Digital rehabilitation guidance system

Abstract

The invention relates to the technical field of medical care informatics, and discloses a digital rehabilitation guidance system which comprises the following steps: decoupling an inertial signal into two paths of independent data streams representing target joint movement and reference part postures; monitoring a fluctuation intensity index of the reference data stream in real time; by means of compliance gating logic, an evaluation process for target joint data is activated only when the fluctuation intensity is lower than a validity threshold value, otherwise, output is blocked, and invalidation is marked; according to the method, a reverse negative mechanism based on non-target channel background noise is established, pseudo standard data caused by compensation leveraging is effectively eliminated, the problem of rehabilitation assessment distortion in a weak supervision environment is solved, and the data credibility and the training safety are improved while the computing power load is reduced.

Description

Technical Field

[0001] This invention relates to a digital rehabilitation guidance system, belonging to the field of healthcare informatics technology. Background Technology

[0002] In current remote and home-based digital rehabilitation technology systems, data processing generally adopts trajectory similarity fitting logic. By collecting time series data of movement of the affected limb, dynamic time warping algorithm is used to calculate the geometric similarity with the gold standard trajectory in the standard database, and rehabilitation scores or guidance feedback are generated.

[0003] However, in unsupervised, uncontrolled home settings, the fundamental assumption that correct trajectory morphology equates to rehabilitation faces challenges. During neuromuscular control remodeling, patients often resort to non-target areas like the trunk and scapula to participate in movement due to insufficient strength or control impairment of the target muscle groups. This pathological compensatory behavior creates the trajectory amplitude at the limb extremities. The geometric features of the data stream generated by these compensatory movements highly fit a standard template, making it difficult to distinguish between true and false signals from a single target channel, leading to false positive assessments. This results in ineffective rehabilitation training and induces patients to solidify incorrect movement patterns. Existing technologies attempt to improve this at the hardware level. For example, the utility model patent CN215195282U discloses a digital rehabilitation training platform, which uses a lifting component in conjunction with an adjustable interactive panel from 0 to 90 degrees to achieve patient-controlled sitting, standing, and wheelchair-assisted postures. While static adaptation to the same scenario is possible, solutions that focus on physical ergonomic adjustments only address static guidance for body alignment and lack dynamic control over the movement process. The system cannot perceive in real time the compensatory force signals from non-target areas during dissection. Even when patients meet the static positioning requirements, they still complete tasks through incorrect compensatory force patterns, rendering rehabilitation training ineffective and failing to guarantee the authenticity of neural remodeling. To address these issues, conventional industry improvements tend to increase the number of sensors to cover key nodes throughout the body or introduce high-performance visual models for full-pose 3D reconstruction. However, under the constraints of universal medical IoT device engineering, increasing sensors leads to higher hardware costs and greater complexity in wearing them. High-performance visual solutions are limited by the computing power bottlenecks in home environments, occlusion interference, and privacy sensitivity, making large-scale deployment difficult. Existing technologies are caught in a dilemma of not being able to balance the reliability of evaluation results with low cost.

[0004] Therefore, the technical problem to be solved by this invention is how to identify and eliminate compensatory invalid data through logical mining of limited sensor data streams without increasing hardware and computing load, thereby resolving the contradiction that low-cost equipment cannot achieve accurate rehabilitation assessment. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A digital rehabilitation guidance system, comprising: The channel mapping and decoupling module is used to receive the raw time-series signals collected by the inertial measurement unit and, based on a preset human kinematic topology model, decompose the raw signals into a first data stream representing the motion trajectory of the target joint and a second data stream representing the attitude of the associated reference part. The stability measurement calculation unit is used to perform time-domain energy analysis on the second data stream and calculate the state dispersion index that characterizes the intensity of unexpected fluctuations in the associated reference part within a unit time window. A logic arbitration controller is configured with compliance gating logic. The logic arbitration controller is used to monitor the state dispersion index in real time, and when the state dispersion index is less than a preset validity threshold, it generates an activation signal to trigger the amplitude evaluation logic for the first data stream, and when the state dispersion index is greater than or equal to the validity threshold, it generates a blocking signal to block the output of the amplitude evaluation logic and mark the current event as invalid. The threshold adaptive drift unit is used to count the frequency of invalid events and automatically increase the validity threshold based on the preset fatigue decay function in response to the rising trend of the occurrence frequency.

[0006] Preferably, the system further includes a dynamic validity verification unit, which is used to construct a velocity-position phase trajectory reflecting the motion state of the first data stream, and calculate the convergence slope of the angular velocity parameter within a preset control time window before the displacement amplitude represented by the first data stream reaches its peak value; in response to the angular velocity parameter not showing a trend of convergence to zero within the preset control time window or the convergence slope exceeding a preset controlled limit value, an inertial fuse signal is generated; the logic arbitration controller is also used to preferentially execute the event invalidation marking operation in response to the inertial fuse signal.

