Edge adaptive wearable gait freezing monitoring and closed-loop intervention system

Abstract

An edge adaptive wearable gait freezing monitoring and closed-loop intervention system, comprising a gait freezing monitoring module, a gait freezing edge detection module and a vibration intervention module; the gait freezing monitoring module comprises a multi-node inertial sensing unit and a multi-node sensing data fusion unit, which collect and fuse lower limb gait data; the gait freezing edge detection module comprises a motion state edge quantization unit, a personalized gait baseline reconstruction unit and a gait freezing multi-level detection model, which are deployed on an edge microprocessor, and the edge microprocessor quantizes the fused gait data to generate a gait feature vector; the vibration intervention module comprises an adaptive intervention parameter mapping model and a vibration application unit, which receive freezing information, judge the freezing state in stages, gradually transition from the initial personalized static intervention parameters to dynamic intervention parameters, and perform smoothing processing on the intervention parameters; the present application realizes smooth, adaptive closed-loop physical intervention on the gait freezing state of the patient.

Description

Technical Field

[0001] This invention relates to the field of wearable health monitoring device technology, specifically to an edge-adaptive wearable gait freeze monitoring and closed-loop intervention system. Background Technology

[0002] With the increasing trend of population aging, the demand for health care and rehabilitation intervention for patients with neurodegenerative diseases is becoming increasingly prominent. Parkinson's disease patients are highly susceptible to severe gait freezing during daily walking, manifested as difficulty initiating movement, sudden cessation of movement, shuffling steps, or extremely asymmetrical gait, greatly increasing the risk of falls and disability. Therefore, continuous monitoring of patients' daily gait and providing immediate physical intervention at the threshold of freezing events have significant clinical and engineering application value in improving patients' walking rhythm and preventing accidental risks.

[0003] However, while existing gait freeze monitoring and assisted intervention solutions can collect gait data through wearable sensors and output assessment results, prompt signals, or assisted intervention control signals after identifying abnormal gait or gait freeze events, for example, patent application CN122117324A discloses a Parkinson's rehabilitation data processing method, patent application CN107361773B discloses a device for detecting and alleviating abnormal gait in Parkinson's disease, and patent application CN112996464B discloses a prompting device and method for handling walking obstacles, these solutions, although involving gait freeze identification, abnormal gait relief, or prompting intervention, still have the following shortcomings in applications aimed at daily continuous wear and real-time closed-loop intervention.

[0004] First, the personalized baseline maintenance mechanism is insufficient. Taking the patent application with publication number CN122117324A as an example, this solution can determine motion state information, candidate gait freeze segments, and target gait freeze events based on sensor data, and generate rehabilitation assessment data or auxiliary intervention control signals. However, this solution mainly focuses on the identification and assessment of gait freeze events, and does not clearly propose a closed-loop maintenance mechanism that continuously updates the wearer's personalized gait baseline from normal walking segments during daily continuous wear, or pauses baseline updates during freeze activation and resumes baseline updates after freeze is lifted.

[0005] Secondly, there is insufficient closed-loop linkage between detection and intervention. Taking the patent application with publication number CN107361773B as an example, although the scheme involves the detection and mitigation of Parkinson's abnormal gait, its focus is on the detection of abnormal gait state and corresponding mitigation measures. It does not highlight the simultaneous completion of gait feature quantification, personalized baseline update control, freeze discrimination and intervention parameter generation on the edge microprocessor side, nor does it form a closed-loop operation mechanism driven by freeze activation signal and freeze release signal.

[0006] Third, the intervention parameters lack personalized smooth adjustment. Taking the patent application with publication number CN112996464B as an example, the solution can provide prompts or assistance for gait disorders, but it usually focuses on the output of prompt signals and does not fully combine the wearer's personalized gait baseline, frozen subtype and comprehensive deviation value to generate continuously changing vibration control parameters, making it difficult to achieve a smooth transition from personalized static intervention, nonlinear dynamic enhancement to intervention decay stage.

