitation stick itself is about its geometric center or the location of the IMU, while inverse kinematics requires the pose of the end effector (i.e., the hand) of the human kinetic chain. Therefore, a transformation from the center of the device to the center of the hand is needed.

[0048] This transformation relies on a pre-defined grip point model. This model defines a fixed translation vector of the user's hand grip center relative to the origin of the local coordinate system of the rehabilitation stick. . Reference Figure 2 The local coordinate system of the rehabilitation bar It can be defined as: the origin is located at the geometric center of the rehabilitation bar, and its... The shaft runs along the length of the rod. and The axis perpendicular to it forms a right-handed coordinate system. Translation vector This describes the position of the center of the wrist relative to the origin when the user holds the device in a standard manner. This vector can be pre-calibrated or dynamically determined based on grip markings on the rehabilitation bar (such as grip positions identified by capacitive sensors or Hall sensor arrays).

[0049] The specific calculation process for determining the hand's end-effector pose is as follows: First, obtain the rehabilitation stick from S110 at any time. The pose, including its position in the world coordinate system Position vector in And represent from the local coordinate system To the world coordinate system Rotation matrix of rotation The rotation matrix can be directly derived from the attitude quaternions. Obtained through conversion.

[0050] Then, the gripping point translation vector in the local coordinate system is... Transform to world coordinates. This transformation is achieved by left-multiplying by a rotation matrix: This new vector This represents the offset vector from the center of the rehabilitation stick to the center of the hand in the world coordinate system.

[0051] Finally, by adding the world coordinates of the center of the rehabilitation stick to the offset vector in this world coordinate system, we can obtain the world coordinates of the hand, that is, the position components of the end-effector pose: The directional component of the hand's end-effector pose is usually aligned with the direction of the rehabilitation bar or has a fixed rotational relationship. In most upper limb motion analyses, the position component is the primary input for inverse kinematics solutions.

[0052] For example, assume that the origin of the local coordinate system of the rehabilitation bar is located at its center, and its length is... Meters. The user holds it with their right hand at a distance from the center. At a distance of meters. Therefore, the translation vector in the gripping point model can be defined as... (Unit: meters). At a certain moment The system obtains the pose of the center of the rehabilitation stick through EKF fusion as follows: Position Posture Quaternion First, convert the quaternion into a rotation matrix. For this quaternion, the corresponding rotation matrix is ​​(rotating 90 degrees around the Y-axis): Next, calculate the offset vector in the world coordinate system: Finally, calculate the world coordinates of the hand: Therefore, the system determines that at that moment, the position of the user's hand tip is... This precise hand position will serve as the input for the next step of inverse kinematics solving.

[0053] S130: Based on the human upper limb kinematics model, an inverse kinematics solution algorithm is used, and a joint angle sequence representing the human upper limb posture is obtained by inverse calculation according to the hand end pose and a preset shoulder joint position.

[0054] In S130, the system performs the core inverse kinematics (IK) calculations. This step utilizes the hand end-effector pose obtained in S120. Based on preset human body parameters, the angle values ​​of each joint that drives the upper limb to reach the end pose are calculated in reverse, forming a continuous sequence of joint angles. .

[0055] This calculation is based on a human upper limb kinematic model. This model abstracts the upper limb as an open-chain or closed-loop kinematic chain composed of multiple rigid links (upper arm, forearm) connected by joints (shoulder joint, elbow joint). The key parameter of the model is the segment length, i.e., the upper arm length. (Distance from shoulder joint to elbow joint) and forearm length (Distance from the elbow joint to the center of the hand grip). These parameters can be demographic averages or personalized for specific users through a simple calibration procedure upon first use. Additionally, a fixed reference point is required, namely the position of the shoulder joint in a world coordinate system. This position can also be determined before training begins through a simple calibration action (for example, having the user stand naturally, recording the hand position at this time, and estimating the shoulder position based on the height model).

[0056] Depending on whether the action being analyzed is in a plane or in three-dimensional space, models and solution algorithms of different complexities can be selected.