[0007] Preferably, the stability measurement calculation unit further includes a frequency domain arbitration logic module, which is used to: perform spectral decomposition on the second data stream to separate the low-frequency drift component located in the first frequency range and the high-frequency jitter component located in the second frequency range, wherein the first frequency range is lower than the second frequency range; calculate the first energy density of the low-frequency drift component and the second energy density of the high-frequency jitter component; the logic arbitration controller is also used to perform weighted determination: when the state dispersion index is greater than the validity threshold and the proportion of the second energy density in the total energy exceeds the preset exemption ratio, bypass the signal and maintain the active state of the amplitude evaluation logic.

[0008] Preferably, the system further includes a collaborative analysis logic module, which is used to: extract the first motion envelope features of the first data stream and the second motion envelope features of the second data stream, and calculate the cross-correlation coefficient of the first motion envelope features and the second motion envelope features within a single action window; the logic arbitration controller is also used to generate a collaborative compensation blocking signal in response to the cross-correlation coefficient exceeding a preset collaborative threshold value, and to block the output of the amplitude evaluation logic in response to the collaborative compensation blocking signal.

[0009] Preferably, the system further includes a reference state verification logic module, which is used to: extract the static gravity projection vector of the second data stream during the static gap before the motion event represented by the first data stream occurs, and calculate the spatial deviation angle between the static gravity projection vector and the preset standard anatomical zero vector; the logic arbitration controller is also used to release the monitoring lock of subsequent motion events only when the spatial deviation angle is within the preset allowable conical domain.

[0010] Preferably, the threshold adaptive drift unit performs automatic adjustment based on the following logic: a fatigue tolerance model that monotonically decreases over time is established, which associates the validity threshold with the cumulative duration of the current training cycle and the historical average compensation rate; under the premise that continuous invalid event markers are detected and the amplitude characteristics of the first data stream remain stable, the validity threshold is gradually increased by a preset step size until the frequency of invalid events decreases to a preset maintenance level.

[0011] Preferably, the logic arbitration controller is also configured with data integrity pre-check logic, which is used to: detect the timestamp synchronization of the first data stream and the second data stream before monitoring the execution state dispersion index; if the time deviation between the two is detected to exceed the preset synchronization tolerance, generate a resynchronization instruction to calibrate the data stream, and suspend the execution of the compliance gating logic before the calibration is completed.

[0012] Preferably, the channel mapping and decoupling module is also used to: dynamically select the source channel of the second data stream according to the preset rehabilitation action definition template; for upper limb rehabilitation actions, lock the data channel representing the trunk tilt angle as the source of the second data stream; for lower limb rehabilitation actions, lock the data channel representing the pelvic rotation angle as the source of the second data stream.

[0013] Preferably, the stability metric calculation unit uses the following weighted calculation logic when calculating the state dispersion index: Where D is the state dispersion index. This represents the variance of the low-frequency drift component. Let be the variance of the high-frequency tremor component, and α and β be the preset first weighting coefficient and second weighting coefficient, respectively, and satisfy α>β.

[0014] Preferably, the amplitude evaluation logic is specifically used to: calculate the geometric similarity or Euclidean distance of the first data stream relative to the preset rehabilitation target trajectory within the effective time window of receiving the activation signal; when the activation signal continues to exist and the geometric similarity exceeds the preset pass score, generate and render a feedback signal representing the qualified action on the user terminal; if an blocking signal is received during the action, terminate the current similarity calculation process and output an error correction instruction prompting the user to correct the posture, instead of outputting an action score.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the digital rehabilitation guidance system, a data gating architecture based on non-target channel states is constructed. The rehabilitation movement evaluation logic is extended from single target trajectory fitting to background posture stability constraints. By utilizing the temporal discreteness characteristics of the second data stream representing the associated body parts, logical circuit breaking is performed on the first data stream representing the target joint, establishing a logical mutual exclusion relationship between signal sources. Without the need to construct a high-computational-power whole-body skeletal model, the low-dimensional sensor data stream is used to identify and eliminate false achievement data generated by trunk leverage or posture compensation. The data processing architecture reduces the consumption of computing resources at the edge, blocking invalid data from entering the rehabilitation assessment record from the source. This solves the technical problem of low signal-to-noise ratio and evaluation distortion of rehabilitation data caused by patient compensatory behavior in a home-based, weakly supervised environment.

[0016] 2. A dynamic verification logic for the movement process is introduced. By constructing the velocity-position phase trajectory of the target joint movement, the convergence characteristics of angular velocity within the time window before the movement reaches its peak are monitored. This distinguishes between controlled muscle contraction and uncontrolled inertial swinging. The logic deepens the granularity of rehabilitation data assessment from simply achieving the positional result to the quality of process control. The system records and feeds back quasi-static motion data with clear neural control characteristics. The mechanism uses temporal logic constraints at the algorithm level to prevent patients from using physical inertia to deceive the system and generate data distortion, ensuring that the output rehabilitation progress data truly reflects the patient's active control ability over the target muscle groups.

