Lower limb rehabilitation training method, system and device based on kinematics-virtual scene bidirectional mapping and medium

Abstract

The present application relates to the technical field of lower limb rehabilitation training, in particular to a lower limb rehabilitation training method, system, device and medium based on kinematics-virtual scene bidirectional mapping. Kinematics data and physiological signal data of the lower limbs of a patient are collected and time-synchronized to obtain training characteristic data; a training state evaluation model is used to identify the training adaptation state of the patient based on the training characteristic data, and a training regulation strategy is generated; avatar posture data in a virtual scene is converted from the kinematics data based on a space mapping model, and virtual scene task parameters are determined in combination with the regulation strategy, based on which the scene content is updated and presented to the patient; mechanical execution parameters are converted from the virtual scene task parameters based on the space mapping model, based on which the lower limbs of the patient are adjusted. The present application establishes a real-time correspondence between the virtual scene and the actual movement of the lower limbs of the patient through bidirectional mapping, and dynamically adjusts the training intensity in combination with the training state, thereby improving the adaptive safety of the training process.

Description

Technical Field

[0001] This invention relates to the field of lower limb rehabilitation training technology, and in particular to lower limb rehabilitation training methods, systems, equipment and media based on kinematic-virtual scene bidirectional mapping. Background Technology

[0002] With the deepening of population aging and the rising incidence of neurological diseases such as stroke and spinal cord injury, the number of people with lower limb motor dysfunction is increasing year by year, and the demand for lower limb rehabilitation training continues to grow. At present, lower limb rehabilitation training mainly relies on therapists to provide one-on-one movement guidance to patients, and helps patients restore lower limb motor function through repeated mechanical movement training. With the development of virtual reality (VR) technology, some lower limb rehabilitation training equipment has begun to try to introduce virtual reality visual feedback, replacing the monotonous training environment with virtual scene images to enhance patients' willingness to participate in the training process.

[0003] However, existing lower limb rehabilitation training methods still have the following shortcomings: First, the traditional rehabilitation training process is boring and tedious, patients are not very enthusiastic about participating, and there is a lack of objective quantitative assessment methods for training intensity and training effect, making it difficult to judge whether the training has achieved the expected results. Secondly, most existing virtual reality rehabilitation training devices only use virtual reality as a simple form of screen display. There is a lack of real-time and accurate correspondence between the content presented in the virtual scene and the actual movement of the patient's lower limbs. This can easily lead to a lack of synchronization between the visual image and the limb movement, which in turn can cause discomfort to the patient. Third, existing training methods are usually conducted according to a pre-set fixed intensity, failing to dynamically adjust the training intensity based on changes in the patient's physical condition during training. This results in training intensity that is too high, causing excessive burden on the patient, or training intensity that is too low, making it difficult to achieve effective training results. The personalization and adaptability of training programs need to be improved. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by this invention is how to address the issues in existing training methods, such as the lack of objective quantitative evaluation of training effects, the lack of precise real-time correspondence between the content presented in the virtual scene and the actual movement of the patient's lower limbs, and the failure to dynamically adjust the training intensity in accordance with the patient's training status.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping, which includes collecting patient lower limb kinematic data through a motion sensing unit, collecting patient physiological signal data through a physiological signal acquisition unit, and synchronizing the two in time to obtain training feature data; Based on the training feature data, the patient's current training adaptation status is identified through the training status assessment model, and corresponding training regulation strategies are generated. Based on the spatial mapping model, the kinematic data is converted into avatar posture data in the virtual scene, and the virtual scene task parameters are determined in combination with the training and control strategy. The virtual scene generation unit updates the scene content accordingly and presents it to the patient. Based on the spatial mapping model, the virtual scene task parameters are converted into execution parameters of the mechanical execution unit, which then provides assistance or resistance adjustment to the patient's lower limbs accordingly.

[0007] As a preferred embodiment of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described in this invention, the training adaptation state includes three categories: insufficient training intensity, appropriate training intensity, and excessive training load. The training status assessment model identifies the patient's current training adaptation status category based on the preset interval combination of the training feature data, and evaluates the confidence level of the identification results. If the confidence level is not lower than the preset threshold, the training control strategy is generated according to the preset strategy rule corresponding to the training adaptation state category; if the confidence level is lower than the preset threshold, the training control strategy generated last time is used.

[0008] As a preferred embodiment of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described in this invention, the spatial mapping model transforms the kinematic data into initial body posture data through coordinate transformation; and the initial body posture data is proportionally adapted according to the patient's body size information. Based on the changing trend of the kinematic data, it is determined whether there is a posture deviation. If there is a deviation, the adapted avatar posture data is predicted and compensated to obtain the avatar posture data that is synchronized with the patient's lower limb movement in real time. If there is no deviation, the adapted avatar posture data will be used directly as the avatar posture data.

