Hand function rehabilitation evaluation and training method based on virtual reality technology

Abstract

The invention relates to the technical field of hand function rehabilitation, in particular to a hand function rehabilitation evaluation and training method based on a virtual reality technology, which comprises the following steps: constructing a reference attitude sequence according to an evaluation sample, comparing real-time attitudes frame by frame to extract a difference trend, recording a turn-back path and a turn-back frequency, and calculating a redundant path and a stability index. According to the method, by accurately constructing a personalized training target and contrasting real-time attitude information frame by frame, the rollback state in training can be recognized in time, redundant paths and stability characteristics can be analyzed, instability in the training process can be quantified, and the training efficiency can be improved. According to the innovative mode, the accuracy of training feedback is improved, a personalized and intelligent rehabilitation adjustment scheme is provided for a patient, rhythm control and tolerance band setting are combined, the effectiveness of a training task closed loop is ensured, blind adjustment of training loads is avoided, and the rehabilitation effect and the pertinence and effectiveness of progress are improved.

Description

Technical Field

[0001] This invention relates to the field of hand function rehabilitation technology, and in particular to a hand function rehabilitation assessment and training method based on virtual reality technology. Background Technology

[0002] The field of hand function rehabilitation technology refers to a set of techniques for assessing and implementing rehabilitation training for patients with hand dysfunction, focusing on motor and sensory abilities. Its core components include hand joint range of motion measurement, grip and pinch strength testing, fine motor ability assessment, sensory threshold and tactile recognition testing, functional motor task training, and data recording and follow-up analysis throughout the rehabilitation process. Initial and phased assessments are completed using standardized scales combined with clinical observation. Rehabilitation plans are developed and continuously intervened through passive stretching training, active joint mobility training, strength training, coordination and dexterity training, and daily task simulation training. Traditional hand function rehabilitation assessment and training methods refer to the assessment and training conducted by physicians based on clinical findings during the hand function rehabilitation process. This technique, which uses observation and scale scoring to assess motor and sensory functions and arrange repetitive hand movement training for patients, addresses the issues of insufficient quantitative assessment and weak interactive training in hand function rehabilitation. Traditional methods use articular goniometers to measure finger and wrist joint range of motion, grip dynamometers or pinch dynamometers to measure hand strength, and score assessments based on the modified Barthel Index Upper Limb Function Scale or Hand Function Grading Scale. Training is arranged through exercises such as finger flexion and extension, wrist flexion, extension and rotation, clay grasping and kneading, bean stringing, towel wringing, and daily living activities. During training, manual recording of training repetitions and completion status is mainly used for periodic review and adjustment.

[0003] Current technologies for hand function rehabilitation rely on manual measurement and simple scale assessments, often failing to accurately reflect the dynamic changes in patients during rehabilitation in real time. Traditional methods primarily collect physical data using tools such as articular goniometers and hand dynamometers, and assessments mainly depend on periodic manual recording and periodic review. These traditional assessment and training methods lack real-time feedback, often failing to promptly identify instability or fatigue issues during patient training, resulting in insufficient targeting and personalization of training. For example, traditional methods, through fixed training tasks and hand movement simulations, easily overlook subtle postural differences that occur during specific training, leading to the failure to optimize or adjust individual training items in a timely manner. Furthermore, current technologies rely heavily on manual data recording during training, which not only increases the operational burden but also makes it difficult to ensure the accuracy and completeness of the data, thus affecting the sustainability and precision of rehabilitation effects. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a hand function rehabilitation assessment and training method based on virtual reality technology, including the following steps: S1: Based on the target task samples extracted during the evaluation phase, the training workstation constructs a reference posture sequence using wrist joint direction and finger joint angle data, combines the sampling time to form a target frame structure, and generates a posture comparison benchmark group. S2: Based on the posture comparison benchmark group, the real-time posture during the training process is compared frame by frame, the trend of posture difference change is extracted, the backtracking state is identified, the corresponding path length and backtracking number are recorded, and a posture backtracking record set is generated. S3: Using the path and event data in the attitude reversal record set, the training workstation calculates the redundant path ratio and extracts instability features. At the same time, it identifies the continuous stable angular velocity segment, calculates its duration ratio, and outputs the training stability analysis results. S4: Call the instability index and the proportion of stable segments in the training stability analysis results, combine them with the preset judgment threshold, determine whether to update the parameters, and generate rhythm level control parameters; S5: Construct a control instruction data packet based on the rhythm level control parameters, including rhythm frequency and tolerance band parameters, and send it from the training workstation to the VR host and motion capture component. After the device performs the adjustment task, it returns a feedback segment and generates a training task closed-loop configuration group.

[0005] As a further aspect of the present invention, the posture reference set includes wrist joint direction information, knuckle angle data, and target frame structure; the posture return record set includes posture difference change trend, return status, path length, and return count; the training stability analysis results include redundant path ratio index, instability characteristics, continuous stable angular velocity segments, and duration proportion; the rhythm level control parameters include instability index, stable segment proportion, and judgment threshold; the training task closed-loop configuration set includes control instruction data package, rhythm frequency parameters, tolerance band range, and feedback segments.