[0017] 3. Establish a spectral separation-based differential compliance judgment standard. Utilize the frequency domain distribution differences between postural compensatory drift and muscle-limit tremor, and perform frequency-band energy density analysis on the background data stream. Based on this, adjust the gating logic blocking threshold. This strategy enhances the system's ability to analyze disturbance signals of different natures. When blocking low-frequency, large-amplitude compensatory movements, a logic exemption is implemented for high-frequency, micro-amplitude physiological tremors. Through refined spectral arbitration, the system avoids the accidental deletion of high-intensity training data due to simple threshold judgment, thereby improving the evaluation algorithm's inclusiveness and accuracy in discriminating the patient's true physiological state under complex pathological conditions. Attached Figure Description

[0018] Figure 1 This is a system logic architecture diagram of the present invention based on non-target channel decoupling and adaptive gating; Figure 2 This invention provides a frequency domain energy analysis and weighted calculation diagram to distinguish between compensatory and physiological tremors. Figure 3 This is a sequence diagram of the compliance gating logic and interaction timing in the action effectiveness evaluation of this invention. Detailed Implementation

[0019] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0020] This invention provides a digital rehabilitation guidance system, including channel mapping and decoupling, background channel stability measurement, multidimensional logic arbitration, and adaptive threshold adjustment under fatigue conditions. The channel mapping and decoupling module is responsible for processing the time-series raw signals collected by the inertial measurement unit, including triaxial acceleration and triaxial angular velocity data with a sampling frequency of 50Hz to 100Hz. Based on a preset rehabilitation movement definition template, this module dynamically locks the logical attributes of the data source. In upper limb movements such as shoulder abduction training, the system locks the sensor data worn on the distal humerus as the first data stream to characterize the target joint movement trajectory; it locks the sensor data worn on the thoracic spine or pelvis as the second data stream to characterize the posture of the associated reference part, i.e., the trunk. The decoupling operation uses a quaternion rotation matrix to project the raw data in the sensor coordinate system to an anatomical reference coordinate system based on the gravity vector, outputting two kinematically independent data streams. The first data stream contains large-amplitude motion signals reflecting the displacement of the limb end, while the second data stream presents a quasi-static signal approaching zero under standard movements.

[0021] The stability measurement calculation unit performs a hybrid time-domain and frequency-domain feature analysis on the second data stream, calculating the state dispersion index that characterizes the intensity of unexpected fluctuations at the associated reference location. The calculation process follows a sliding time window logic, with the system setting a time window T of length such as 0.5s. w Within this window, the acceleration modulus or Euler angle data of the second data stream are sampled. For each sampling window, its variance or standard deviation is calculated to quantify the dispersion of the signal. To distinguish between compensatory trunk swaying and physiological tremor, the system uses weighted calculation logic to determine the state dispersion index. The specific calculation formula is as follows: Where D is the state dispersion index; The variance of the low-frequency drift component separated after low-pass filtering at a cutoff frequency of 3Hz represents the compensatory swaying of the torso. The variance of the high-frequency tremor component separated after bandpass filtering (e.g., 8Hz to 12Hz) represents the physiological tremor caused by muscle fatigue or the limits of neural control. α and β are the preset first and second weighting coefficients, respectively, and satisfy the relationship α>β. For example, if α=0.8 and β=0.2 are set, this weighting logic sets a higher penalty weight for low-frequency large-amplitude compensation signals.

[0022] The logic arbitration controller, as the core decision-making unit, is equipped with compliance gating logic to perform data validity screening at the signal source. The controller monitors the calculated state dispersion index D in real time and compares it with the preset validity threshold D. th A comparison is performed, and this comparison process constitutes a logical mutual exclusion gate: when D is satisfied... <D th When conditions are met, the logic arbitration controller generates an activation signal, connecting the first data stream to the amplitude evaluation logic, allowing the system to calculate rehabilitation indicators such as the target joint's motion amplitude or trajectory similarity and generate rehabilitation progress data. Generating the activation signal requires synchronously sending a transmission enable instruction to the direct memory access controller, writing the first data stream to the shared memory circular buffer, calling the vector multiplication unit in the digital signal processor, and performing a dot product operation between the input data and a mask matrix consisting entirely of 1s. This ensures that the subsequent amplitude evaluation logic reads complete original frame data. When D≥D is detected... th At this time, the controller flips the logic state and generates a blocking signal, shields the output of the amplitude evaluation logic, directly marks the motion events within the current time window as invalid and triggers error correction instructions. The blocking signal triggers the underlying hardware interrupt routine, injects a zero mask matrix into the rendering buffer, performs bitwise operations on the target joint coordinate vector and the Boolean NOT value of the blocking flag, the graphics processing unit skips the vertex update instruction of the current frame, performs a null operation (NOP) overwrite using the residual data of the previous frame's video memory, and cuts off the transmission of invalid data to the user terminal at the physical link level to prevent erroneous display caused by software layer logic delay. This mechanism ensures that rehabilitation scoring is only performed under the effective premise that non-target parts remain relatively still.