[0057] In a preferred embodiment, for situations where many rehabilitation movements (such as shoulder abduction and flexion) primarily occur within a specific plane (such as the coronal or sagittal plane), the system employs a two-dimensional planar chain model. The geometric relationships of this model are as follows: Figure 8 As shown in the diagram. In this model, the shoulder joint, elbow joint, and hand form a triangle, with the three sides being the upper arm length L1, forearm length L2, and shoulder-to-hand distance ld, respectively. In this case, the inverse kinematics problem has an analytical solution. The solution process is as follows: Calculate the vector from the shoulder point to the hand point: .

[0058] Calculate the magnitude of this vector, which is the straight-line distance from the shoulder to the hand: .

[0059] Solving the elbow flexion-extension angle using the law of cosines In a company , and In the triangle formed, there are Therefore, the intermediate variables can be solved. The elbow angle is then... The choice of positive or negative sign corresponds to the two possible postures of the elbow joint (flexion or hyperextension), and the solution of elbow flexion is usually chosen based on human kinematic constraints.

[0060] Solving for shoulder joint angles This angle consists of two parts: one part is the shoulder-hand vector. Angle with the coordinate axis The other part is the upper arm. Shoulder-hand vector The angle between . This can also be solved using the Law of Cosines or the Law of Sines. The final shoulder joint angle is .

[0061] In another embodiment, to handle more complex spatial motions, the system employs a three-dimensional spatial chain model. This model typically has redundant degrees of freedom (e.g., the shoulder joint has 3 degrees of freedom, and the elbow joint has 1-2 degrees of freedom), resulting in the IK problem not having a unique analytical solution. In this case, the system uses an optimization-based iterative solution algorithm, such as the Levenberg-Marquardt (LM) algorithm. The goal of this algorithm is to find a set of joint angles. This allows the model's hand position to be calculated using forward kinematics (FK) from these angles. Compared with the actual measured hand position The error between them is minimized. Its objective function can be defined as: The first term is the position error term. The second term is a regularization term, used to increase the stability of the solution and make it more human-like. It is the regularization weight. It is a reference posture (e.g., the solution from the previous moment or a natural, relaxed posture). It is a diagonal weight matrix used to weight the importance of different joints (for example, the weight of the shoulder joint can be set higher than that of the elbow joint). The LM algorithm updates the joint angles by iteratively calculating the Jacobian matrix until the error converges to a sufficiently small value or the maximum number of iterations is reached.

[0062] For example, a two-dimensional planar chain model is used for the calculation. Assume the user's limb length is... rice, Meters. The shoulder joint position is fixed at... At a certain moment, S120 calculates the position of the hand's end point as follows: A two-dimensional analysis is performed on the projection onto the XZ plane, with the shoulder point being... , hand point .

[0063] Calculate shoulder and hand vectors: .

[0064] Calculate the distance between the shoulder and hand: rice.

[0065] Calculate the intermediate variables of the elbow angle: .

[0066] Solve for the elbow angle (choose the elbow flexion solution): This is a reasonable elbow flexion angle.

[0067] Solve for the shoulder angle: The final shoulder angle is... .

[0068] By repeating this calculation at each time step, the system generates a continuous sequence of joint angles. These sequences accurately reflect the changes in the user's upper limb posture throughout the entire movement.

[0069] S200: Real-time detection of posture compensation movements during training. like Figure 4 As shown, after S100 converts the movement of the rehabilitation bar into a high-fidelity sequence of human joint angles, the system then executes S200. This step utilizes this real-time generated biomechanical data to instantly detect and quantify any non-standard compensatory movements that may occur during rehabilitation training. The goal of rehabilitation training is to exercise specific target muscle groups and joints, while compensatory movements (such as tilting the body to assist in raising the arm) shift the training load to non-target areas, not only reducing the rehabilitation effect but also potentially causing secondary injuries. Therefore, S200 constructs a multi-dimensional compensatory intensity function to integrate various subtle erroneous movement patterns into a quantifiable risk indicator, thereby solving the key technical problem that traditional assessment methods cannot effectively distinguish between normal movement variations and pathological compensation.

[0070] Before performing real-time compensation detection, the system requires a standard for comparison, namely a standard action template. The quality of this template directly determines the accuracy of the angle offset calculation in subsequent S220. This step details the offline construction and online loading process of this template.