[0006] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Based on the target task samples extracted during the evaluation phase, call the wrist joint direction information and finger joint angle data, and synchronize them according to the sampling time point to construct the wrist joint and finger joint posture matching matrix at the time point and generate a synchronized posture matrix group. S102: Based on the synchronized attitude matrix group, extract the attitude change trajectory at each time point, compare the attitude change amplitude with the overall change level, filter the frame data with prominent changes, and combine the corresponding time point and attitude matrix to generate a key frame attitude sequence group. S103: For the key frame pose sequence group, extract the angle parameters of the wrist joint and finger joints, set the interval boundary values ​​according to the distribution interval, and construct the pose reference frame structure in combination with the key frame time points to generate a pose comparison benchmark group.

[0007] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the posture reference group, collect real-time posture information during the training process, and compare the real-time posture frames with the corresponding reference frames in the posture reference group frame by frame according to the sampling time point order. Perform difference calculation on the wrist joint direction parameters and finger joint angle parameters, summarize the frame difference amount and arrange them in time order to generate a posture difference sequence. S202: Based on the attitude difference sequence, calculate the magnitude of the difference between adjacent frames, and judge the direction of change according to the preset attitude change judgment threshold. Mark the frame segments where the continuous change direction reverses as back segments, and record the start and end time points of the corresponding frame segments to obtain the back segment identifier set. S203: For the set of backtracking segment identifiers, call the attitude path data of consecutive frames within the segment, calculate the path length value of the backtracking segment, and count the number of times the backtracking segment appears in the same training process. Summarize and record the path length value and the number of backtracking according to the segment number to establish an attitude return record set.

[0008] As a further aspect of the present invention, the setting of the posture change determination threshold specifically involves limiting the change amplitude of the wrist joint direction parameter and the change amplitude of the finger joint angle parameter to not less than a preset first amplitude threshold and a second amplitude threshold, respectively. The first amplitude threshold and the second amplitude threshold are set in a graded manner according to the movement difficulty level of different rehabilitation training stages. When generating the regression segment identifier set, it is determined that there is a frame segment where the continuous change direction reverses only when the change amplitude of the difference between adjacent frames simultaneously meets the first amplitude threshold and the second amplitude threshold. The start and end times of each frame segment in the rollback segment identifier set are determined by identifying the time point when the direction of the difference change first reverses and the time point when it first recovers to the same direction as before the reversal.

[0009] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the path length data and training process timeline information in the attitude return record set, extract the total path value of each backtracking path, and perform difference calculation with the standard path value of the corresponding training stage, and then normalize it. Summarize the normalization results according to the training process stage to generate a redundancy path ratio index. S302: Based on the redundant path ratio index, filter the ratio intervals that are greater than the instability judgment threshold, extract the time period sequence within the corresponding interval, and combine the number of backtracking times and the path fluctuation frequency within the sequence to construct a feature vector set of unstable state and obtain the instability feature coefficient. S303: Combining the aforementioned instability characteristic coefficients, call the angular velocity sequence data during the training process, filter out segments with continuous changes less than the stability threshold based on the angular velocity change values ​​between adjacent frames, count the number of continuous frames in the stable segments, and calculate the ratio of the number to the total number of frames in the entire training process to establish the training stability analysis results.

[0010] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the instability feature coefficient and stable segment ratio value in the training stability analysis results, and perform interval assignment judgment on the two indicators according to the preset instability ratio threshold and ratio judgment threshold in the training workstation, generate a binary judgment label sequence, and obtain stability judgment label; S402: Based on the stability judgment marker, filter the sequence segments whose marker values ​​do not meet the conditions, extract the parameter fluctuation amplitude and feedback frequency during the training process in the corresponding time period, perform amplitude level mapping operation on the two, establish parameter adjustment interval feature values, and generate parameter update trigger signals. S403: Call the parameters to update the trigger signal, determine whether the trigger points meet the level allocation standard according to the training period sequence, set the corresponding level code for the time points that meet the conditions, and construct a rhythm mapping table in combination with the basic rhythm parameters corresponding to the training stage to establish rhythm level control parameters.

[0011] As a further aspect of the present invention, the instability ratio threshold is specifically set by dividing it into three numerical ranges according to the precision of the training task. Each level of the instability ratio threshold corresponds to a tolerance standard for different training precision requirements. When generating the stability judgment label, the corresponding position in the binary judgment label sequence is marked as not meeting the condition only when the instability feature coefficient exceeds the instability ratio threshold of the corresponding training task level and the stable segment proportion is lower than the proportion judgment threshold. The mapping operation between the parameter fluctuation amplitude and the feedback frequency amplitude level is as follows: according to the five-level amplitude mapping standard set in the training workstation, the parameter fluctuation amplitude and the feedback frequency are respectively classified into the corresponding amplitude level intervals, and the characteristic value of the parameter adjustment interval is determined by the product of the amplitude level corresponding to the parameter fluctuation amplitude and the feedback frequency.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the level code and rhythm mapping table in the rhythm level control parameters, extract the corresponding rhythm frequency value and tolerance band boundary value, bind the two parameters into a continuous parameter group structure according to the time series, and encapsulate them into a unified format data field in combination with the data transmission protocol format requirements to generate a control instruction data packet; S502: According to the control instruction data packet, the data fields are sent to the VR host and the motion capture component respectively, and the instruction execution feedback information responded by the device in the task phase is detected. The key frame number and rhythm response interval in the feedback content are filtered, and the feedback information in the current execution phase is summarized to obtain a set of feedback segments. S503: Combine the feedback fragment set, extract the training path data corresponding to the frame number, and perform frame position matching processing with the control parameters bound before sending. Based on the feedback matching rate and execution duration, construct the closed-loop execution segment sequence of the training process and establish a training task closed-loop configuration group.