[0023] The dynamics validity verification unit is used to identify ballistic cheating behavior, such as using explosive force to swing limbs to deceive and achieve the required angle. Within a preset control time window (e.g., 200ms before the peak value of the displacement amplitude represented by the first data stream), this unit constructs a velocity-position phase trajectory and monitors the convergence characteristics of the angular velocity parameter. The verification logic calculates the first derivative of the angular velocity in the first data stream, i.e., angular acceleration. In controlled rehabilitation exercise, the angular velocity exhibits a linear or quasi-linear monotonically decreasing trend when the limb approaches its maximum range of motion. If the angular velocity parameter does not show a linear or quasi-linear monotonically decreasing trend within the preset control time window... If the trend towards zero or the absolute value of the convergence slope of the angular velocity exceeds the preset controlled limit, indicating an impact-induced emergency stop or overshoot due to inertia, the unit generates an inertial fuse signal. The logic arbitration controller responds to this signal by prioritizing the execution of an invalid event marker. The collaborative analysis logic module is used to identify smooth compensatory patterns where the target limbs and torso move in a highly synchronized rhythm. The module extracts the motion envelope features of the first and second data streams within a single action window, which can be extracted using Hilbert transform. The cross-correlation coefficient between the two envelope sequences is then calculated. In normal rehabilitation exercises, the target limbs move while the trunk remains stationary; the two are statistically independent in terms of temporal sequence. If the value approaches zero, the calculated cross-correlation coefficient... If the value exceeds the preset coordination threshold, such as 0.6, it indicates that the patient is using rhythmic trunk swinging to assist limb movement. At this time, the logic arbitration controller generates a coordination compensation blocking signal to block the output of the amplitude evaluation logic.

[0024] The threshold adaptive drift unit establishes a fatigue tolerance model that evolves over time and based on historical data performance to address the decline in control ability caused by muscle fatigue during rehabilitation training. This unit counts the frequency of invalid events; if an increasing trend in the frequency of invalid events is detected over a continuous time period, and the amplitude characteristics of the first data stream remain relatively stable, the system determines that the patient has entered a fatigue phase. The fatigue decay function follows a nonlinear drift model. Here, I is a Boolean indicator variable representing the occurrence of invalid events, and λ is the fatigue coefficient calculated based on the historical maximum endurance measured by the user's offline calibration. The calculation logic only activates the integration term when the integration ratio of the high-frequency component power spectral density S(f) in the 8-12Hz range exceeds 30%, thereby filtering out the threshold false rise caused by non-fatigue compensation actions. Based on the preset fatigue decay function, the unit responds to this trend by automatically increasing the effectiveness threshold D by a preset step size, such as increasing by 5% each time. th This drift mechanism ensures training safety while maintaining patient confidence and avoids consecutive negative scores due to physiological limits. The baseline verification logic module is used to eliminate systematic measurement errors caused by slight sensor movement or initial body position deviation. During the static interval before the motion event represented by the first data stream, such as 1 to 2 seconds before the start of the movement, this module activates a static scan of the second data stream. The module extracts the mean acceleration vector of the second data stream during this period and defines it as the current static gravity projection vector. It then calculates the spatial deviation angle θ between this vector and the preset standard anatomical zero vector, i.e., the gravity direction under ideal sitting or standing posture. dev The logic arbitration controller only applies to θ dev Located within a preset permissible conical region, such as a semi-apex angle of 10°. ∘ When the object is inside the cone, the monitoring lock for subsequent motion events is released. If the deviation angle exceeds the limit, the system suspends the evaluation process and outputs an attitude reset prompt.

[0025] Example 1: In a home rehabilitation training scenario for a stroke patient with hemiplegia, the patient is in the transitional stage from flaccid paralysis to spasticity. The abductor muscle strength of the affected upper limb shoulder joint is only grade 2, accompanied by severe scapular compensatory habits. When the patient performs shoulder abduction training, due to insufficient strength in the middle deltoid muscle, the patient subconsciously tilts the trunk to the healthy side and shrugs to pull the affected arm up. At this time, if a traditional inertial system that only monitors the trajectory of the affected limb's distal end is used, the sensor will record a seemingly acceptable movement from 0... ∘ Rise to 90 ∘The smooth arc trajectory of the movement leads to a misjudgment of the action as effective and positive feedback is given. This misjudgment will cause the patient to continuously reinforce the incorrect compensatory movement pattern. When the digital rehabilitation guidance system of this invention faces this situation, the channel mapping and decoupling module separates the collected inertial data stream. The first data stream generated by the sensor worn on the affected upper arm shows a large angular displacement signal. At the same time, the stability measurement calculation unit monitors the second data stream locked in the thoracic spine segment of the trunk in real time. The system performs weighted dispersion calculation on the second data stream within the sliding time window. Due to the patient's trunk tilting and leveraging behavior, the energy of the low-frequency drift component in the second data stream increases sharply, causing the calculated state dispersion index D to quickly exceed the preset effectiveness threshold D at the beginning of the movement. th .