[0009] As a preferred embodiment of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described in this invention, the spatial mapping model obtains the adjustment direction and adjustment range of the execution parameters of the mechanical execution unit according to the scene difficulty level corresponding to the virtual scene task parameters; The adjusted execution parameters are compared with the previous execution parameters. If the difference exceeds the preset safety threshold, the execution parameters are output in stages. If the difference does not exceed the preset safety threshold, the execution parameters are output directly. When a preset abnormal change is detected in the patient's physiological signal data, a control instruction is generated to interrupt the current training, and the execution priority of the control instruction is higher than the adjustment of the execution parameters.

[0010] As a preferred embodiment of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described in this invention, the kinematic data includes at least one of the patient's lower limb joint angle data and spatial pose data, and the physiological signal data includes at least one of the patient's lower limb muscle electromyography signal, heart rate-related signal and skin conductance response signal. If the sampled values ​​in the kinematic data or physiological signal data exceed a preset reasonable range, the transmission of the corresponding data to subsequent steps will be paused, and data compensation or a prompt to re-collect data will be performed. If the data does not exceed a preset reasonable range, the kinematic data and the physiological signal data will be used for time-series synchronization processing.

[0011] As a preferred embodiment of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described in this invention, the training feature data includes movement completion feature representing the patient's degree of movement completion, fatigue feature representing the patient's degree of physical fatigue, and participation feature representing the patient's degree of training engagement. If the coefficient of variation of the training feature data is within a preset stable range, then the training feature data is used to identify the training adaptation state. If the coefficient of variation of the training feature data continues to exceed the preset stable range, the generation of the training control strategy will be paused, and the current training control strategy will be maintained.

[0012] As a preferred embodiment of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described in this invention, the virtual scene content includes scene difficulty elements and interactive feedback elements corresponding to the training task parameters, and the updated virtual scene content is output to the patient in an immersive visual form. Based on the difference in the patient's motor completion characteristics before and after the update, it is determined whether the update of the virtual scene content has a positive impact on the patient's training performance; If a positive impact is generated, the corresponding training control strategy will be recorded as a valid case. If no positive impact is generated, the training control strategy will not be recorded.

[0013] Secondly, embodiments of the present invention provide a lower limb rehabilitation training system based on kinematic-virtual scene bidirectional mapping, which includes a motion sensing unit for collecting kinematic data of the patient's lower limbs; Physiological signal acquisition unit, used to acquire the patient's physiological signal data; The main control unit is used to perform time-series synchronization processing on the kinematic data and the physiological signal data to obtain training feature data, identify the patient's current training adaptation state and generate training regulation strategies through a training state evaluation model, and is also used to obtain avatar posture data in the virtual scene based on the kinematic data through a spatial mapping model, determine virtual scene task parameters based on the training regulation strategies, and obtain execution parameters of the mechanical execution unit based on the virtual scene task parameters. A virtual scene generation unit is used to update the virtual scene content based on the avatar posture data and the virtual scene task parameters and present it to the patient. A mechanical actuator is used to provide assistance or resistance adjustment to the patient's lower limbs according to the execution parameters.

[0014] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the lower limb rehabilitation training method of kinematic-virtual scene bidirectional mapping as described in the first aspect of the present invention.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the lower limb rehabilitation training method with kinematic-virtual scene bidirectional mapping as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: The simultaneous acquisition of kinematic and physiological signals and extraction of training features at the front end provide a quantitative basis for judging training status, integrating motor performance and physiological state, overcoming the problem of training effects relying on subjective observation and being difficult to objectively evaluate; the intermediate training status assessment identifies the patient's training adaptation state based on training features and generates control strategies, enabling dynamic adjustment of training intensity according to the patient's state, avoiding mismatch between fixed intensity plans and the patient's actual state; the spatial mapping stage establishes a real-time correspondence between the virtual scene and the patient's actual lower limb movements through bidirectional conversion between kinematic data and avatar posture, scene task parameters and mechanical execution parameters, reducing discomfort caused by asynchrony between visual feedback and limb movements; the terminal mechanical execution and safety monitoring stage, while physically implementing training difficulty, monitors abnormal physiological signals and promptly interrupts training to ensure training safety; each stage is sequentially connected and operates cyclically, forming an overall rehabilitation training program with intensity matching the patient's state and scene corresponding to limb movements. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for a lower limb rehabilitation training method using a two-way mapping of kinematics and virtual scene; Figure 2 A computer device diagram for a lower limb rehabilitation training method using a two-way mapping between kinematics and virtual scenes; Figure 3 A schematic diagram of the process for judging the stability of training feature data in a lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping; Figure 4 A schematic diagram of the mechanical execution parameter adjustment and safety monitoring and judgment process for a lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Example 1 Reference Figure 1 This is the first embodiment of the present invention, which provides a lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping, including: S100: Collects the patient's lower limb kinematic data through the motion sensing unit and the patient's physiological signal data through the physiological signal acquisition unit, and synchronizes the two in time to obtain training feature data.