[0013] As a further aspect of the present invention, the rhythm frequency values ​​included in the rhythm mapping table are set within a range of 0.2Hz to 2.0Hz, and a correspondence between the rhythm level code and the rhythm frequency value is constructed with a step size of 0.1Hz. The tolerance band boundary value is set according to the error tolerance standard corresponding to different rhythm levels, and the upper and lower boundary values ​​are determined by extending 0.05Hz upwards and downwards according to the frequency level. Each group in the continuous parameter group structure includes a rhythm level code, a rhythm frequency value, and a pair of tolerance band boundary values. When the rhythm frequency value and the tolerance band boundary value are encapsulated into a unified format data field, the fields are arranged in order of high byte first and low byte last according to the data transmission protocol format requirements, and a single byte field identifier is added to each group of data to form a data field structure with a total length of fixed bytes.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by accurately constructing personalized training goals and comparing them frame by frame with real-time posture information, it is possible to promptly identify regression states during training, analyze redundant paths and stability characteristics, and quantify instability during training. This innovative approach improves the accuracy of training feedback, provides patients with personalized and intelligent rehabilitation adjustment plans, and, combined with rhythm control and tolerance zone settings, ensures the effectiveness of the training task closed loop, avoids blindly adjusting the training load, and improves the targetedness and effectiveness of rehabilitation effects and progress. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in 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.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1This invention provides a method for hand function rehabilitation assessment and training based on virtual reality technology, including the following steps: S1: Based on the target task samples extracted during the evaluation phase, the training workstation uses wrist joint direction information and finger joint angle data as training standards to construct a reference posture sequence, and combines the sampling time points to form a target frame structure, generating a posture comparison benchmark group. S2: Based on the attitude comparison benchmark group, by comparing the real-time attitude information during the training process frame by frame, the trend of attitude difference changes is extracted, the backtracking state is identified, and the path length and backtracking number of related segments are recorded to generate an attitude backtracking record set. S3: Based on the path and event data in the attitude return record set, the training workstation calculates the redundant path ratio index and extracts instability features. At the same time, it identifies the continuous stable angular velocity segment during the training process, calculates the proportion of the duration in the training time, and generates training stability analysis results. S4: Call the instability index and stable segment ratio in the training stability analysis results, and determine whether to perform parameter updates based on the preset judgment threshold in the training workstation, and generate rhythm level control parameters. S5: Call the rhythm level control parameters to construct a control instruction data package, including rhythm frequency parameters and tolerance band range. The instructions are sent from the training workstation to the VR host and motion capture component. After the associated device performs the task adjustment, it returns a feedback segment as training input to generate a training task closed-loop configuration group.

[0023] The posture reference set includes wrist joint direction information, knuckle angle data, and target frame structure; the posture return record set includes posture difference change trend, return status, path length, and number of return times; the training stability analysis results include redundant path ratio index, instability characteristics, continuous stable angular velocity segments, and duration percentage; the rhythm level control parameters include instability index, stable segment percentage, and judgment threshold; the training task closed-loop configuration set includes control instruction data package, rhythm frequency parameters, tolerance band range, and feedback segments.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Based on the target task samples extracted during the evaluation phase, call the wrist joint direction information and finger joint angle data, and synchronize them according to the sampling time point to construct the wrist joint and finger joint posture matching matrix at the time point and generate a synchronized posture matrix group. Consistency verification was performed on the wrist joint orientation information and finger joint angle data recorded in the target task samples. Verification included the data acquisition object identifier, task number, and acquisition time range, ensuring that both types of data corresponded to the same training object and were generated within the same training task cycle. Wrist joint orientation information described the wrist's posture changes in space as a continuous angle sequence, while finger joint angle data recorded hand movement states as the flexion and extension angles of each finger's proximal joint. Both types of data were sorted according to their sampling time points, and data records with missing time markers, duplicate time points, or time jumps were removed. For cases where data was missing on one side but the time points were still continuous, the angle values ​​corresponding to the previous valid time point were used for completion processing, and a completion marker field was added to the data record. After completing the timeline organization, the wrist joint orientation information and finger joint angle data were aligned point-by-point using the sampling time point as the synchronization basis. The wrist joint heading, pitch, and roll parameters at the same time point were merged with the finger joint angle parameters into a unified attitude vector. The attitude vectors were arranged in time point order, and an attitude matching matrix was constructed with time points as row indices and attitude parameters as column indices. For multiple attitude matching matrices formed from samples of the same target task, they are merged and organized according to sample number and acquisition order to form a synchronous attitude matrix group, providing a continuous and unified data input basis for subsequent attitude change analysis.