[0026] At this point, compliance gating logic plays a crucial role. Although the trajectory of the first data stream closely matches the standard template, due to D≥D... th When the condition is triggered, the logic arbitration controller generates a blocking signal, shielding the output of the amplitude evaluation logic. The system does not output misleading feedback indicating action completion; instead, it triggers a correction command to keep the trunk upright based on the abnormal characteristics of the second data stream. This background noise-based reverse rejection mechanism resolves the contradiction between trajectory achievement and force error within a single system architecture. That is, without adding whole-body motion capture equipment, it achieves deep logical verification of the target channel's motion quality through rigid constraints on the silence level of non-target channels. As training continues, the patient's control ability decreases due to muscle fatigue, and the trunk begins to exhibit slight, involuntary swaying. The frequency domain analysis logic in the stability measurement calculation unit identifies this fluctuation as mainly concentrated in the high-frequency range of 8Hz to 12Hz, belonging to physiological tremors under muscle strength limits rather than malignant compensation. At this point, the smaller second weighting coefficient β in the weighted calculation formula keeps the calculated D value within the range of D. th The following or threshold adaptive drift unit is based on a fatigue model with D appropriately increased. th The system maintains the activation of the amplitude assessment logic, recognizes the patient's effective efforts in a state of fatigue, and ultimately, through the coordinated operation of the above logic, the system enables the patient to complete the movement by relying solely on the target muscle groups in a home-based, low-monitored environment while inhibiting compensation, thereby ensuring that every recorded rehabilitation data has clear neural remodeling value.

[0027] Example 2: This example verifies the core discrimination capability, anti-interference performance, and adaptive fault-tolerance mechanism of the technical solution of the present invention in a real engineering environment. The experimental design constructs a three-act evidence chain containing a multi-dimensional control system to confirm the advantages of the non-target channel noise rejection and synergistic compensation analysis mechanism compared with traditional technologies. The experimental platform consists of a high-precision optical motion capture system as the anatomical truth and a prototype machine integrating the algorithm of the present invention. The sampling frequency of the optical system is set to 200Hz. The prototype machine contains two inertial measurement units (IMUs), which are worn on the lateral side of the humerus of the subject's right arm (target channel) and the T4 segment of the thoracic spine (non-target channel), respectively. The sampling frequency is 100Hz. In order to simulate the electromagnetic noise and micro-motion interference in a real home environment, Gaussian white noise with a signal-to-noise ratio of 20dB is actively superimposed on the original IMU signal.

[0028] Sliding time window length T w Based on spectral analysis of typical compensatory movements such as shoulder shrugs and tilting (typically <3Hz), and combined with the Nyquist sampling theorem, T... w The time was set to 0.5s to achieve effective smoothing filtering while maintaining capture sensitivity; the effectiveness threshold D... th Initial D th The value was set to 0.050 (dimensionless normalized value), derived from statistical analysis of the baseline variance of thoracic spine acceleration during simulated standard movements in 30 healthy subjects, covering 99.7% of the normal physiological fluctuation range; synergistic threshold value. The value is set to 0.6. This value is based on the signal cross-correlation theory and is used to define whether there is an unexpected linkage between two independent physical processes. Ten subjects with different degrees of shoulder joint mobility limitation were recruited. Each subject had to complete three sets of prescribed movements in a noisy environment, corresponding to three typical movement patterns: Control group A (standard movement): Subjects kept their trunks still and used only the deltoid muscles to complete shoulder abduction movements under professional supervision, simulating the ideal rehabilitation state. Experimental group B (explicit compensation): Subjects simulated insufficient muscle strength and completed the movements by obvious trunk tilting, simulating common erroneous compensation patterns. Experimental group C (implicit synergy): Subjects simulated smooth compensation, with the limbs raised and the trunk slightly swaying in sync, simulating hidden pathological synergy patterns. The experiment collected raw sensor data under each set of movements and processed them using the algorithm of this invention to obtain key intermediate feature values. The table below presents a comparison of data from typical samples.

[0029] Table 1: Comparison of Key Intermediate Feature Data and Judgment Results Data Phenomenon and Underlying Mechanism Interpretation: Effective Blockage of Overt Compensation; in Experimental Group B, the maximum angle of the target channel reached 91.2°. ∘If judged solely by the traditional trajectory method, this action would be mistakenly deemed valid because the angle meets the standard. However, this invention detected that the state dispersion index D soared to 0.185, far exceeding the threshold D. th (0.050), which directly triggers the compliance gating logic and outputs an invalid judgment. The data shows that the dispersion index D can penetrate the trajectory illusion and accurately capture abnormal agitation in non-target parts; in experimental group C, the state dispersion index D is 0.045, slightly lower than D th If only the dispersion criterion is relied upon at this point, this action may be missed, but the cross-correlation coefficient... Reaching 0.78, exceeding the synergy threshold of 0.6, activated the synergy analysis logic, identifying a strong pathological synchronization between the target and background signals. Based on this, the system correctly outputs an invalidation judgment. To verify the system's adaptability during long-term training, subjects were arranged to undergo 30 minutes of continuous high-intensity repetitive training. During the fatigue phase from the 25th to the 30th minute, the subjects exhibited high-frequency muscle tremors.

[0030] Table 2: Comparison of Threshold Drift and Judgment Results in the Fatigue Stage Data Phenomenon and Underlying Mechanism Interpretation: During the fatigue stage, due to the introduction of physiological tremors, the calculated... If a value naturally increases, and a fixed threshold (0.050) is used, the system will incorrectly reject the subject's effective effort (false negative). However, in this invention, the threshold adaptive drift unit identifies an increase (65%) in the energy percentage of the tremor frequency band (8-12Hz) and a continuous training trend, automatically raising the effectiveness threshold to 0.065 (i.e., ...). This invention, through the dual logic of non-target channel noise rejection and cooperative cross-correlation analysis, can still achieve 100% interception of explicit and implicit compensatory actions in a noisy environment with a signal-to-noise ratio of only 20dB, thus eliminating the false positives of traditional trajectory fitting schemes.