[0007] In summary, there is an urgent need in the field for a gait freeze detection and closed-loop intervention system that can dynamically calibrate personalized baselines, quantize in real time at the edge, and provide adaptive smooth mapping to overcome the shortcomings of existing technologies. Summary of the Invention

[0008] To overcome the shortcomings of the prior art, the present invention aims to provide an edge-adaptive wearable gait freeze monitoring and closed-loop intervention system. Its gait freeze monitoring module achieves real-time fusion acquisition of lower limb motion data through a dual-node inertial sensing unit, and performs edge quantization of the motion state at the edge, extracting gait feature vectors, thus solving the high latency problem caused by traditional cloud computing. The gait freeze edge detection module performs in-depth analysis and processing of the quantified features, dynamically reconstructing a personalized gait baseline, and accurately comparing and classifying the freeze state through a multi-level detection model, thereby quantifying the specific freeze subtype and its comprehensive deviation value, overcoming the limitations of poor adaptability of traditional fixed thresholds, and achieving high-precision identification of pathological states. The vibration intervention module extracts features from the comprehensive deviation value and freeze subtype, and based on a stepped state machine and multi-dimensional time window control, integrates personalized static and nonlinear dynamic intervention mapping functions to output multi-dimensional collaborative smooth intervention parameters, ultimately driving a vibration motor, achieving smooth, adaptive closed-loop physical intervention for the patient's gait freeze state.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: An edge-adaptive wearable gait freeze monitoring and closed-loop intervention system, characterized in that it includes a gait freeze monitoring module, a gait freeze edge detection module, and a vibration intervention module; The gait freeze monitoring module includes a multi-node inertial sensing unit and a multi-node sensing data fusion unit; the multi-node inertial sensing unit includes at least two inertial sensing nodes configured on the user's lower limbs, each inertial sensing node having a built-in microcontroller, a three-axis accelerometer and a three-axis gyroscope; the output of the multi-node inertial sensing unit is connected to the input of the multi-node sensing data fusion unit. The gait freeze edge detection module includes an edge microprocessor and a motion state edge quantization unit, a personalized gait baseline reconstruction unit, and a gait freeze multi-level detection model deployed on the edge microprocessor; the output of the multi-node sensor data fusion unit is connected to the input of the motion state edge quantization unit, the output of the motion state edge quantization unit is connected to the input of the personalized gait baseline reconstruction unit and the gait freeze multi-level detection model, and the output of the personalized gait baseline reconstruction unit is connected to the gait freeze multi-level detection model and the vibration intervention module. The vibration intervention module includes an adaptive intervention parameter mapping model and a vibration application unit; the output of the gait freeze multi-level detection model is connected to the input of the adaptive intervention parameter mapping model, and the output of the adaptive intervention parameter mapping model is connected to the vibration application unit; the vibration application unit includes at least one vibration motor configured on the user's lower limb. The gait freeze multi-level detection model is used to output freeze activation signals or freeze deactivation signals to the personalized gait baseline reconstruction unit and the adaptive intervention parameter mapping model, and to output freeze subtype and comprehensive deviation value to the adaptive intervention parameter mapping model. The adaptive intervention parameter mapping model generates vibration control parameters based on the freeze subtype, comprehensive deviation value and personalized reference parameters output by the personalized gait baseline reconstruction unit, so as to drive the vibration motor to apply tactile stimulation to the user, thereby forming a closed loop connection between freeze monitoring, personalized baseline maintenance and vibration intervention.

[0010] At least two inertial sensor nodes are worn on the user's left and right lower limbs respectively, and the timestamps of the bilateral motion data are aligned through a time synchronization mechanism. Each inertial sensor node synchronously collects the acceleration and angular velocity signals of its own lower limb at a preset sampling rate, and sends the collected motion data to the multi-node sensor data fusion unit.

[0011] The multi-node sensor data fusion unit is used to perform noise filtering and attitude calculation on the time-aligned motion data to obtain the attitude information of both lower limbs, and extract gait spatiotemporal feature parameters including gait frequency, swing phase duration and gait symmetry.