[0025] S102: Based on the synchronized attitude matrix group, extract the attitude change trajectory at each time point, compare the attitude change amplitude with the overall change level, filter the frame data with prominent changes, and combine the corresponding time point and attitude matrix to generate a key frame attitude sequence group. After the synchronous attitude matrix is ​​formed, attitude change features during training are extracted. Attitude vectors in the attitude matching matrix are read sequentially over time. Difference processing is performed on the attitude vectors at adjacent time points along their corresponding parameter dimensions to obtain a sequence of changes describing the magnitude of attitude changes. This sequence of changes, indexed by time points, reflects the attitude changes between adjacent sampling points during training. Within the same target task sample range, the sequence of changes is statistically analyzed, and the concentrated distribution interval of changes is extracted as a reference for the overall change level. The changes at each time point are compared point-by-point with the overall change level. When the change at a certain time point consistently falls within the high range of the overall change level, that time point is marked as a candidate frame with prominent changes. For candidate frames continuously distributed along the time axis, a merging process is performed based on the magnitude of the changes, retaining only the time point with the largest change as the representative frame of that continuous segment. The representative frame is considered the key time point where attitude changes are most concentrated during training. The time stamp corresponding to the representative frame and its complete attitude data in the attitude matching matrix are written into the keyframe attitude sequence group, ensuring that the keyframe attitude sequence fully reflects the time nodes with significant attitude change features during training.

[0026] S103: For the keyframe pose sequence group, extract the angle parameters of the wrist joint and finger joint, set the interval boundary value according to the distribution interval, and construct the pose reference frame structure in combination with the keyframe time points to generate the pose comparison benchmark group. The pose data in the keyframe pose sequence is read one by one, and wrist joint direction parameters and knuckle angle parameters are extracted and classified. The parameter values ​​corresponding to all keyframes are then aggregated within the same training phase. Based on the aggregated results, the distribution of each angle parameter in the keyframe set is statistically analyzed, and the effective interval boundaries of each parameter are determined by analyzing the concentrated distribution intervals. The interval boundaries are determined based on the value range of most keyframes, and extreme values ​​that significantly deviate from the concentrated intervals are removed to ensure that the interval boundaries reflect the main pose features within the training phase. After the interval boundaries are determined, the time point information, wrist joint direction parameters, knuckle angle parameters, and corresponding interval boundaries of each keyframe are associated to construct a pose reference frame structure containing time markers and parameter interval information. The pose reference frame structure is arranged chronologically and grouped according to the training phase to form a pose comparison benchmark group. The pose comparison benchmark group is used for subsequent real-time pose comparison operations, providing a stable and consistent reference for pose deviation analysis during training.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the attitude reference group, collect real-time attitude information during the training process, and compare the real-time attitude frames with the corresponding reference frames in the attitude reference group frame by frame according to the sampling time point order. Perform difference calculation on wrist joint direction parameters and finger joint angle parameters, summarize the frame difference amount and arrange them in time order to generate attitude difference sequence. During training, real-time posture information is continuously collected, including wrist joint orientation parameters and finger joint angle parameters. The real-time posture information is used to generate a posture frame sequence at fixed sampling time points. Before entering the comparison operation, the posture frames undergo timeline processing to ensure continuous and valid time markers. Based on the time markers of each posture reference frame in the posture reference group, time alignment is performed between the real-time posture frames and their corresponding reference frames. After alignment, item-by-item interpolation is performed on the real-time posture frames and reference frames in the dimensions of wrist joint orientation parameters and finger joint angle parameters to obtain the deviation of each parameter. The deviation reflects the degree of change of the real-time posture relative to the reference posture. The deviations of each parameter at the same time point are summarized to form a single-frame posture difference. All single-frame posture difference quantities are arranged in chronological order to form a posture difference sequence. The posture difference sequence is used to describe the continuous deviation of the real-time posture relative to the posture reference throughout the entire training process.

[0028] S202: Based on the attitude difference sequence, calculate the magnitude of the difference between adjacent frames, and judge the direction of change according to the preset attitude change judgment threshold. Mark the frame segments where the continuous change direction reverses as backtracking segments, and record the start and end time points of the corresponding frame segments to obtain the backtracking segment identifier set. After the attitude difference sequence is formed, the attitude change trend and regression characteristics are determined. The magnitude of the difference between adjacent time points in the attitude difference sequence is calculated, and the positive and negative directions of the magnitude are recorded. The magnitude of the change is compared point by point with a preset attitude change judgment threshold. When the magnitude of the change exceeds the threshold, the time point is marked as a significant change state. The change direction information is continuously scanned on the time axis. When the change direction changes from positive to negative or vice versa in a significant change state, the time range is determined as a regression segment. Each regression segment corresponds to its start and end time points, and a segment number is generated. The segment numbers, start and end time points, and their position indices in the attitude difference sequence of all regression segments are compiled and written into a regression segment identifier set. The regression segment identifier set is used to centrally describe the time distribution of attitude adjustment regression corrections during training.