[0031] Example 3: This example combines Figures 1 to 3 A description of a digital rehabilitation guidance system, such as... Figure 1As shown, the logical architecture mainly consists of a closed loop of data acquisition, processing, and arbitration feedback. The inertial measurement unit (IMU) is responsible for acquiring the raw time-series signals and transmitting them to the channel mapping and decoupling module. Based on the topology model, the signals are decomposed into a first data stream representing the motion of the target joint and a second data stream representing the attitude of the associated reference part. The second data stream enters the stability measurement calculation unit to calculate the state dispersion index reflecting the intensity of unexpected fluctuations and sends it to the logic arbitration controller. The controller is equipped with compliance gating logic to compare the state dispersion index with the validity threshold in real time: when the dispersion is lower than the threshold, an activation signal is generated to trigger the amplitude evaluation logic and output rehabilitation feedback; when the dispersion is greater than or equal to the threshold, a blocking signal is generated to shield the output and mark the event as invalid. At the same time, the threshold adaptive drift unit dynamically adjusts the validity threshold by statistically analyzing the invalid frequency and combining it with the fatigue model, and feeds it back to the logic arbitration controller, thereby forming a closed-loop control system with anti-fatigue characteristics.

[0032] like Figure 2 As shown, the horizontal axis represents frequency in Hz, and the vertical axis represents energy density. The figure depicts three key curves: the solid line in the 0-3Hz range represents the high-energy low-frequency drift component, mainly corresponding to compensatory trunk swaying; the dashed line in the 8-12Hz range represents the relatively low-energy high-frequency tremor component, corresponding to physiological muscle tremors; and the dotted line spanning the entire frequency band represents the weighted state dispersion D. This curve demonstrates how the system performs weighted calculations based on the energy characteristics of different frequency bands, thereby amplifying the impact of low-frequency drift on the dispersion index while preserving high-frequency tremor characteristics. Figure 3 As shown, the temporal logic from patient action to system feedback encompasses six interactive nodes: patient action execution, upper arm sensor, trunk sensor, dispersion calculation, compliance gating, and feedback system. The process begins with the patient performing a shoulder abduction movement accompanied by a trunk posture change. The upper arm sensor and trunk sensor output the first data stream (angular displacement signal) and the second data stream (posture signal), respectively. The dispersion calculation node performs sliding window variance calculation and low-frequency drift component extraction, generating a dispersion index D and transmitting it to the compliance gating node. Here, the process is divided into two mutually exclusive branches: in the case where the trunk remains stationary (i.e., D is low), D < threshold D0. th Under the condition that the system executes the evaluation path and ultimately outputs effective feedback from the feedback system, and in the case of compensatory leverage (i.e., D is high), D ≥ threshold D is satisfied. th Under certain conditions, the system will trigger a path blocking action and output a prompt to keep the torso upright via the feedback system.

[0033] Example 4: This example addresses the threshold adaptive adjustment and stability boundary under complex conditions in the aforementioned technical solutions, aiming to eliminate the potential black box of adaptive logic and ensure that the parameter evolution of the system has a deterministic and controllable procedure under long-term operation and extreme conditions. During rehabilitation training, the patient's muscle strength decline is not a linear process, and there are individual differences in anti-fatigue characteristics. To solve the subjective problem of setting fatigue compensation function parameters, this example constructs a standardized parameter calibration and dynamic evolution procedure based on individual baselines. This procedure performs baseline measurements during the initial system configuration phase. The system guides the patient to perform continuous repetitive movements at 50% to 70% of maximum voluntary contractile force (MVC) in an uncompensated and muscle-strength state until the movement amplitude naturally decays to 80% of the initial amplitude or the subjective fatigue score (RPE) reaches a preset threshold. The system records the time T at this point. fatigue and the total number of actions N fatigue And calculate the average attenuation rate λ of the action frequency. freq Based on this individualized data, the system initializes the key parameters of the fatigue compensation function, specifically the effectiveness threshold D. th The dynamic adjustment no longer relies on the frequency of a single invalid event, but is driven by the following multi-factor logic: when the frequency of actions within three consecutive time windows is detected to be lower than (1-λ) of the initial frequency... freq When the high-frequency jitter energy (8Hz to 12Hz) of the background channel accounts for more than 40% of the total energy, the system determines that it has entered the fatigue compensation critical region. At this time, the system adjusts the frequency according to the step size. The validity threshold is increased step by step until the upper limit is reached. Or, for the recovery of movement frequency, this procedure transforms the adjustment of adaptive thresholds from a vague trend response into a deterministic closed-loop control process with clear triggering conditions, step size control, and boundary constraints by introducing clear physiological characteristic anchors (tremor energy ratio, frequency decay rate) and individualized baseline data.