[0023] S200: Based on training feature data, the system identifies the patient's current training adaptation status through a training status assessment model and generates corresponding training regulation strategies.

[0024] S300: Based on a spatial mapping model, kinematic data is converted into avatar posture data in a virtual scene, and the virtual scene task parameters are determined by combining training and control strategies. The virtual scene generation unit then updates the scene content and presents it to the patient accordingly.

[0025] S400: Based on a spatial mapping model, virtual scene task parameters are converted into execution parameters of the mechanical execution unit, which then provides assistance or resistance adjustment to the patient's lower limbs.

[0026] It should be noted that during lower limb rehabilitation training, the range of motion, speed, fatigue, and tension of the patient's lower limbs will continuously change with the training process. If the virtual scene simply follows the patient's lower limb kinematic data without combining training characteristic data to identify the patient's current training adaptation state, it is easy for the virtual scene to be out of sync with the patient's actual movement and for the training intensity to be mismatched with the patient's current physical state. This can lead to discomfort for the patient and affect the progress of the training. At the same time, if a corresponding conversion relationship is not established between the virtual scene task parameters and the mechanical execution parameters, the training difficulty presented in the virtual scene will not be reflected in the actual assistance or resistance experienced by the patient's lower limbs, resulting in a disconnect between the virtual scene and the actual training process of the patient's lower limbs.

[0027] Therefore, by collecting and synchronizing patients' kinematic and physiological signal data to obtain training feature data, identifying the patient's current training adaptation state, and generating training regulation strategies, the kinematic data is converted into avatar posture data in the virtual scene based on a spatial mapping model. Combined with the training regulation strategies, the virtual scene task parameters are determined to update the scene content. Then, the virtual scene task parameters are converted into mechanical execution parameters to adjust the patient's lower limbs. This achieves real-time correspondence between the virtual scene content and the patient's actual lower limb movements, as well as dynamic adjustment of training intensity according to the patient's training state.

[0028] Example 2 Reference Figures 1-4 This is the second embodiment of the present invention.

[0029] In this embodiment, step S100 involves collecting the patient's lower limb kinematic data through the motion sensing unit and the patient's physiological signal data through the physiological signal acquisition unit, and synchronizing the two in time to obtain training feature data, including the following steps A1-A2: A1: Kinematic data includes at least one of the patient's lower limb joint angle data and spatial pose data, and physiological signal data includes at least one of the patient's lower limb muscle electromyography signal, heart rate-related signal, and skin conductance response signal.

[0030] Specifically, before formally acquiring kinematic and physiological signal data, initialization and parameter calibration are first performed to establish the baseline working state upon which subsequent data acquisition and training regulation depend. The initialization process includes three steps: device self-test, zero-point calibration, and compatibility identification.

[0031] During the equipment self-test, the communication links of each functional link are polled with heartbeat packets. Each link returns a "ready" or "faulty" flag according to a preset cycle. If there is a functional link that is not ready, the process will pause and proceed to the next link, and the corresponding fault location will be indicated, so as to ensure that the kinematic data and physiological signal data collected later come from the normally functioning acquisition channel.

[0032] During the zero-point calibration process, joint angle and torque baseline data are continuously collected for 60 seconds while the patient remains stationary, either standing or sitting. Zero drift values ​​are calculated and automatically compensated. Simultaneously, electromyographic signals are collected for 60 seconds at rest, and their root mean square baseline values ​​are calculated. Heart rate baseline and skin conductance response baseline are also collected. These baseline values ​​serve as reference benchmarks in the subsequent training feature data extraction process to eliminate the impact of individual differences and sensor zero-point drift on the accuracy of feature calculation.

[0033] In the adaptation recognition stage, based on the patient's input of height, weight, affected side information, and rehabilitation stage information, the system matches the initial posture parameters and the proportion and viewing angle of the avatar in the virtual scene from the preset database. After the above calibration is completed, an initial working state ready flag is generated, and the formal acquisition process of kinematic data and physiological signal data begins.