[0029] S203: For the set of backtracking segment identifiers, call the attitude path data of consecutive frames within the segment, calculate the path length value of the backtracking segment, and count the number of times the backtracking segment appears in the same training process. Summarize and record the path length value and the number of backtracking according to the segment number to establish an attitude return record set. Based on the start and end times recorded in the rollback segment identifier set, continuous attitude data within the corresponding time range is retrieved. Adjacent attitude data within a segment undergo frame-by-frame interpolation across each attitude parameter dimension. The frame differences are accumulated chronologically to form the attitude path length value for that rollback segment. The attitude path length value describes the cumulative magnitude of attitude changes within the rollback segment. In addition to path length statistics, the frequency of occurrence of rollback segments within the same training process is counted, and the occurrence sequence number of each rollback segment during the training process is recorded. The path length values, frequency of occurrence, and corresponding segment numbers of all rollback segments are uniformly organized to form an attitude rollback record set. This attitude rollback record set provides quantitative input data for subsequent redundant path analysis and training stability evaluation.

[0030] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the path length data and training timeline information in the attitude return record set, extract the total path value of each backtracking path, and normalize it after performing a difference operation with the standard path value of the corresponding training stage. Summarize the normalization results according to the training stage to generate a redundancy path ratio index. Each backtracking segment record in the attitude reversal log set is read, and the corresponding attitude path length value is extracted. This value is then combined with the training timeline information to determine the training phase to which the backtracking segment belongs. The division of training phases is based on the pre-defined phase time boundaries of the training task, ensuring that each backtracking segment can be clearly assigned to a unique training phase. After phase assignment, the attitude path statistics corresponding to the same training phase in the evaluation phase are retrieved to obtain the standard path value for that phase. The standard path value is obtained by summarizing the path length statistics of multiple target task samples within the evaluation phase. It is used to characterize the typical attitude path level in a stable state during that training phase. The path length value of the backtracking segment is compared with the standard path value of the corresponding phase to obtain a path difference quantity reflecting the relative deviation of the backtracking segment. The path difference quantity is further proportionalized to the standard path value to form a redundancy path ratio. The redundancy path ratio describes the relative redundancy of the backtracking path within that training phase. The redundancy path ratios of all backtracking segments are summarized and organized by training phase to form a set of redundancy path ratio indicators, providing data input with a unified dimension for subsequent unstable state analysis.

[0031] S302: Based on the redundant path ratio index, filter the ratio intervals that are greater than the instability judgment threshold, extract the time period sequence within the corresponding interval, and combine the number of backtracking times and path fluctuation frequency within the sequence to construct a feature vector set of unstable state and obtain the instability feature coefficient. After the redundant path ratio index set is formed, features are constructed for unstable states during training. The redundant path ratio index set is scanned chronologically, and segments with redundant path ratios exceeding a preset instability threshold are extracted to form a candidate unstable time period sequence. Each candidate time period corresponds to a continuous interval of abnormal redundant path ratios. For each candidate time period, the number of backtracking segments in the attitude reversal record set within that time period is counted to reflect the frequency of attitude reversal corrections. Simultaneously, the attitude difference sequence is read within that time period, and the number of significant fluctuations in attitude difference is counted. Combined with the duration of the time period, a path fluctuation frequency index is obtained. The redundant path ratio, number of backtrackings, and path fluctuation frequency undergo unified dimension processing before aggregation to ensure comparability of different types of indicators on a numerical scale. A weighted aggregation operation is performed on the three types of indicators according to preset weights to generate a feature vector describing the degree of instability in that time period. The feature vectors corresponding to all candidate time periods are organized into an instability feature coefficient set, providing input for subsequent comprehensive stability analysis.

[0032] S303: Combining the instability feature coefficient, calling the angular velocity sequence data during the training process, filtering out segments with continuous changes in angular velocity values ​​between adjacent frames that are less than the stability threshold, counting the number of continuous frames in the stable segments, and calculating the ratio of the number to the total number of frames in the entire training process to establish the training stability analysis results. During training, angular velocity sequence data is continuously collected. Angular velocity data and attitude data are recorded using the same sampling time point order to ensure timeline consistency. Before analysis, time stamp verification and outlier removal are performed on the angular velocity sequence to ensure the continuity and validity of angular velocity change recording. The change amplitude of angular velocity values ​​at adjacent time points in the sequence is calculated to obtain the angular velocity change amplitude sequence. The angular velocity change amplitude is compared with a preset stability threshold. When the change amplitude is consistently below the stability threshold for a continuous time period, that time period is determined as a stable segment. The number of continuous frames in each stable segment is counted, and the number of continuous frames in all stable segments throughout the entire training process is accumulated. The accumulated number of stable frames is proportionalized to the total number of frames in the entire training process to obtain the stable segment proportion. The stable segment proportion is correlated with the instability characteristic coefficient of the corresponding time period to form the training stability analysis result. This analysis result reflects the comprehensive distribution of stable and unstable attitude states during training.

[0033] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the instability feature coefficient and stable segment ratio from the training stability analysis results. Based on the instability ratio threshold and ratio judgment threshold preset in the training workstation, perform interval assignment judgment on the two indicators respectively, generate a binary judgment label sequence, and obtain stability judgment label. The instability characteristic coefficients and the proportion of stable segments in the training stability analysis results are read time-by-time and compared with preset instability ratio thresholds and proportion judgment thresholds. Based on the comparison results, time periods with instability characteristic coefficients exceeding the thresholds are marked as unstable states, and time periods with stable segment proportions below the thresholds are marked as low-stability states. These two types of state labels are combined within the same time period to generate corresponding stability judgment labels. The stability judgment labels are recorded in binary form to distinguish between training time periods that meet and do not meet the stability conditions. The stability judgment labels for all time periods are arranged in chronological order to form a stability judgment label sequence. This sequence serves as an important basis for subsequent parameter adjustment analysis and can intuitively reflect the temporal distribution of stable states during training.