[0071] S205: Construction and Loading of Standard Action Templates The standard motion template is not a single trajectory, but a statistical model that includes the average trajectory and the normal range of variation. Its construction process is an offline, data-driven process.

[0072] In a specific implementation, the template construction process is as follows: First, standard movement data is collected. The system developers will invite several (e.g., 5) senior physical therapists with extensive clinical experience as demonstrators. Each therapist will wear the same smart rehabilitation bar device used by the patient and, under the synchronous monitoring of a high-precision optical motion capture system (as ground truth), repeatedly perform a specific rehabilitation movement (e.g., "shoulder abduction to 90 degrees in the coronal plane and return"). Each therapist will repeat this movement 20 times. Using the S100's posture inversion method, the system can generate a corresponding, multi-dimensional joint angle trajectory sequence for each demonstration. ,in .

[0073] Secondly, data alignment and statistical modeling were performed. Due to subtle differences in the speed and rhythm of the actions performed by different presenters, the first step was to align all 100 trajectory sequences in time. The system employed a Dynamic Time Warping (DTW) algorithm, selecting the highest-quality trajectory as a baseline and non-linearly aligning the remaining 99 trajectories to this baseline, ensuring that all trajectories remained synchronized at feature points (such as start, peak, and end). After alignment, all trajectories were resampled to the same length (e.g., 101 phase points, corresponding to...). Next, the system employs functional data analysis methods at each phase point. Above, calculate the mean vector of joint angles for all 100 trajectories at that point. Covariance Matrix .

[0074] Finally, the standard templates are defined and stored. The standard action templates are ultimately defined and stored as a data structure. This structure must contain at least: Average trajectory sequence This represents the most ideal and standard execution paradigm for this action.

[0075] Normal variation range: This is typically defined as a confidence band. For example, at each phase point... The upper and lower boundaries of the confidence band can be defined as follows: The confidence band is the mean plus or minus 1.5 times the standard deviation. Any measured movement falling within this confidence band can be considered normal, non-pathological movement variation.

[0076] During rehabilitation training, the system loads the corresponding standard movement templates from storage based on the selected training program. When calculating the angular offset in S220, the system not only compares the measured angle with the average trajectory but also determines whether it falls within the normal variation range (confidence band). If it falls within the band, the angular offset can be considered 0 or multiplied by an attenuation coefficient less than 1, thereby improving the specificity of compensation detection and avoiding misjudgment of normal movement fluctuations.

[0077] S210: Estimate the user's real-time torso pose parameters In S210, the system needs to estimate the user's trunk posture because abnormal trunk movements are one of the most common forms of upper limb compensation. For example, when shoulder abduction is insufficient, the patient may unconsciously tilt their trunk to the opposite side to trick their hand into reaching the target height. This embodiment proposes a method to infer trunk tilt based solely on the wrist joint trajectory and shoulder joint reference point already present in S100, avoiding the need for additional sensors (such as a chest strap IMU).

[0078] This method is based on the following kinematic principle: in proper upper limb training movements, the trunk should remain neutral and stable, therefore the position of the shoulder joint... The position should remain essentially unchanged in the world coordinate system. At this point, the wrist joint's trajectory is primarily contributed by the rotation of the shoulder and elbow joints. However, once trunk tilt occurs, the trunk will swing around the hip joint like an inverted pendulum, which will affect the shoulder joint position. Translation in the world coordinate system, thus causing the trajectory of the wrist joint. An abnormal horizontal displacement has occurred.

[0079] Based on this principle, the system calculates the trunk tilt angle. The process is as follows: Obtain real-time wrist joint trajectory points and the preset shoulder joint reference point .

[0080] Calculate the vertical distance difference between the wrist joint and the shoulder joint. Difference in horizontal (left and right) distance .

[0081] The torso tilt angle is defined as the angle between the vector formed by these two distance differences and the vertical axis: .

[0082] When the torso remains upright The value should be very small (ideally 0). It is also close to 0. When the torso tilts to one side, It will increase significantly, leading to Deviation from 0. This angle visually quantifies the degree of torso tilt.