[0012] The motion state edge quantization unit is used to perform time-domain statistical quantization of gait spatiotemporal feature parameters based on a sliding window within the edge microprocessor, generate standardized gait feature vectors, and output the gait feature vectors in parallel to the personalized gait baseline reconstruction unit and the gait freeze multi-level detection model.

[0013] The personalized gait baseline reconstruction unit is used to select normal walking segments from gait feature vectors according to preset time-domain rhythm criteria during the system cold start phase and daily operation after baseline establishment, and to perform incremental moving average calculation on the normal walking segments in order to construct and dynamically update personalized baseline parameters; when a freeze activation signal is received, the update of personalized baseline parameters is paused, and the update is resumed after a freeze release signal is received.

[0014] The described gait freeze multi-level detection model performs two levels of detection: First, before personalized baseline parameters are established, a built-in universal conservative threshold is used; after personalized baseline parameters are established, a relatively adaptive threshold based on dynamic calibration of personalized baseline parameters is used to initially screen gait feature vectors and mark suspected freeze windows. Second, multi-dimensional features of suspected freeze windows are extracted, and a classifier is used to determine whether gait freeze has occurred and identify the freeze subtype. When gait freeze is confirmed, dynamic deviation quantization of freeze is performed, and the freeze subtype, comprehensive deviation value, and freeze activation signal are output. If, under the condition that a freeze activation signal has been output but not yet deactivated, a freeze deactivation signal is output if multiple consecutive sliding windows fail to identify suspected freeze windows or the classification result is continuously non-frozen.

[0015] The adaptive intervention parameter mapping model includes a stepped state machine and a multi-dimensional time window control module, and is configured to: in the initial stage of freezing, input personalized benchmark parameters and freezing subtype into a personalized static intervention mapping function to generate personalized static intervention parameters; if no freeze release signal is received within the first preset time window, switch to nonlinear dynamic intervention, so that the intervention intensity is dynamically adjusted with the freezing duration and comprehensive deviation value; input personalized static intervention parameters and dynamic adjustment amount into a multi-dimensional collaborative smoothing mapping function to generate continuously gradually changing vibration control parameters.

[0016] The adaptive intervention parameter mapping model simultaneously determines whether a freeze release signal is received and whether the intervention duration has reached the maximum intervention time within each preset cycle during the nonlinear dynamic intervention. If a freeze release signal is received or the intervention duration reaches the maximum intervention time, the adjustment of the intervention intensity is stopped, and the vibration control parameters are smoothly returned to zero along the decay curve. When the intervention is stopped due to reaching the maximum intervention time, the intervention cooling period is entered, and the response to new freeze activation signals is suspended during the intervention cooling period.

[0017] The vibration motor is fixedly installed on the stimulated part of the user's lower limb and performs mechanical vibration according to the instantaneous frequency, duty cycle and amplitude parameters in the vibration control parameters to output physical tactile stimulation with continuously changing amplitude, frequency and beat.

[0018] Compared with the prior art, the present invention has the following beneficial technical effects: This invention uses a personalized gait baseline reconstruction unit to automatically filter normal walking segments during the cold start phase using time-domain rhythm criteria, and constructs personalized baseline parameters for the wearer using incremental moving averages. This allows the discrimination criteria for frozen detection to be dynamically adjusted according to the wearer's actual gait characteristics, avoiding the problem of poor adaptability of fixed thresholds to patients with different disease stages, and reducing the misjudgment rate of daily turning, active stopping and other actions.

[0019] This invention deploys the motion state edge quantization, baseline reconstruction, and freeze detection processes in an edge microprocessor. Motion state edge quantization relies solely on time-domain statistical operations to generate gait feature vectors, eliminating the need for frequency-domain transformation or uploading to the cloud. This eliminates the delays caused by remote data transmission and centralized computation, enabling the system to complete feature extraction and early warning at the edge of a freeze, thus improving the real-time performance of daily continuous monitoring.