[0034] S402: Based on stability judgment, filter sequence segments whose label values ​​do not meet the conditions, extract the parameter fluctuation amplitude and feedback frequency during the training process in the corresponding time period, perform amplitude level mapping operation on the two, establish parameter adjustment interval feature values, and generate parameter update trigger signals. The time periods marked as not meeting the conditions are extracted from the stability judgment marker sequence, and the time periods that do not meet the conditions and are continuously distributed on the time axis are merged to form parameter adjustment candidate segments. For each candidate segment, the attitude difference sequence within the corresponding time range is read, and the fluctuation amplitude of the attitude parameters within the segment is statistically analyzed. The attitude parameter fluctuation amplitude is obtained by comparing the maximum and minimum values ​​of the attitude difference within the segment, which is used to reflect the concentration of attitude changes within the segment. At the same time, the frequency of feedback information within the segment is statistically analyzed. The feedback information comes from the aforementioned rollback segment records and attitude difference anomaly markers. The attitude parameter fluctuation amplitude and feedback frequency are mapped to discrete levels, and the mapping rules are completed according to preset intervals. The combination of the two levels forms the parameter adjustment interval feature value. Based on the parameter adjustment interval feature value, it is determined whether to generate a parameter update trigger signal, and the corresponding time position is recorded for each trigger signal.

[0035] S403: Call parameters to update the trigger signal, determine whether the trigger point meets the level allocation standard according to the training period sequence, set the corresponding level code for the time point that meets the condition, and build a rhythm mapping table in combination with the basic rhythm parameters corresponding to the training stage to establish rhythm level control parameters. After the parameter update trigger signal is generated, it is read sequentially over time, and its distribution is statistically analyzed within a sliding time window to reflect its density. The characteristic value of the parameter adjustment interval corresponding to each trigger signal describes its intensity. The trigger density and intensity are jointly determined, and a corresponding level code is generated based on a preset level classification rule. The level code characterizes the rhythm control level required for the current training period. The level code is then associated with the basic rhythm parameters corresponding to the training phase to construct a rhythm mapping relationship. The mapping result is written into the rhythm level control parameters, providing a clear basis for subsequent control command generation.

[0036] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the level code and rhythm mapping table in the rhythm level control parameters, extract the corresponding rhythm frequency value and tolerance band boundary value, bind the two parameters into a continuous parameter group structure according to the time series, and encapsulate them into a unified format data field in combination with the data transmission protocol format requirements to generate a control instruction data packet; The rhythm level control parameters are read, and the recorded level codes and corresponding rhythm mapping relationships for each training stage are extracted. The rhythm mapping relationship explicitly records the rhythm frequency values ​​corresponding to different level codes and the allowed rhythm fluctuation range boundaries. The corresponding rhythm frequency parameters are matched based on the level codes, and the lower and upper boundary values ​​corresponding to each rhythm frequency are extracted to form a complete set of rhythm control parameters. The rhythm control parameter set is bound to trigger time points in chronological order, with each set of parameters corresponding to a clearly defined effective start time and duration range. To ensure the continuity of parameters on the timeline, a parameter connection relationship is established between adjacent trigger time points. When no new rhythm level control parameters are generated within an adjacent time period, the most recently effective rhythm control parameters are used. After time binding is completed, the rhythm frequency parameters and tolerance band boundary parameters are uniformly structured and encapsulated into a continuous parameter group structure. The continuous parameter group structure contains a time identifier field, a rhythm frequency field, a lower boundary field, and an upper boundary field. These fields are arranged in a predetermined order. The continuous parameter group structure is formatted and encapsulated according to data transmission protocol requirements, converting all fields into a unified format data field set to form a control instruction data packet. The control instruction data packet is used to describe the rhythm control requirements to be executed within a specific time period, and serves as the input object for subsequent instruction sending and execution feedback collection.

[0037] S502: Based on the control command data packet, send the data fields to the VR host and motion capture component respectively, detect the command execution feedback information responded by the device during the task phase, filter the key frame number and rhythm response interval in the feedback content, summarize the feedback information in the current execution phase, and obtain a set of feedback segments. After the control command data packet is generated, the data fields are sent to the execution terminals participating in the interaction during training in a predetermined order according to the time identifier and parameter fields recorded in the control command data packet. The control commands maintain parameter integrity during transmission, ensuring that the rhythm frequency and tolerance band boundary values ​​are not rearranged or missing during transmission. After the control commands enter the execution phase, the command execution feedback information returned by the execution terminals is continuously monitored within the corresponding task time range. The feedback information includes the actual rhythm response record and key time markers related to the training posture. The feedback information is parsed, data items related to keyframe numbers are filtered, and response interval information reflecting the rhythm execution status is extracted. The rhythm response interval is obtained by determining the continuity of the rhythm response in the feedback, and is used to describe whether the rhythm remains stable within the allowable range during the execution phase. Feedback information collected within the effective time range of the same control command is summarized and organized to form a set of feedback segments. Each feedback segment corresponds to a clear start and end time range and includes the set of keyframe numbers and rhythm response interval markers within that time range.