[0083] For example, suppose the shoulder joint reference point is... In an ideal shoulder abduction movement, the wrist joint may move to... ,at this time , If the patient tilts their trunk to achieve the same hand position, their actual shoulder joint position may shift to a different location. However, to compensate, the actual position of the wrist joint may change. At this time, the system observed for The system calculates based on this: This value deviates significantly from 0, clearly indicating severe lateral tilting of the torso. This effectively captures the primary torso compensation patterns without requiring additional hardware.

[0084] It is important to emphasize that the method proposed in this embodiment for inferring trunk tilt solely from the wrist joint trajectory is an effective approximation under the premise of minimal hardware (no wearable devices). This method has high accuracy and sensitivity in most planar upper limb rehabilitation movements (such as shoulder abduction in the coronal plane and flexion in the sagittal plane), because in these movements, abnormal lateral trunk movements are the main compensatory mode and are directly and significantly reflected in the horizontal trajectory of the wrist joint.

[0085] However, the accuracy of this simplified model may decrease for rehabilitation movements involving complex three-dimensional motion or requiring significant trunk rotation (e.g., diagonal movements in PNF patterns). To address such complex scenarios and further improve the system's versatility and robustness, this application also provides an optional enhancement scheme. In this enhancement scheme, the system can support data communication with a miniature wireless IMU sensor worn on the user's sternum. This chest-strap IMU can directly and accurately measure the rotation matrix or quaternion of the trunk's own coordinate system. When training complex movements, the system can fuse the direct measurement results from the chest-strap IMU to correct or replace trunk posture parameters indirectly derived from wrist joint trajectories. For example, the trunk roll angle measured by the chest-strap IMU can be used as... The truth value of the system. This flexible architecture of basic (non-wearable) + enhanced (single-point wearable) allows the system to strike a balance between convenience and high precision, adapting to more diverse clinical needs.

[0086] S220: Calculate the angular offset of each joint. In S220, the system will use the real-time joint angle sequence generated in S100. With the preset standard action template Compare to calculate the angular offset of each joint. Here It is the normalized action phase.

[0087] First, time alignment is required. Since the patient's movement speed may differ from the standard template, directly comparing angles at the same timestamp is meaningless. The system employs simple linear time normalization: assuming the total duration of a complete movement is... So at time The action phase is calculated as Then, the system uses this phase value... The corresponding reference joint angle is found in the standard motion template through linear interpolation. .

[0088] Then, calculate the angular offset: .

[0089] This offset It quantified the patient's first [action / phase] at the same stage of the action. The difference between the actual angle and the standard angle of a joint. For example, a positive value may indicate excessive joint movement, while a negative value indicates insufficient joint movement.

[0090] For example, suppose in a standard shoulder abduction movement template, when the movement is halfway through ( The shoulder abduction angle should be: The total time a patient takes to perform this action. Seconds. Seconds, its phase The system obtains the measured shoulder joint angle at that moment from S100. The angular offset of the joint is then: .

[0091] This means that during this phase of the movement, the patient's shoulder abduction angle was less than the standard. This offset is one of the important inputs for subsequent calculations of compensation strength.

[0092] S230: Generate real-time compensation strength values ​​using the compensation strength function. In S230, the system fuses the torso posture parameters and joint angle offsets obtained in the first two steps, and then uses the log intensity function. This generates a single, quantified real-time compensation strength value. The function aims to integrate errors from different sources and in different units into a unified indicator to reflect the overall compensation risk.

[0093] In this embodiment, the compensation strength function is defined as a weighted summation, and its model structure is as follows: Figure 9 As shown: in: and These are the weighting coefficients for each item, and Generally, trunk compensation has the greatest impact on training effectiveness, therefore It is given the highest weight.

[0094] This is the trunk error term, derived from the trunk tilt angle in S210. The result is obtained through calculation. To make it dimensionless and comparable, it is defined as: ,in It is a maximum permissible trunk tilt angle (e.g.) ).

[0095] It is the first The angle error term for each joint is determined by the angle offset in S220. The calculation yielded this result. To avoid penalizing normal, minor fluctuations in movement, a "tolerance band" is introduced into this error term. Its definition is: ,in This is the maximum permissible error angle for the joint. This means that only when the absolute value of the angle deviation exceeds the tolerance zone... Only the excess portion will be included in the compensation intensity.

[0096] For example, continuing with the previous example, let's set the parameters as follows: weight , , Maximum permissible error: rad, rad. Tolerance zone: rad.