[0020] The gait freeze multi-level detection model of the present invention adopts a two-level identification structure of "dynamic threshold initial screening + multi-dimensional feature classification". The dynamic threshold operates based on a built-in general conservative threshold during the cold start stage. After the personalized baseline is established, it automatically switches to a relatively adaptive threshold based on the wearer's own baseline. In conjunction with the classifier to identify the freeze subtype, it reduces the situation of missed detection of severe patients or misdetection of mild patients due to a single fixed threshold, and reduces the false alarm rate while ensuring sensitivity.

[0021] The vibration intervention module of this invention adopts a phased adaptive intervention strategy. In the initial stage of freezing, personalized static intervention parameters are applied. If the freezing is not lifted, nonlinear dynamic intervention is switched to the next stage. Vibration control parameters are generated through a multidimensional collaborative smoothing mapping function, so that the amplitude, frequency and beat of vibration stimulation can be smoothly transitioned, avoiding startle reflex and balance disorder caused by on-off rigid stimulation, and delaying the development of intervention tolerance.

[0022] This invention feeds a freeze activation signal to the personalized gait baseline reconstruction unit through a multi-level gait freeze detection model, so that the baseline updates are paused during the freeze and resumed after the freeze is lifted, forming a closed-loop feedback between the detection and the baseline. This prevents abnormal gait data in the frozen state from contaminating the normal baseline and ensures the accuracy of personalized reference parameters during long-term wear. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the gait freezing multi-level detection model and the adaptive intervention parameter mapping model in an embodiment of the present invention.

[0025] Figure 3 This is a flowchart of the multi-level gait freeze detection process according to an embodiment of the present invention.

[0026] Figure 4 This is a flowchart of the gait freeze closed-loop adaptive intervention process according to an embodiment of the present invention.

[0027] Figure 5 This is a schematic diagram of the closed-loop operation experiment results of an embodiment of the present invention. Detailed Implementation

[0028] The present invention will be further described below with reference to the embodiments and accompanying drawings.

[0029] Reference Figure 1 A wearable edge-adaptive gait freeze monitoring and closed-loop intervention system is disclosed. It includes a gait freeze monitoring module, a gait freeze edge detection module, and a vibration intervention module. The gait freeze monitoring module comprises a multi-node inertial sensing unit and a multi-node sensing data fusion unit. The multi-node inertial sensing unit includes at least two inertial sensing nodes configured on the user's lower limbs, each inertial sensing node housing a microcontroller, a three-axis accelerometer, and a three-axis gyroscope. The output of the multi-node inertial sensing unit is connected to the input of the multi-node sensing data fusion unit. The gait freeze edge detection module includes an edge microprocessor and a motion state edge quantization unit, a personalized gait baseline reconstruction unit, and a gait freeze multi-level detection model deployed on the edge microprocessor. The output of the multi-node sensor data fusion unit is connected to the input of the motion state edge quantization unit. The output of the motion state edge quantization unit is connected to the input of the personalized gait baseline reconstruction unit and the gait freeze multi-level detection model. The output of the personalized gait baseline reconstruction unit is connected to the gait freeze multi-level detection model and the vibration intervention module. The vibration intervention module includes an adaptive intervention parameter mapping model and a vibration application unit. The output of the gait freeze multi-level detection model is connected to the input of the adaptive intervention parameter mapping model, and the output of the adaptive intervention parameter mapping model is connected to the vibration application unit. The vibration application unit includes at least one vibration motor configured on the user's lower limb. The gait freeze multi-level detection model is used to output freeze activation signals or freeze deactivation signals to the personalized gait baseline reconstruction unit and the adaptive intervention parameter mapping model, and to output freeze subtype and comprehensive deviation value to the adaptive intervention parameter mapping model. The adaptive intervention parameter mapping model generates vibration control parameters based on the freeze subtype, comprehensive deviation value, and personalized baseline parameters output by the personalized gait baseline reconstruction unit to drive the vibration motor to apply tactile stimulation to the user, thereby forming a closed-loop connection between freeze monitoring, personalized baseline maintenance, and vibration intervention.