[0034] After initialization and calibration, kinematic and physiological signal data are acquired in parallel at a synchronous sampling frequency of no less than 100Hz.

[0035] The acquisition of kinematic data specifically includes: acquiring joint angle sequences of the hip, knee, and ankle joints through angle sensors installed at the axis of key joints; acquiring the pose and angular velocity of the lower limb linkage in space through an inertial measurement unit; acquiring joint interaction torque through a torque sensor installed at the joint drive output end; and acquiring gait phase information and center of gravity offset through a pressure distribution sensor integrated into the sole of the foot.

[0036] The acquisition of physiological signal data specifically includes: acquiring electromyographic signals of the main muscle groups of the lower limbs through a surface electromyography sensor; extracting heart rate and time-domain indices of heart rate variability through a heart rate acquisition module; and acquiring skin conductivity and its time-varying slope through a skin conductance response sensor.

[0037] A2: Training feature data should include at least the following: exercise completion features, which characterize the patient's level of exercise completion; fatigue features, which characterize the patient's level of physical fatigue; and participation features, which characterize the patient's level of training engagement.

[0038] Specifically, refer to Figure 3 After obtaining valid kinematic and physiological signal data, the two are first processed for time synchronization: linear interpolation is used to resample the data of each channel to a unified time axis, and the cross-correlation delay between kinematic and physiological signals is calculated. If the delay exceeds the preset range, the time deviation is compensated to obtain multi-source data with consistent time reference.

[0039] Based on this, training feature data for characterizing patient training performance is extracted from the synchronized data, specifically including four categories: exercise completion features, fatigue features, participation features, and load features.

[0040] Among them, the movement completion feature is obtained by temporally aligning and comparing the actual joint angle trajectory with the preset reference gait template, and is used to characterize the degree of matching between the patient's lower limb movement trajectory and the target trajectory; the fatigue feature is obtained by performing frequency domain transformation on the electromyographic signal and extracting the trend of the signal frequency feature quantity over time. The more obvious the decrease in the trend, the higher the degree of fatigue accumulation.

[0041] Engagement features are obtained by fusing heart rate variability features with the phase synchronization of skin conductance response, and are used to characterize the patient’s level of engagement during training; load features are obtained by comprehensively processing the ratio of heart rate to the preset physiological upper limit and respiratory rate, and are used to characterize the patient’s current physiological load level.

[0042] The aforementioned training feature data are continuously generated at preset time intervals and stored in a circular queue with a fixed duration. When the queue is full, the moving average and coefficient of variation are calculated for each feature value. If the coefficient of variation is within a preset stable range, the training feature data is determined to be in a stable state, and the next step of training adaptation state identification is initiated. If the coefficient of variation continues to exceed the preset range, the training feature data is determined to be in a fluctuating state, and subsequent adaptive adjustment is paused, maintaining the current training parameters until the data returns to stability.

[0043] In this embodiment, step S200, based on training feature data, identifies the patient's current training adaptation state through a training state evaluation model and generates a corresponding training regulation strategy, including the following step B1: B1: Training adaptation status includes at least three categories: insufficient training intensity, appropriate training intensity, and excessive training load; The training status assessment model is used to identify the patient's current training adaptation status category based on the preset interval combination of training feature data, and to generate training regulation strategies according to the preset strategy rules corresponding to the training adaptation status category.

[0044] Specifically, the training state assessment model outputs three categories of training adaptation states: insufficient training intensity, appropriate training intensity, and excessive training load.

[0045] The training status assessment model is constructed as follows: the exercise completion characteristics, fatigue characteristics, participation characteristics, load characteristics, and the coefficient of variation representing the stability of the data in the training feature data are used as input variables. Each input variable is divided into multiple preset value intervals, and each value interval corresponds to the typical value range of the input variable in a specific training adaptation state. Then, based on the combination of the value intervals of each input variable, the patient's current training adaptation state category is identified through the pre-constructed correspondence.

[0046] Specifically, when the exercise completion characteristics and participation characteristics are both at a moderate or high level, the fatigue characteristics are at a moderate level, the load characteristics are at a moderate level, and the coefficient of variation is within a stable range, the patient is determined to be in a training adaptation state with appropriate training intensity. When the exercise completion characteristic is at a high level but the participation characteristic is at a low level, or the fatigue characteristic is at a low level, the patient is determined to be in a training adaptation state with insufficient training intensity. When fatigue characteristics are at a low level and load characteristics are at a high level, or when participation characteristics show a significant decrease, the patient is determined to be in a training adaptation state with excessive training load. When the coefficient of variation exceeds the preset stable range, it is determined that the current training feature data is insufficient to support reliable state recognition, and the training state evaluation model maintains the training adaptation state obtained from the previous recognition.