[0097] At some point The system calculates that: Trunk tilt angle: rad.

[0098] Shoulder joint angle deviation: rad.

[0099] Elbow joint angle deviation (assuming it is): (Within the tolerance zone).

[0100] Calculate each error term: .

[0101] .

[0102] (because (Not exceeding the tolerance zone).

[0103] Calculate the final compensation strength value: .

[0104] This value (typically ranging from 0 to 1) quantifies the overall severity of compensation at the current moment.

[0105] It should be further explained that the weighting coefficients in the compensation strength function Maximum permissible error (e.g.) and joint error tolerance zone The values ​​of these hyperparameters are crucial to the sensitivity and specificity of the detection. In addition to expert-set values ​​based on clinical experience, this application provides a data-driven systematic optimization method to address the difficulty in determining these parameters.

[0106] In a preferred embodiment, the parameters can be determined through an offline machine learning process. First, a labeled training dataset is constructed. This dataset contains a large amount of complete joint angle trajectory data of patients performing various rehabilitation movements. Simultaneously, at least two senior physical therapists review this data via video, annotating frame by frame the "time segments" where compensation clearly occurs and their "compensation types" (such as trunk tilting, shoulder shrugging, etc.).

[0107] Then, the compensation detection problem is transformed into a supervised classification problem. For each time point in the training data, the system calculates a series of candidate features (including trunk tilt angle, joint angle offset, etc.) based on the joint angles. The compensation strength function itself can be viewed as a linear model that maps these features to a single risk value. The optimization goal is to find an optimal set of parameters (…). (etc.), so that the compensation strength function calculated The data showed significantly higher values ​​than the threshold during time segments labeled "compensated" by the therapist, and lower values ​​than the threshold during "normal" time segments. This optimization process can employ standard machine learning algorithms, such as using logistic regression to learn the weights. Alternatively, grid search combined with cross-validation can be used to find the optimal combination of thresholds and tolerance bands. This data-driven approach allows for the customization of an optimal set of clinically validated compensatory testing parameters for each rehabilitation exercise, greatly improving the robustness and clinical applicability of the approach.

[0108] S240: Making a judgment on compensation and providing feedback In S240, the system calculates the real-time compensation strength value based on S230. The final decision is made. To avoid false alarms caused by transient noise or accidental motion fluctuations, the decision logic introduces dual thresholds for time and amplitude.

[0109] The system sets a compensation trigger threshold. (For example, ) and a duration threshold (For example, (seconds). The judgment logic is as follows: The system continuously monitors... .once The value exceeds The system starts timing. If in the following... Within a time period, The value has remained at If the above is true, then the system will ultimately determine that a valid compensation action has occurred.

[0110] Once a case is determined to be compensation, the system will immediately trigger a feedback mechanism. This may include: Visual feedback: Display a red warning light on the screen, or highlight the main source of compensation (through analysis). The error term that contributes the most to the calculation is used to determine this.

[0111] Auditory feedback: Play prompts or voice prompts, such as "Please keep your body upright" or "Relax your shoulders and do not shrug".

[0112] Tactile feedback: If the rehabilitation stick integrates a vibration motor, it can alert the user through vibration.

[0113] This immediate and targeted feedback helps patients recognize and correct their incorrect movements, thereby greatly improving the quality and efficiency of rehabilitation training.

[0114] S300: Perform trajectory comparison and quantitative scoring of training movements. like Figure 6 As shown, after a user completes a full rehabilitation movement, the system executes S300. The purpose of this step is to provide a comprehensive and objective quantitative score for the quality of the entire movement process. This solves the technical problem that traditional assessment methods can only make rough pass / fail judgments and cannot reflect subtle differences in the degree of movement completion. The core of S300 is to use the Dynamic Time Warping (DTW) algorithm to nonlinearly compare the complete measured joint angle trajectory sequence generated by S100 with a standard template, thereby calculating a comprehensive score that reflects the similarity of trajectory morphology.

[0115] S310: Preprocessing the measured joint angle trajectory sequence In S310, the system first processes the recorded sequence of measured joint angle trajectories for the complete motion. Preprocessing is performed to eliminate noise and make it suitable for comparison.