[0038] S503: Combine the feedback fragment set, extract the training path data corresponding to the frame number, and perform frame position matching processing with the control parameters bound before sending. Based on the feedback matching rate and execution duration, construct the closed-loop execution segment sequence of the training process and establish a training task closed-loop configuration group. Each feedback segment in the feedback segment set is read one by one. Based on the keyframe number information recorded in the feedback segment, the corresponding training path data is retrieved. The training path data comes from the real-time posture sequence. The path data reflects the posture change trajectory of the training object within the corresponding time range in chronological order. The training path data corresponding to the keyframe is aligned in time with the rhythm control parameters generated in step 501 to ensure that the path data and control parameters are under the same time reference. After time alignment, the keyframe hit rate is statistically analyzed, recording the actual number of hit keyframes and the number of keyframes that should have been hit, and the keyframe hit rate is calculated accordingly. At the same time, combined with the rhythm response interval marker in the feedback segment, it is determined whether the rhythm control within the execution segment remains within the allowable range. The keyframe hit rate and rhythm response status are used together to construct the feedback matching rate index. The parameter effective duration recorded in the control command data packet is further read, and the feedback matching rate index and execution duration are jointly judged. Execution segments that meet the preset matching conditions and duration requirements are organized into effective closed-loop execution segments. All effective closed-loop execution segments are arranged in chronological order to form a closed-loop execution segment sequence. The closed-loop execution segment sequence records the rhythm level, keyframe hit status, and path change characteristics corresponding to each segment, and establishes a training task closed-loop configuration group to fully describe the closed-loop correspondence between control commands, execution feedback, and posture changes during the training process.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for hand function rehabilitation assessment and training based on virtual reality technology, characterized in that, Includes the following steps: S1: Based on the target task samples extracted during the evaluation phase, the training workstation constructs a reference posture sequence using wrist joint direction and finger joint angle data, combines the sampling time to form a target frame structure, and generates a posture comparison benchmark group. S2: Based on the posture comparison benchmark group, the real-time posture during the training process is compared frame by frame, the trend of posture difference change is extracted, the backtracking state is identified, the corresponding path length and backtracking number are recorded, and a posture backtracking record set is generated. S3: Using the path and event data in the attitude reversal record set, the training workstation calculates the redundant path ratio and extracts instability features. At the same time, it identifies the continuous stable angular velocity segment, calculates its duration ratio, and outputs the training stability analysis results. S4: Call the instability index and the proportion of stable segments in the training stability analysis results, combine them with the preset judgment threshold, determine whether to update the parameters, and generate rhythm level control parameters; S5: Construct a control instruction data packet based on the rhythm level control parameters, including rhythm frequency and tolerance band parameters, and send it from the training workstation to the VR host and motion capture component. After the device performs the adjustment task, it returns a feedback segment and generates a training task closed-loop configuration group.

2. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 1, characterized in that, The posture reference set includes wrist joint direction information, knuckle angle data, and target frame structure; the posture return record set includes posture difference change trend, return status, path length, and number of return times; the training stability analysis results include redundant path ratio index, instability characteristics, continuous stable angular velocity segments, and duration percentage; the rhythm level control parameters include instability index, stable segment percentage, and judgment threshold; the training task closed-loop configuration set includes control instruction data package, rhythm frequency parameters, tolerance band range, and feedback segments.

3. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Based on the target task samples extracted during the evaluation phase, call the wrist joint direction information and finger joint angle data, and synchronize them according to the sampling time point to construct the wrist joint and finger joint posture matching matrix at the time point and generate a synchronized posture matrix group. S102: Based on the synchronized attitude matrix group, extract the attitude change trajectory at each time point, compare the attitude change amplitude with the overall change level, filter the frame data with prominent changes, and combine the corresponding time point and attitude matrix to generate a key frame attitude sequence group. S103: For the key frame pose sequence group, extract the angle parameters of the wrist joint and finger joints, set the interval boundary values ​​according to the distribution interval, and construct the pose reference frame structure in combination with the key frame time points to generate a pose comparison benchmark group.

4. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Based on the posture reference group, collect real-time posture information during the training process, and compare the real-time posture frames with the corresponding reference frames in the posture reference group frame by frame according to the sampling time point order. Perform difference calculation on the wrist joint direction parameters and finger joint angle parameters, summarize the frame difference amount and arrange them in time order to generate a posture difference sequence. S202: Based on the attitude difference sequence, calculate the magnitude of the difference between adjacent frames, and judge the direction of change according to the preset attitude change judgment threshold. Mark the frame segments where the continuous change direction reverses as back segments, and record the start and end time points of the corresponding frame segments to obtain the back segment identifier set. S203: For the set of backtracking segment identifiers, call the attitude path data of consecutive frames within the segment, calculate the path length value of the backtracking segment, and count the number of times the backtracking segment appears in the same training process. Summarize and record the path length value and the number of backtracking according to the segment number to establish an attitude return record set.

5. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 4, characterized in that, The setting of the posture change judgment threshold is specifically to limit the change amplitude of wrist joint direction parameter and the change amplitude of finger joint angle parameter to not less than the preset first amplitude threshold and second amplitude threshold, respectively. The first amplitude threshold and the second amplitude threshold are set according to the movement difficulty level of different rehabilitation training stages. When generating the backtracking segment identifier set, it is determined that there is a frame segment with the continuous change direction reversed only when the change amplitude of the difference between adjacent frames simultaneously meets the first amplitude threshold and the second amplitude threshold. The start and end times of each frame segment in the rollback segment identifier set are determined by identifying the time point when the direction of the difference change first reverses and the time point when it first recovers to the same direction as before the reversal.

6. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the path length data and training process timeline information in the attitude return record set, extract the total path value of each backtracking path, and perform difference calculation with the standard path value of the corresponding training stage, and then normalize it. Summarize the normalization results according to the training process stage to generate a redundancy path ratio index. S302: Based on the redundant path ratio index, filter the ratio intervals that are greater than the instability judgment threshold, extract the time period sequence within the corresponding interval, and combine the number of backtracking times and the path fluctuation frequency within the sequence to construct a feature vector set of unstable state and obtain the instability feature coefficient. S303: Combining the aforementioned instability characteristic coefficients, call the angular velocity sequence data during the training process, filter out segments with continuous changes less than the stability threshold based on the angular velocity change values ​​between adjacent frames, count the number of continuous frames in the stable segments, and calculate the ratio of the number to the total number of frames in the entire training process to establish the training stability analysis results.

7. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 6, characterized in that, The specific steps of S4 are as follows: S401: Call the instability feature coefficient and stable segment ratio value in the training stability analysis results, and perform interval assignment judgment on the two indicators according to the preset instability ratio threshold and ratio judgment threshold in the training workstation, generate a binary judgment label sequence, and obtain stability judgment label; S402: Based on the stability judgment marker, filter the sequence segments whose marker values ​​do not meet the conditions, extract the parameter fluctuation amplitude and feedback frequency during the training process in the corresponding time period, perform amplitude level mapping operation on the two, establish parameter adjustment interval feature values, and generate parameter update trigger signals. S403: Call the parameters to update the trigger signal, determine whether the trigger points meet the level allocation standard according to the training period sequence, set the corresponding level code for the time points that meet the conditions, and construct a rhythm mapping table in combination with the basic rhythm parameters corresponding to the training stage to establish rhythm level control parameters.

8. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 7, characterized in that, Specifically, the instability ratio threshold is set in three numerical ranges according to the precision of the training task. Each level of the instability ratio threshold corresponds to a tolerance standard for different training precision requirements. When generating the stability judgment label, the corresponding position in the binary judgment label sequence is marked as not meeting the condition only if the instability feature coefficient exceeds the instability ratio threshold of the corresponding training task level and the proportion of the stable segment is lower than the proportion judgment threshold. The mapping operation between the parameter fluctuation amplitude and the feedback frequency amplitude level is as follows: according to the five-level amplitude mapping standard set in the training workstation, the parameter fluctuation amplitude and the feedback frequency are respectively classified into the corresponding amplitude level intervals, and the characteristic value of the parameter adjustment interval is determined by the product of the amplitude level corresponding to the parameter fluctuation amplitude and the feedback frequency.

9. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 7, characterized in that, The specific steps of S5 are as follows: S501: Call the level code and rhythm mapping table in the rhythm level control parameters, extract the corresponding rhythm frequency value and tolerance band boundary value, bind the two parameters into a continuous parameter group structure according to the time series, and encapsulate them into a unified format data field in combination with the data transmission protocol format requirements to generate a control instruction data packet; S502: According to the control instruction data packet, the data fields are sent to the VR host and the motion capture component respectively, and the instruction execution feedback information responded by the device in the task phase is detected. The key frame number and rhythm response interval in the feedback content are filtered, and the feedback information in the current execution phase is summarized to obtain a set of feedback segments. S503: Combine the feedback fragment set, extract the training path data corresponding to the frame number, and perform frame position matching processing with the control parameters bound before sending. Based on the feedback matching rate and execution duration, construct the closed-loop execution segment sequence of the training process and establish a training task closed-loop configuration group.

10. The hand function rehabilitation assessment and training method based on virtual reality technology according to claim 9, characterized in that, The rhythm frequency values ​​contained in the rhythm mapping table are set in the range of 0.2Hz to 2.0Hz, and the correspondence between the rhythm level code and the rhythm frequency value is constructed in steps of 0.1Hz. The tolerance band boundary value is set according to the error tolerance standard corresponding to different rhythm levels, and the upper and lower boundary values ​​are determined by extending 0.05Hz up and down according to the frequency level. Each group in the continuous parameter group structure contains a rhythm level code, a rhythm frequency value and a pair of tolerance band boundary values. When the rhythm frequency value and the tolerance band boundary value are encapsulated into a unified format data field, the fields are arranged in order of high byte first and low byte last according to the data transmission protocol format requirements, and a single byte field identifier is added to each group of data to form a data field structure with a total length of fixed bytes.

Images