[0047] Specifically, the training state assessment model, while obtaining the patient's current training adaptation state category, also performs a confidence assessment on the identification result. The assessment process includes calculating the degree of belonging to a preset value range for each input variable, taking the minimum degree of belonging as the activation intensity of the corresponding identification criterion, and then aggregating the activation intensities of different identification criteria. The maximum degree of belonging value in the aggregated result is taken as the confidence score of this training adaptation state identification, which can be expressed by the following formula: in, This represents the degree of belonging to the output result after aggregation, and y represents the continuous output value obtained after weighting.

[0048] If the confidence level is lower than the preset threshold, the current recognition result is determined to be in an uncertain state. The training state evaluation model continues to use the training control strategy generated last time, while improving the collection accuracy of training feature data to obtain more stable data to support the next recognition. If the confidence level is not lower than the preset threshold, the training state evaluation model generates a training control strategy based on the identified training adaptation state category and the preset strategy rule corresponding to that category. The training control strategy contains control direction information for determining the adjustment direction of virtual scene task parameters and mechanical execution parameters, which is used by the spatial mapping model in subsequent steps.

[0049] In this embodiment, step S300 converts kinematic data into avatar pose data in a virtual scene based on a spatial mapping model, and determines virtual scene task parameters by combining training and control strategies. The virtual scene generation unit then updates the scene content accordingly and presents it to the patient, including the following steps C1-C3: C1: The spatial mapping model obtains the initial body posture data by transforming the kinematic data into coordinates; The system adapts the initial body posture data proportionally based on the patient's body size information, and predicts and compensates the adapted body posture data based on the changing trend of kinematic data, thus obtaining body posture data that is synchronized with the patient's lower limb movements in real time.

[0050] Specifically, the spatial mapping model first converts kinematic data into avatar pose data in the virtual scene. This conversion process includes three sub-processes: coordinate transformation, human-scale adaptive scaling, and low-latency pose prediction compensation. In the coordinate transformation sub-process, three coordinate systems are defined: the sensor coordinate system, the human anatomical coordinate system, and the virtual scene coordinate system. The sensor coordinate system has its origin at the base of the mechanical actuator, and the raw angle values ​​output by each joint angle sensor are based on this coordinate system. For implementations with active joint angle measurement capabilities, the joint angle sequence is converted into the spatial pose of the foot in the sensor coordinate system through forward kinematics. Specifically, the hip joint coordinate is set as follows: Thigh length is The length of the lower leg is Knee joint center With the center of the ankle joint The positions are calculated using the recursive formula for the knee joint position in forward kinematics: in , These represent the hip and knee flexion / extension angles, respectively. The recursive formula for the ankle joint position in forward kinematics is as follows: Furthermore, the foot pose in the sensor coordinate system is transformed to the avatar foot pose in the virtual scene coordinate system using a rigid body transformation matrix. Specifically, this is calculated using the coordinate system transformation formula: in, For rotation matrix, It is a translation vector, determined by the relative relationship between the sensor coordinate system and the virtual scene coordinate system.

[0051] For mechanical actuators where joint angles cannot be directly measured, a reverse kinematics method is adopted to infer joint angles from the foot position and link spatial position. Specifically, the knee joint angle is calculated using the reverse kinematics formula. Formula for solving the hip joint angle using inverse kinematics: Where (x, y, z) are the position coordinates of the foot tip in the human anatomical coordinate system.

[0052] In the human-scale adaptive scaling process, the bone length and joint positions of the avatar in the virtual scene are scaled according to the patient's individual parameters such as height and weight. The bone length is calculated according to the bone length scaling formula: in, The actual limb length of the patient. Preset the height for the avatar. The patient's actual height is used as the reference, and the lengths of the remaining bones are calculated according to the same proportional relationship.

[0053] Simultaneously, during the initialization phase, the pose offset of each joint is collected when the patient is standing still. This offset is then added during each coordinate transformation to perform joint position compensation, specifically calculated using the joint position compensation formula: in, This represents the initial offset of the corresponding joint in the human anatomical coordinate system.

[0054] In the low-latency attitude prediction compensator process, the end-to-end delay from kinematic data acquisition to the driving avatar completing attitude update is first measured during the initialization phase. Specifically, it is calculated using the end-to-end delay measurement formula: in, The time when the synchronization signal is emitted. To determine the moment when the avatar completes the posture update, the delay is retested at preset intervals during training.