[0030] At least two inertial sensor nodes are worn on the user's left and right lower limbs respectively, and the timestamps of the motion data from both sides are aligned through a time synchronization mechanism. Each inertial sensor node synchronously collects the acceleration and angular velocity signals of its own lower limb at a preset sampling rate, and sends the collected motion data to a multi-node sensor data fusion unit. In this embodiment, the two inertial sensor nodes worn on the left and right lower limbs respectively employ a wireless time synchronization mechanism.

[0031] The multi-node sensor data fusion unit is used to perform noise filtering and attitude calculation on the time-aligned motion data to obtain the attitude information of both lower limbs, and extract gait spatiotemporal feature parameters including gait frequency, swing phase duration and gait symmetry.

[0032] Reference Figure 2 The motion state edge quantization unit is used to perform time-domain statistical quantization of gait spatiotemporal feature parameters based on a sliding window within the edge microprocessor, generate standardized gait feature vectors, and output the gait feature vectors in parallel to the personalized gait baseline reconstruction unit and the gait freeze multi-level detection model.

[0033] Reference Figure 3 The personalized gait baseline reconstruction unit is used to select normal walking segments from gait feature vectors according to preset time-domain rhythm criteria during the system cold start phase and daily operation after baseline establishment, and to perform incremental moving average calculation on the normal walking segments in order to construct and dynamically update personalized baseline parameters; when a freeze activation signal is received, the update of personalized baseline parameters is paused, and the update is resumed after a freeze release signal is received.

[0034] Reference Figure 2 and Figure 3 The gait freeze multi-level detection model performs two levels of detection: First, before personalized baseline parameters are established, a built-in universal conservative threshold is used; after personalized baseline parameters are established, a relatively adaptive threshold dynamically calibrated based on personalized baseline parameters is used to initially screen gait feature vectors and mark suspected freeze windows. Second, multi-dimensional features of suspected freeze windows are extracted, and a classifier is used to determine whether gait freeze has occurred and identify the freeze subtype. When gait freeze is confirmed, dynamic deviation quantization of freeze is performed, and the freeze subtype, comprehensive deviation value, and freeze activation signal are output. If, under the condition that a freeze activation signal has been output but not yet deactivated, a freeze deactivation signal is output if multiple consecutive sliding windows fail to identify suspected freeze windows or the classification result is continuously non-frozen. In this embodiment, multiple consecutive sliding windows refer to N consecutive sliding windows.

[0035] Reference Figure 2 and Figure 4The adaptive intervention parameter mapping model includes a stepped state machine and a multi-dimensional time window control module, and is configured as follows: In the initial freezing stage, personalized baseline parameters and freezing subtype are input into a personalized static intervention mapping function to generate personalized static intervention parameters; if no freeze release signal is received within the first preset time window, nonlinear dynamic intervention is initiated, and the intervention intensity is dynamically adjusted according to the freezing duration and comprehensive deviation value; the personalized static intervention parameters and dynamic adjustment amount are input into a multi-dimensional collaborative smoothing mapping function to generate continuously gradually changing vibration control parameters; during each preset cycle of the nonlinear dynamic intervention, the adaptive intervention parameter mapping model simultaneously determines whether a freeze release signal is received and whether the intervention duration has reached the maximum intervention time; if a freeze release signal is received or the intervention duration has reached the maximum intervention time, the adjustment of the intervention intensity is stopped, and the vibration control parameters are smoothly returned to zero along the decay curve; when intervention is stopped due to reaching the maximum intervention time, the intervention cooling period is entered, and the response to new freeze activation signals is suspended during the intervention cooling period.

[0036] The vibration motor is fixedly installed on the stimulated part of the user's lower limb and performs mechanical vibration according to the instantaneous frequency, duty cycle and amplitude parameters in the vibration control parameters to output physical tactile stimulation with continuously changing amplitude, frequency and beat; the physical tactile stimulation is used to help the user safely get out of the gait freeze state and restore independent walking.