[0116] Preprocessing typically includes: Filtering and smoothing: The angle sequence of each joint is processed using, for example, a 5-point moving average filter or a low-pass filter to eliminate glitches caused by sensor noise or tiny muscle tremors, making the trajectory smoother.

[0117] Time normalization: The timeline of the entire action sequence is from Linear mapping to normalized phase axis This allows action sequences of different durations to be compared within the same logical timeframe, which is a prerequisite step for the DTW algorithm.

[0118] S320: Calculates the overall trajectory difference value using a dynamic time warping algorithm. In the S320, the system performs the core calculations of the DTW algorithm. DTW is an algorithm that finds the optimal matching path between two time series of different lengths. It finds the alignment that makes the two sequences most similar in shape by non-linearly "stretching" or "compressing" the time axis. The basic principle of the DTW algorithm is as follows: Figure 10 As shown.

[0119] like Figure 10 As shown, the calculation process of DTW is as follows: Constructing the cost matrix: Let the normalized measured sequence be The standard template sequence is .in, and These are all n-dimensional joint angle vectors. The system constructs a... Cost matrix , of which elements Represents the first of the measured sequences The nth point and the template sequence The "local distance" between points. This distance is defined as the weighted squared Euclidean distance between the multidimensional joint angle vectors: in, It is the first The weights of each joint are used to highlight the importance of the joints in core training. For example, in shoulder abduction training, the weight of the shoulder joint... It will be set to be larger than the elbow joint. high.

[0120] Finding the optimal regular path: The DTW algorithm uses dynamic programming to optimize the cost matrix. Find a path starting from the bottom left corner. Go to the top right corner path Points on the path The algorithm satisfies the constraints of monotonicity and continuity. The goal is to find a path that minimizes the sum of the local distances of all points along that path. This minimum cumulative distance is the DTW distance between the two sequences, i.e., the overall trajectory difference value. .

[0121] this The value precisely quantifies the overall difference in morphology between the measured motion trajectory and the standard template. The smaller the value, the more standard the motion.

[0122] For example, consider a simplified single-joint trajectory alignment. Measured sequence Standard template .

[0123] Constructing a cost matrix (value is) ): By filling the cumulative cost matrix using dynamic programming, the optimal path found may be... 12 and Align 14 in the middle, 15 and Align with 15 in the middle, 13 and Align the 11 in the middle, 11 and Aligned with 11 in the middle.

[0124] By accumulating costs along this path, we obtain To avoid unfairness caused by different path lengths, path length is usually used. The cumulative cost is normalized to obtain the average variance value. Assuming the calculated (unit: ).

[0125] S330: Convert difference values ​​into quantified scores In S330, the system will use the overall trajectory difference value calculated by S320, which is meaningful to technicians but not intuitive to users. It is converted into a quantitative score, such as 0-100, that is easy for users to understand through a rating mapping function.

[0126] This embodiment uses an exponential decay function as the scoring mapping function: in, It is the average trajectory difference value. It is a baseline error energy parameter used to adjust the sensitivity of the score. The value can be obtained through statistical analysis of a large number of standard motion samples. When the motion is perfect, , With differences in movement As the value increases, the score will decrease non-linearly and rapidly.

[0127] In addition, the number of compensatory actions detected in S200 can also be... Introduced as a penalty item to further refine the scoring: in It is the penalty points for each compensation payment.

[0128] For example, continuing the previous example, suppose the average difference value is calculated. Let the reference error energy be... Furthermore, during this operation, the S200 detected one compensatory action, resulting in a penalty. .

[0129] Basic rating: .

[0130] Final rating: .

[0131] The system will eventually show the user feedback "Score: 73 points" and may include an error heatmap, visually indicating which stage and joint of the movement had the greatest deviation, providing specific improvement guidance for the user's next training session.

[0132] Example 2 This application also provides a human posture inversion system 100, which is configured to perform the method described in S100 and its sub-steps (S105 to S130) of the aforementioned embodiment. See also... Figure 3 The system 100 can be an integrated intelligent rehabilitation device or a distributed system consisting of an intelligent rehabilitation bar and a back-end computer. In the figure, solid arrows represent data volume, and dashed double arrows represent control flow.