[0055] Furthermore, a Kalman filter-based attitude prediction algorithm is used to predict joint angles. A state vector containing joint angles, angular velocities, and angular accelerations is defined, and its state transition equation and observation equation are expressed by the Kalman filter state transition equation: Where A is the state transition matrix and Q is the process noise covariance. Kalman filter observation equation: Where C is the observation matrix and R is the measurement noise covariance, the predicted value of the joint angle is output through this filtering process to offset the delay introduced by the data transmission and processing process.

[0056] When the end-to-end delay exceeds one sampling period, a second-order extrapolation compensation is performed based on the Kalman filter prediction. Specifically, it is calculated using the second-order extrapolation compensation formula: in, , These are the angular velocity and angular acceleration at the current moment, respectively.

[0057] To prevent high-frequency jitter introduced by prediction compensation, a low-pass filter is applied to smooth the output prediction value. The specific calculation is performed using the low-pass filter smoothing formula: in, As a smoothing coefficient, the value of this coefficient is adaptively adjusted according to the patient's current movement speed to ensure smooth and continuous changes in the avatar's posture in the virtual scene.

[0058] After the above three sub-processes of coordinate transformation, scale scaling, and prediction compensation, the final avatar posture data that is synchronized with the patient's lower limb movements in real time is obtained.

[0059] C2: The spatial mapping model obtains the adjustment direction and adjustment range of the execution parameters of the mechanical execution unit based on the scene difficulty level corresponding to the virtual scene task parameters; The mechanical actuator adjusts the assistance or resistance to the patient's lower limbs in stages based on the difference between the current execution parameters and the previous execution parameters; When a preset abnormal change is detected in the patient's physiological signal data, a control command is generated to interrupt the current training.

[0060] Specifically, refer to Figure 4 The spatial mapping model obtains avatar posture data and, in conjunction with training and control strategies, determines virtual scene task parameters. It then converts these virtual scene task parameters into mechanical execution parameters, thus achieving a correspondence between virtual space difficulty settings and physical space execution intensity.

[0061] Specifically, the spatial mapping model determines the adjustment direction and magnitude of the mechanical execution parameters relative to their current values ​​based on the scene difficulty level corresponding to the virtual scene task parameters. The adjustment direction includes increasing or decreasing the execution intensity, and the adjustment magnitude is related to the degree of change in the scene difficulty level.

[0062] After obtaining the adjusted mechanical execution parameters, the parameters are compared with the previously effective mechanical execution parameters, and the difference between the two is calculated. If the difference exceeds the preset safety threshold, the target mechanical execution parameters are not output all at once. Instead, the target value is divided into multiple stages and output gradually at preset time intervals to avoid sudden changes in execution parameters causing discomfort to the patient's lower limbs.

[0063] During the above parameter adjustment and phased output process, the patient's physiological signal data is monitored for comfort and safety. If the physiological signal data shows a preset abnormal change in a short period of time, it is determined that the patient has an adverse reaction. At this time, a control command to interrupt the current training is immediately generated. The execution priority of this control command is higher than the conventional mechanical execution parameter adjustment process, thereby ensuring the gradual adaptation of training intensity while protecting the patient's training safety.

[0064] C3: The virtual scene content includes scene difficulty elements and interactive feedback elements corresponding to the training task parameters. The virtual scene generation unit presents the updated virtual scene content to the patient in an immersive visual form through a display device.

[0065] Specifically, the virtual scene content includes scene difficulty elements corresponding to the virtual scene task parameters and interactive feedback elements used to present training feedback. The virtual scene content is updated based on avatar posture data and virtual scene task parameters, and the updated virtual scene content is output to the patient in an immersive visual form.

[0066] To ensure the consistency between the updated virtual scene content and the patient's actual lower limb movements in time, multiple layers of low-latency technologies are employed during the virtual scene content update process: Firstly, before the virtual scene rendering begins, the currently collected joint angle data is not used directly. Instead, the avatar posture data obtained through prediction compensation in C1 is used to drive the update of the virtual scene content, so that when the virtual scene is displayed, its content is as consistent as possible with the actual position of the patient's lower limbs. Secondly, the process of generating virtual scene content and the process of outputting the screen are designed as two independent and asynchronous processing steps. After the previous processing step completes the generation of the current frame content, when the latest kinematic data arrives, the generated screen content is corrected for posture before being output, so as to compensate for posture deviation at the last moment before the screen is output. Third, the area of ​​the patient's gaze in the virtual scene is processed with a higher resolution, while the area outside the field of vision is processed with a lower resolution. This reduces the amount of computation required for processing a single frame without affecting the patient's perception of the scene content, thereby shortening the time required to update the virtual scene content. Fourth, the posture-driven data during the virtual scene content update process is set to a high real-time processing priority, taking precedence over the update processing of non-real-time elements in the scene.