[0034] To verify the system's reliability under non-ideal environments, this embodiment further clarifies the boundary protection logic for signal anomalies and severe disturbances. In actual home environments, sensors may be subjected to instantaneous strong impacts such as collisions or continuous electromagnetic interference, causing non-physiological abrupt changes in the data stream. To address such challenges, the system incorporates an anomaly-breaking mechanism based on signal morphology. At the front end of the stability measurement calculation unit, a signal preprocessing and feature screening stage is set up to monitor the first-order difference (rate of change of acceleration, i.e., jerk) and amplitude envelope of the input signal in real time. When the jerk in any axis exceeds the physiological limit of human motion, such as 150 m / s², the system will take action. 3If a signal undergoes a polarity reversal and an amplitude change exceeding 2g (gravitational acceleration) within 50ms, the system determines that the data segment is a non-biological motion artifact, such as a sensor colliding with a hard object. In this case, the logic arbitration controller triggers a signal abnormality circuit breaker, directly discarding the data within the current time window and suspending subsequent compensatory evaluation logic until the signal characteristics return to the physiologically feasible domain for more than 500ms. In addition, for persistent electromagnetic interference, the system uses the IMU's built-in magnetometer to monitor the modulus change of the environmental magnetic field vector. If the variance of the magnetic field modulus exceeds 20% of the geomagnetic field strength within a 1s window, it indicates the presence of strong magnetic interference, such as when near high-power electrical appliances. The system will automatically downgrade the calculation algorithm, temporarily disable the absolute heading correction function that relies on the magnetometer, and only perform short-term attitude estimation based on gyroscope integration. It will also send a prompt to the user to move away from the interference source. This layered defense system ensures that the system will not output incorrect evaluation results when facing uncontrollable environmental disturbances, but will ensure the reliability of the final output data through downgraded operation or active circuit breaker.

[0035] Example 5: This example addresses the adaptive model initialization dependency issue in the aforementioned technical solutions by supplementing a standardized offline calibration and data filling procedure. This provides an objectively reproducible initial parameter set for core models such as the fatigue decay function, ensuring the system's cold-start performance on different individuals. To eliminate the subjectivity of initial parameter settings for the fatigue model, the system incorporates an offline calibration program. This program guides the user through a standardized exhaustion test to obtain an individualized fatigue characteristic baseline. Specifically, the program requires the user to continuously perform a target action at 50% of maximum voluntary contraction force (MVC) until the action amplitude naturally decays to 70% of the initial amplitude. During this process, the system simultaneously records the decay curve of the action frequency and the evolution trajectory of background channel tremor energy. By performing exponential fitting on the collected time-series data, the system extracts the time constant characterizing individual endurance characteristics. With the tremor energy growth rate γ.

[0036] To address calculation errors caused by slight shifts in sensor placement or differences in user body shape, the system enforces a pre-calibration procedure before each evaluation cycle. This requires the user to remain still for 3 seconds in a natural upright or seated posture. During this period, the system collects accelerometer and magnetometer data from each channel, calculates the projection components of the gravity vector in the sensor coordinate system, and constructs the initial attitude rotation matrix R based on this. init Meanwhile, the system calculates the magnetic field interference intensity index by comparing the currently measured geomagnetic vector modulus with the preset local geomagnetic reference value. If the index exceeds the safety threshold, such as 10%, the system will automatically prompt the user to calibrate the magnetometer or adjust the environment to ensure that the initial boundary conditions of the inertial calculation meet the accuracy requirements.

[0037] Example 6: In the mass production and deployment scenario of rehabilitation monitoring equipment, it is necessary to ensure that each device has uniform benchmark accuracy and self-adaptive capability before leaving the factory. To eliminate the impact of sensor manufacturing tolerances and individual wearing differences, this example constructs a standardized system deployment pre-deployment calibration and model building procedure. It performs zero-bias calibration of the inertial measurement unit. On a constant temperature static platform, the system continuously collects static output data of the accelerometer and gyroscope in three orthogonal axes. The sampling time is set to 60 seconds. By calculating the arithmetic mean of the data of each axis, the zero-bias accelerometer b is determined. a With angular velocity zero bias b g These parameters are then stored in the sensor's non-volatile memory as a correction reference for subsequent attitude calculations.

[0038] The system enters the model building phase for specific users. After the user wears the device for the first time, the system guides them to maintain an upright and still posture. This captures the gravitational acceleration vector projection at the current wearing position, and the system uses this projection vector to calculate the initial rotation matrix R of the sensor coordinate system relative to the anatomical coordinate system. init This completes the alignment of the spatial reference. Finally, the system performs personalized calibration of the dynamic threshold parameters, guiding the user to repeat the target action 10 times with standard intensity. The system records the discrete sequence of the background channel data in real time and calculates the mean μ of the sequence. D with standard deviation And set the initial validity threshold to This setting logic is based on the normal distribution. The principle is to ensure that the system covers 99.7% of normal physiological fluctuations in an uncompensated state.

[0039] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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.