[0133] System 100 includes: A data acquisition module 110, whose physical carrier is a smart rehabilitation stick and an optional external positioning system, is responsible for collecting raw sensor data (IMU data and position data) from the rehabilitation stick and executing sensor fusion algorithms (such as EKF) through an internal processor, ultimately outputting a high-precision real-time pose data stream.

[0134] An end-effector pose determination module 115, typically implemented in software on a backend computer, receives pose data from the data acquisition module 110 and performs calculations in S120. This module internally stores a grip point model, and its function is to determine the end-effector pose of the user's hand in the world coordinate system based on the pose of the rehabilitation stick and the grip point model.

[0135] A pose calculation module 120, also implemented in software, receives the hand end-effector pose from the end-effector pose determination module 115 and performs the calculation in step S130. This module internally stores personalized human upper limb kinematic model parameters (such as segment lengths and shoulder joint positions calibrated in step S105) and inverse kinematics algorithms. Its function is to inversely solve the hand end-effector pose into a multi-dimensional sequence of joint angles, and use this sequence as the final output of the system for subsequent applications (such as compensation detection or motion scoring).

[0136] A storage module 160 is used to store various data required for system operation, including but not limited to: user's personal information and calibrated limb parameters, kinematic models, etc. A processor 170 serves as the system's computing core, coordinating and executing the functions of all the aforementioned modules.

[0137] Example 3 This application also provides a human posture compensation detection system, which is configured to perform the method described in S200 and its sub-steps (S205 to S240) of the aforementioned embodiment. See also... Figure 5 System 200 is typically implemented in software on a backend computing device (such as a PC, tablet, or cloud server) and interacts with an upstream system that provides motion data (such as the attitude inversion system mentioned above). Solid arrows in the diagram represent data flow, and dashed arrows represent control flow. System 200 includes: A motion data interface 210 is configured to receive real-time joint angle sequences and wrist trajectories representing the user's upper limb movements from an external data source.

[0138] A trunk posture estimation module 220 performs the calculations in S210, receives the wrist joint trajectory and a preset shoulder joint reference point, and estimates the user's real-time trunk posture parameters (such as tilt angle) through geometric relationships. An angle offset calculation module 230 performs the calculations in S220, aligns the real-time joint angle sequence with a standard motion template loaded from storage in time, and calculates the angle offset of each joint.

[0139] A compensation strength calculation module 240 performs the calculation of S230, receives the torso posture parameters and angle offset, and performs a weighted combination through a preset compensation strength function to generate a quantified real-time compensation strength value.

[0140] A compensation decision module 250 executes the logic of S240, compares the compensation intensity value with preset time and amplitude thresholds, and finally determines whether effective compensation has occurred, and outputs an alarm signal when compensation occurs.

[0141] A storage module 260 stores a standard action template library, compensation detection parameters, and historical compensation records. A processor 270 coordinates and executes the functions of all the aforementioned modules.

[0142] Example 4 This application also provides a motion standard trajectory comparison and scoring system 300 based on a smart rehabilitation bar. This system 300 is configured to execute the method described in S300 and its sub-steps (S310 to S330) of the aforementioned embodiment. (Refer to...) Figure 7 System 300 is typically activated after a complete rehabilitation exercise and is implemented in software on a backend computing device. In the diagram, solid arrows represent data flow, and dashed arrows represent control flow. System 300 includes: A trajectory acquisition module 310 is configured to acquire the complete measured joint angle trajectory sequence corresponding to a user completing one rehabilitation training movement. This module is also responsible for the preprocessing steps described in S310.

[0143] A template storage module 320 stores a library of standard joint angle trajectory templates corresponding to various rehabilitation training movements. Based on the user's training program, this module provides the corresponding standard templates to the difference calculation module.

[0144] A difference calculation module 330 performs the calculation of S320, uses the dynamic time warping (DTW) algorithm to perform nonlinear comparison between the measured trajectory and the standard template, and calculates the overall trajectory difference value that can reflect the morphological differences between the two.

[0145] A score generation module 340 performs the calculations of S330, receiving the overall trajectory difference value from the difference calculation module and converting it into a user-friendly quantitative score of 0-100 using a preset score mapping function (such as an exponential decay function). Optionally, this module can also receive externally input compensation frequency information to correct the final score.