[0067] After updating the virtual scene content, the feedback effect of the scene update on the patient's training performance was evaluated by comparing the patient's motor completion characteristics before and after the update. Specifically, the effect was calculated using the real-time feedback gain evaluation formula. This is used to determine whether the current virtual scene content update has a positive impact on the patient's training performance. If the calculation result exceeds the preset positive range, the corresponding training control strategy will be recorded as a valid case.

[0068] In this embodiment, step S400 converts the virtual scene task parameters into execution parameters of the mechanical execution unit based on the spatial mapping model, and then uses the mechanical execution unit to provide assistance or resistance adjustment to the patient's lower limbs, including the following step D1: D1: Based on the spatial mapping model, the virtual scene task parameters are converted into the execution parameters of the mechanical execution unit, which then provides assistance or resistance adjustment to the patient's lower limbs.

[0069] Specifically, the mechanical execution parameters obtained in C2 are output to the mechanical execution unit. The mechanical execution unit applies assistance or resistance to the patient's lower limbs according to the received mechanical execution parameters, which corresponds to the current training adaptation state, so as to realize the physical representation of the training difficulty set in the virtual scene in the actual movement of the patient's lower limbs.

[0070] After the mechanical execution unit completes its response, the patient's lower limb movement state changes accordingly, triggering the acquisition of kinematic and physiological signal data again. The newly acquired data undergoes further processing through steps such as training feature data extraction, training adaptation state identification, spatial mapping transformation, and mechanical execution parameter output. This forms a closed-loop training process that sequentially connects and iterates through data acquisition, feature fusion, state decision-making, parameter execution, and effect feedback. The duration of each loop is controlled within a preset time range, ensuring that changes in the patient's lower limb movement state can be re-acquired and incorporated into the next round of training control within a short period of time, so that the training control strategy always matches the patient's current real-time state.

[0071] Example 3 Reference Figure 2 This is the third embodiment of the present invention.

[0072] The above is a schematic scheme of a lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping. It should be noted that the technical solution of this lower limb rehabilitation training system based on kinematic-virtual scene bidirectional mapping belongs to the same concept as the technical solution of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described above. Details not described in detail in the technical solution of the lower limb rehabilitation training system based on kinematic-virtual scene bidirectional mapping in this embodiment can be found in the description of the technical solution of the lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping described above.

[0073] This embodiment also provides a lower limb rehabilitation training system based on kinematic-virtual scene bidirectional mapping, including: a motion sensing unit for collecting kinematic data of the patient's lower limbs; Physiological signal acquisition unit, used to acquire the patient's physiological signal data; The main control unit is used to perform time-series synchronous processing of kinematic data and physiological signal data to obtain training feature data, identify the patient's current training adaptation state through the training state assessment model and generate training regulation strategies, and also to obtain avatar posture data in the virtual scene based on kinematic data through the spatial mapping model, determine virtual scene task parameters based on the training regulation strategies, and obtain the execution parameters of the mechanical execution unit based on the virtual scene task parameters. The virtual scene generation unit is used to update the virtual scene content based on the avatar posture data and virtual scene task parameters and present it to the patient. The mechanical actuator is used to provide assistance or resistance adjustment to the patient's lower limbs based on the execution parameters.

[0074] This embodiment also provides an electronic device suitable for lower limb rehabilitation training with kinematic-virtual scene bidirectional mapping, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the lower limb rehabilitation training method with kinematic-virtual scene bidirectional mapping as proposed in the above embodiment.

[0075] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the lower limb rehabilitation training method for achieving bidirectional mapping of kinematics and virtual scenes as proposed in the above embodiments.

[0076] The storage medium proposed in this embodiment belongs to the same inventive concept as the lower limb rehabilitation training method for realizing kinematic-virtual scene bidirectional mapping proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0077] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0078] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping, characterized by: This includes collecting the patient's lower limb kinematic data through a motion sensing unit, collecting the patient's physiological signal data through a physiological signal acquisition unit, and synchronizing the two in time to obtain training feature data; Based on the training feature data, the patient's current training adaptation status is identified through the training status assessment model, and corresponding training regulation strategies are generated. Based on the spatial mapping model, the kinematic data is converted into avatar posture data in the virtual scene, and the virtual scene task parameters are determined in combination with the training and control strategy. The virtual scene generation unit updates the scene content accordingly and presents it to the patient. Based on the spatial mapping model, the virtual scene task parameters are converted into execution parameters of the mechanical execution unit, which then provides assistance or resistance adjustment to the patient's lower limbs accordingly.