[0037] Reference Figure 5 The horizontal axis represents the normalization recording process, and the area between the two vertical dashed lines is the marked frozen anomaly interval. After entering this interval, the standardized angular velocity changes from a relatively regular periodic fluctuation before freezing to a fluctuation state with smaller amplitude and disordered rhythm. The overall freezing suspicion increases accordingly and is generally higher than the personalized relative threshold within the anomaly interval.

[0038] The system first outputs an initial screening signal based on the comparison between the overall freeze suspicion level and the personalized relative threshold. Because the overall freeze suspicion level exhibits local fluctuations, the initial screening signal is allowed to switch briefly within the abnormal range. The system further generates a confirmation signal through continuous window verification, ensuring the confirmation result remains continuous and stable during the freeze period, preventing interruptions or repeated triggering of the freeze event due to characteristic fluctuations in a single window. After the confirmation signal is output, the system pauses personalized gait baseline updates; after the freeze is lifted, the confirmation signal is revoked, and baseline updates resume, thereby reducing the contamination of normal gait baselines by frozen abnormal data.

[0039] After confirming a freeze event, the system continuously generates vibration amplitude, vibration frequency, and beat parameters based on the deviation of the overall freeze suspicion level from the personalized relative threshold. These three intervention parameters rise and fall in tandem with the freeze status, rather than being simply started and stopped using fixed parameters. When the freeze severity increases, the intervention parameters increase accordingly; when the freeze severity decreases, the intervention parameters decrease accordingly. After the freeze is lifted, the intervention parameters continue to decline smoothly and gradually return to zero, thereby reducing parameter abrupt changes during intervention initiation, dynamic adjustment, and termination.

[0040] Therefore, this embodiment achieves closed-loop linkage between freeze monitoring, personalized baseline maintenance, and vibration intervention through personalized relative thresholds, two-level judgment of initial screening and confirmation, suspension of baseline updates during freezing, and smooth mapping of multiple intervention parameters. This improves the stability of freeze event output, the anti-contamination capability of personalized baselines, and the continuity of intervention parameter changes.

[0041] The above description is only one embodiment of the present invention, and not all or the only embodiment. Any equivalent modifications made by those skilled in the art to the technical solution of the present invention by reading the present invention specification are covered by the claims of the present invention.

Claims

1. An edge-adaptive wearable gait freeze monitoring and closed-loop intervention system, characterized in that, It includes a gait freeze monitoring module, a gait freeze edge detection module, and a vibration intervention module; The gait freeze monitoring module includes a multi-node inertial sensing unit and a multi-node sensing data fusion unit; the multi-node inertial sensing unit includes at least two inertial sensing nodes configured on the user's lower limbs, each inertial sensing node having a built-in microcontroller, a three-axis accelerometer and a three-axis gyroscope; the output of the multi-node inertial sensing unit is connected to the input of the multi-node sensing data fusion unit. The gait freeze edge detection module includes an edge microprocessor and a motion state edge quantization unit, a personalized gait baseline reconstruction unit, and a gait freeze multi-level detection model deployed on the edge microprocessor; the output of the multi-node sensor data fusion unit is connected to the input of the motion state edge quantization unit, the output of the motion state edge quantization unit is connected to the input of the personalized gait baseline reconstruction unit and the gait freeze multi-level detection model, and the output of the personalized gait baseline reconstruction unit is connected to the gait freeze multi-level detection model and the vibration intervention module. The vibration intervention module includes an adaptive intervention parameter mapping model and a vibration application unit; the output of the gait freeze multi-level detection model is connected to the input of the adaptive intervention parameter mapping model, and the output of the adaptive intervention parameter mapping model is connected to the vibration application unit; the vibration application unit includes at least one vibration motor configured on the user's lower limb. The gait freeze multi-level detection model is used to output freeze activation signals or freeze deactivation signals to the personalized gait baseline reconstruction unit and the adaptive intervention parameter mapping model, and to output freeze subtype and comprehensive deviation value to the adaptive intervention parameter mapping model. The adaptive intervention parameter mapping model generates vibration control parameters based on the frozen subtype, the comprehensive deviation value, and the personalized baseline parameters output by the personalized gait baseline reconstruction unit. These parameters drive the vibration motor to apply tactile stimulation to the user, thereby forming a closed-loop connection between frozen monitoring, personalized baseline maintenance, and vibration intervention.