Claims

1. A digital rehabilitation guidance system, characterized in that, include: The channel mapping and decoupling module is used to receive the raw time-series signals collected by the inertial measurement unit and, based on a preset human kinematic topology model, decompose the raw signals into a first data stream representing the motion trajectory of the target joint and a second data stream representing the attitude of the associated reference part. The stability measurement calculation unit is used to perform time-domain energy analysis on the second data stream and calculate the state dispersion index that characterizes the intensity of unexpected fluctuations in the associated reference part within a unit time window. A logic arbitration controller is configured with compliance gating logic. The logic arbitration controller is used to monitor the state dispersion index in real time, and when the state dispersion index is less than a preset validity threshold, it generates an activation signal to trigger the amplitude evaluation logic for the first data stream, and when the state dispersion index is greater than or equal to the validity threshold, it generates a blocking signal to block the output of the amplitude evaluation logic and mark the current event as invalid. The threshold adaptive drift unit is used to count the frequency of invalid events and automatically increase the validity threshold based on the preset fatigue decay function in response to the rising trend of the occurrence frequency.

2. The digital rehabilitation guidance system according to claim 1, characterized in that, The system also includes a dynamic validity verification unit, which is used to construct a velocity-position phase trajectory reflecting the motion state of the first data stream, and calculate the convergence slope of the angular velocity parameter within a preset control time window before the displacement amplitude represented by the first data stream reaches its peak value; and generate an inertial fuse signal in response to the angular velocity parameter not showing a trend of convergence to zero or the convergence slope exceeding the preset controlled limit value within the preset control time window. The logic arbitration controller is also used to preferentially perform an invalidation flag operation in response to an inertial fuse failure signal.

3. The digital rehabilitation guidance system according to claim 1, characterized in that, The stability measurement calculation unit also includes a frequency domain arbitration logic module, which is used to: perform spectral decomposition on the second data stream to separate the low-frequency drift component located in the first frequency range and the high-frequency jitter component located in the second frequency range, wherein the first frequency range is lower than the second frequency range; Calculate the first energy density of the low-frequency drift component and the second energy density of the high-frequency flutter component; The logic arbitration controller is also used to perform weighted decision-making: when the state dispersion index is greater than the validity threshold and the proportion of the second energy density in the total energy exceeds the preset exemption ratio, the signal is blocked and the amplitude evaluation logic is kept active.

4. The digital rehabilitation guidance system according to claim 1, characterized in that, The system also includes a collaborative analysis logic module, which is used to: extract the first motion envelope features of the first data stream and the second motion envelope features of the second data stream, and calculate the cross-correlation coefficient between the first motion envelope features and the second motion envelope features within a single action window; The logic arbitration controller is also used to generate a collaborative compensation blocking signal in response to the cross-correlation coefficient exceeding a preset collaborative limit value, and to shield the output of the amplitude evaluation logic in response to the collaborative compensation blocking signal.

5. The digital rehabilitation guidance system according to claim 1, characterized in that, The system also includes a baseline verification logic module, which is used to: extract the static gravity projection vector of the second data stream during the static gap before the motion event represented by the first data stream occurs, and calculate the spatial deviation angle between the static gravity projection vector and the preset standard anatomical zero vector; The logic arbitration controller is also used to unlock the monitoring of subsequent motion events only when the spatial deviation angle is within a preset allowable conical range.

6. The digital rehabilitation guidance system according to claim 1, characterized in that, The threshold adaptive drift unit performs automatic adjustment based on the following logic: a fatigue tolerance model that monotonically decreases over time is established, which associates the validity threshold with the cumulative duration of the current training cycle and the historical average compensation rate; under the premise that continuous invalid event markers are detected and the amplitude characteristics of the first data stream remain stable, the validity threshold is gradually increased by a preset step size until the frequency of invalid events decreases to a preset maintenance level.

7. The digital rehabilitation guidance system according to claim 1, characterized in that, The logic arbitration controller is also equipped with data integrity pre-check logic, which is used to: detect the timestamp synchronization of the first data stream and the second data stream before monitoring the execution state dispersion index; if the time deviation between the two exceeds the preset synchronization tolerance, generate a resynchronization instruction to calibrate the data stream, and suspend the execution of the compliance gating logic before the calibration is completed.

8. The digital rehabilitation guidance system according to claim 1, characterized in that, The channel mapping and decoupling module is also used to: dynamically select the source channel of the second data stream according to the preset rehabilitation action definition template; for upper limb rehabilitation actions, lock the data channel representing the trunk tilt angle as the source of the second data stream; For lower limb rehabilitation exercises, the data channel representing the pelvic rotation angle is locked as the source of the second data stream.

9. A digital rehabilitation guidance system according to claim 3, characterized in that, The stability metric calculation unit uses the following weighted calculation logic when calculating the state dispersion index: Where D is the state dispersion index. This represents the variance of the low-frequency drift component. Let be the variance of the high-frequency tremor component, and α and β be the preset first weighting coefficient and second weighting coefficient, respectively, and satisfy α>β.

10. A digital rehabilitation guidance system according to claim 1, characterized in that, The amplitude evaluation logic is specifically used to: calculate the geometric similarity or Euclidean distance of the first data stream relative to the preset rehabilitation target trajectory within the effective time window of receiving the activation signal; when the activation signal continues to exist and the geometric similarity exceeds the preset pass score, generate and render a feedback signal representing that the action is qualified on the user terminal; If an obstruction signal is received during the action, the current similarity calculation process will be terminated, and an error correction instruction prompting the user to correct the posture will be output instead of an action score.

Images