2. The lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping as described in claim 1, characterized in that: The training adaptation states include three categories: insufficient training intensity, appropriate training intensity, and excessive training load. The training status assessment model identifies the patient's current training adaptation status category based on the preset interval combination of the training feature data, and evaluates the confidence level of the identification results. If the confidence level is not lower than the preset threshold, the training regulation strategy is generated according to the preset strategy rule corresponding to the training adaptation state category; If the confidence level is lower than the preset threshold, the training control strategy generated in the previous iteration will be used.

3. The lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping as described in claim 2, characterized in that: The spatial mapping model transforms the kinematic data into initial body posture data through coordinate transformation; and performs proportional adaptation on the initial body posture data according to the patient's body size information. Based on the changing trend of the kinematic data, it is determined whether there is a posture deviation. If there is a deviation, the adapted avatar posture data is predicted and compensated to obtain the avatar posture data that is synchronized with the patient's lower limb movement in real time. If there is no deviation, the adapted avatar posture data will be used directly as the avatar posture data.

4. The lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping as described in claim 3, characterized in that: The spatial mapping model obtains the adjustment direction and adjustment range of the execution parameters of the mechanical execution unit based on the scene difficulty level corresponding to the virtual scene task parameters; The adjusted execution parameters are compared with the previous execution parameters. If the difference exceeds the preset safety threshold, the execution parameters are output in stages. If the difference does not exceed the preset safety threshold, the execution parameters are output directly. When a preset abnormal change is detected in the patient's physiological signal data, a control instruction is generated to interrupt the current training, and the execution priority of the control instruction is higher than the adjustment of the execution parameters.

5. The lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping as described in claim 4, characterized in that: The kinematic data includes at least one of the patient's lower limb joint angle data and spatial pose data, and the physiological signal data includes at least one of the patient's lower limb muscle electromyography signal, heart rate-related signal, and skin conductance response signal. If the sampled values ​​in the kinematic data or physiological signal data exceed a preset reasonable range, the transmission of the corresponding data to subsequent steps will be paused, and data compensation or a prompt to re-collect data will be performed. If the data does not exceed a preset reasonable range, the kinematic data and the physiological signal data will be used for time-series synchronization processing.

6. The lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping as described in claim 5, characterized in that: The training feature data includes exercise completion features that characterize the patient's level of exercise completion, fatigue features that characterize the patient's level of physical fatigue, and participation features that characterize the patient's level of training engagement. If the coefficient of variation of the training feature data is within a preset stable range, then the training feature data is used to identify the training adaptation state. If the coefficient of variation of the training feature data continues to exceed the preset stable range, the generation of the training control strategy will be paused, and the current training control strategy will be maintained.

7. The lower limb rehabilitation training method based on kinematic-virtual scene bidirectional mapping as described in claim 6, characterized in that: The virtual scene content includes scene difficulty elements and interactive feedback elements corresponding to the training task parameters, and the updated virtual scene content is output to the patient in an immersive visual form. Based on the difference in the patient's motor completion characteristics before and after the update, it is determined whether the update of the virtual scene content has a positive impact on the patient's training performance; If a positive impact is generated, the corresponding training control strategy will be recorded as a valid case. If no positive impact is generated, the training control strategy will not be recorded.

8. A lower limb rehabilitation training system based on kinematic-virtual scene bidirectional mapping, employing the lower limb rehabilitation training method of kinematic-virtual scene bidirectional mapping as described in any one of claims 1-7, characterized in that: include, Motion sensing unit, used to collect kinematic data of the patient's lower limbs; Physiological signal acquisition unit, used to acquire the patient's physiological signal data; The main control unit is used to perform time-series synchronization processing on the kinematic data and the physiological signal data to obtain training feature data, identify the patient's current training adaptation state and generate training regulation strategies through a training state evaluation model, and is also used to obtain avatar posture data in the virtual scene based on the kinematic data through a spatial mapping model, determine virtual scene task parameters based on the training regulation strategies, and obtain execution parameters of the mechanical execution unit based on the virtual scene task parameters. A virtual scene generation unit is used to update the virtual scene content based on the avatar posture data and the virtual scene task parameters and present it to the patient. A mechanical actuator is used to provide assistance or resistance adjustment to the patient's lower limbs according to the execution parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the lower limb rehabilitation training method of kinematic-virtual scene bidirectional mapping as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the lower limb rehabilitation training method of kinematic-virtual scene bidirectional mapping as described in any one of claims 1-7.

Images