2. The system according to claim 1, characterized in that, At least two inertial sensor nodes are worn on the user's left and right lower limbs respectively, and the timestamps of the bilateral motion data are aligned through a time synchronization mechanism. Each inertial sensor node synchronously collects the acceleration and angular velocity signals of its own lower limb at a preset sampling rate, and sends the collected motion data to the multi-node sensor data fusion unit.

3. The system according to claim 1, characterized in that, The multi-node sensor data fusion unit is used to perform noise filtering and attitude calculation on the time-aligned motion data to obtain the attitude information of both lower limbs, and extract gait spatiotemporal feature parameters including gait frequency, swing phase duration and gait symmetry.

4. The system according to claim 1, characterized in that, The motion state edge quantization unit is used to perform time-domain statistical quantization of gait spatiotemporal feature parameters based on a sliding window within the edge microprocessor, generate standardized gait feature vectors, and output the gait feature vectors in parallel to the personalized gait baseline reconstruction unit and the gait freeze multi-level detection model.

5. The system according to claim 1, characterized in that, The personalized gait baseline reconstruction unit is used to select normal walking segments from gait feature vectors according to preset time-domain rhythm criteria during the system cold start phase and daily operation after baseline establishment, and to perform incremental moving average calculation on the normal walking segments in order to construct and dynamically update personalized baseline parameters; when a freeze activation signal is received, the update of personalized baseline parameters is paused, and the update is resumed after a freeze release signal is received.

6. The system according to claim 1, characterized in that, The described gait freeze multi-level detection model performs two levels of detection: First, before personalized baseline parameters are established, a built-in universal conservative threshold is used; after personalized baseline parameters are established, a relatively adaptive threshold based on dynamic calibration of personalized baseline parameters is used to initially screen gait feature vectors and mark suspected freeze windows. Second, multi-dimensional features of suspected freeze windows are extracted, and a classifier is used to determine whether gait freeze has occurred and identify the freeze subtype. When gait freeze is confirmed, dynamic deviation quantization of freeze is performed, and the freeze subtype, comprehensive deviation value, and freeze activation signal are output. If, under the condition that a freeze activation signal has been output but not yet deactivated, a freeze deactivation signal is output if multiple consecutive sliding windows fail to identify suspected freeze windows or the classification result is continuously non-frozen.

7. The system according to claim 1, characterized in that, The adaptive intervention parameter mapping model includes a ladder state machine and a multi-dimensional time window control module, and is configured to: in the initial stage of freezing, input personalized baseline parameters and freezing subtype into a personalized static intervention mapping function to generate personalized static intervention parameters; If no freeze release signal is received within the first preset time window, nonlinear dynamic intervention is initiated, and the intervention intensity is dynamically adjusted according to the freeze duration and the comprehensive deviation value. The personalized static intervention parameters and dynamic adjustment values ​​are input into the multidimensional collaborative smoothing mapping function to generate continuously changing vibration control parameters.

8. The system according to claim 7, characterized in that, The adaptive intervention parameter mapping model simultaneously determines whether a freeze release signal is received and whether the intervention duration has reached the maximum intervention time within each preset cycle during the nonlinear dynamic intervention. If a freeze release signal is received or the intervention duration reaches the maximum intervention time, the adjustment of the intervention intensity is stopped, and the vibration control parameters are smoothly returned to zero along the decay curve. When the intervention is stopped due to reaching the maximum intervention time, the intervention cooling period is entered, and the response to new freeze activation signals is suspended during the intervention cooling period.

9. The system according to claim 1, characterized in that, The vibration motor is fixedly installed on the stimulated part of the user's lower limb and performs mechanical vibration according to the instantaneous frequency, duty cycle and amplitude parameters in the vibration control parameters to output physical tactile stimulation with continuously changing amplitude, frequency and beat.

Images