[0146] A storage module 360 ​​is used to store historical rating data, analysis reports, etc., and a processor 370 coordinates and executes the functions of all the above modules.

[0147] It should be noted that, in order to clearly illustrate the different technical solutions provided in this application, this specification provides multiple embodiments (such as Embodiment 2, Embodiment 3, and Embodiment 4) and corresponding multiple accompanying drawings (such as...). Figure 3 , Figure 5 , Figure 7 In these different embodiments and figures, components or modules with the same or similar functions may appear, but are given different reference numerals. For example, the "processor" for performing computational coordination functions is... Figure 3 It may be labeled as 370, while Figure 5 It may be marked as 270.

[0148] Those skilled in the art will understand that these different reference numerals are merely for maintaining internal clarity and consistency in their respective figures and embodiment descriptions, and do not imply any structural or essential differences in these functionally identical components. Therefore, in the context of this application, unless otherwise specified, functional modules with the same or similar names (such as "processor," "storage module," "data acquisition module," etc.), regardless of their reference numerals, should be understood to refer to technical entities that perform the corresponding functions and have the same or equivalent structure and principles. The differences in reference numerals should not be construed as limiting the scope of the claims.

[0149] Example 5 This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement any of the method steps in Embodiment 1. The storage medium can be non-volatile, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), or it can be volatile, such as random access memory (RAM).

[0150] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0151] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0152] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for comparing and scoring standard movement trajectories based on a smart rehabilitation bar, characterized in that, include: Obtain the measured joint angle trajectory sequence corresponding to a user completing one rehabilitation training movement; Provide a standard joint angle trajectory template consisting of multiple joint angle dimensions, corresponding to the rehabilitation training movements described above; A dynamic time warping algorithm is used to calculate an optimal matching path between the measured joint angle trajectory sequence and the standard joint angle trajectory template, and an overall trajectory difference value is obtained by accumulating along the path. The overall trajectory difference value is converted into a quantitative score representing the quality of action completion through a preset scoring mapping function.

2. The method according to claim 1, characterized in that, The local distance metric in the dynamic time warping algorithm is defined as the weighted squared Euclidean distance between the measured joint angle trajectory sequence and the standard joint angle trajectory template across all joint angle dimensions at the same matching time.

3. The method according to claim 2, characterized in that, In the weighted squared Euclidean distance, the weight coefficients assigned to the angular differences of the core training joints are higher than the weight coefficients assigned to the angular differences of the auxiliary joints.

4. The method according to claim 1, characterized in that, The dynamic time warping algorithm also includes a phase matching supplementation step to handle the problem of inconsistent starting phases of periodic actions.

5. The method according to claim 1, characterized in that, The scoring mapping function is an exponential decay function.

6. The method according to claim 1, characterized in that, The method further includes: During the rehabilitation training exercises, the user's postural compensatory movements are detected in real time. The number or severity of detected compensatory actions is used as a penalty to adjust the quantified score.

7. The method according to claim 1, characterized in that, The process of obtaining the measured joint angle trajectory sequence includes: The measured joint angle trajectory sequence is obtained by performing inverse kinematics on the pose data of an intelligent rehabilitation bar.

8. A motion standard trajectory comparison and scoring system based on an intelligent rehabilitation bar, characterized in that, include: A trajectory acquisition module is configured to acquire the measured joint angle trajectory sequence corresponding to a user completing a rehabilitation training movement; A template storage module is configured to provide a standard joint angle trajectory template consisting of multiple joint angle dimensions, corresponding to the rehabilitation training movements. A difference calculation module is configured to use a dynamic time warping algorithm to calculate an optimal matching path between the measured joint angle trajectory sequence and the standard joint angle trajectory template, and accumulate along the path to obtain an overall trajectory difference value. A scoring generation module is configured to convert the overall trajectory difference value into a quantitative score representing the quality of action completion through a preset scoring mapping function.

9. The system according to claim 8, characterized in that, The system also includes: A compensatory detection module is configured to detect the user's postural compensatory movements in real time during the rehabilitation training exercises; Furthermore, the scoring generation module is also configured to use the number or severity of detected compensatory actions as a penalty item to correct the quantified score.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images