Gait monitoring and muscle force monitoring fusion algorithm and rehabilitation walking aid using same

Abstract

The invention relates to a gait monitoring algorithm which is used for a traditional four-foot walking aid, a six-axis gyroscope is installed in the middle of four feet of the walking aid, pressure sensor data is installed at the bottom of a handle and the bottoms of the four feet, and the gait monitoring algorithm comprises a sensor data acquisition module used for acquiring original sensor data related to gaits of a user, acquiring three-axis acceleration information; acquiring triaxial angular velocity information according to gyroscope sensor data; acquiring handle pressure and ground contact pressure information according to data of a pressure sensor; the gait feature extraction system is used for extracting gait features from original sensor data and comprises a gait feature extraction module used for extracting gait feature values from the original sensor data, the gait feature values comprise stride frequency, and the number of steps per second is calculated by analyzing periodic fluctuation of pressure or acceleration; and calculating the distance between the two steps through acceleration integration or the ground contact interval between the left foot and the right foot.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of rehabilitation equipment, and particularly relates to a gait monitoring and muscle strength monitoring fusion algorithm and a rehabilitation walking aid using the same. BACKGROUND

[0002] With the aggravation of population aging and the increasing incidence of nervous system diseases such as stroke, early rehabilitation is crucial to avoid and reduce disability. The most urgent rehabilitation for hemiplegia and limb dysfunction is gait rehabilitation training. Traditional rehabilitation training is mostly carried out in professional medical institutions, and patients need to complete walking training under the guidance of rehabilitation therapists to improve gait coordination, improve stride and step frequency, and correct body posture. However, due to factors such as shortage of rehabilitation resources, high cost, and strong long-term dependence of patients, many patients are difficult to maintain continuous high-quality rehabilitation training, which affects the rehabilitation effect, and even may cause secondary injury due to incorrect gait.

[0003] Old people, stroke hemiplegia and limb dysfunction patients are high-risk groups of sarcopenia, and need to be detected as soon as possible, to predict the risk of falling and intervene in time to reduce the harm to patients and promote rehabilitation.

[0004] In order to improve the popularity and convenience of rehabilitation training, portable walking aids have gradually appeared on the market, mainly used to improve the stability and safety of patients when walking. However, the existing walking aid products are generally single-function, only having basic physical support function, lacking intelligent auxiliary system, and being unable to give real-time correction according to the specific gait state of the patient, continuous normal feedback guidance, especially being unable to help the patient to carry out standardized and self-rehabilitation training in the home environment, missing the best rehabilitation window period, because of inaccurate rehabilitation exercise, appearing of modified walking gait, and even falling, causing fractures, and resistance to rehabilitation, which seriously affects the quality of life of patients and increases the social and family burden.

[0005] In recent years, with the development of sensor technology, artificial intelligence algorithm and embedded voice feedback technology, gait monitoring and analysis have become possible under the premise of controllable hardware cost. Some researches have used accelerometers, gyroscopes and other devices to detect the step frequency, stride and posture characteristics of users, and applied them in wearable devices or insoles. However, these systems are usually positioned for data collection and post-evaluation, or require professional operating environment, and have not formed an integrated, real-time voice correction guidance application solution for rehabilitation walking aids.

[0006] Therefore, it is urgent to provide a rehabilitation walking aid system integrating gait monitoring, behavior judgment, muscle strength judgment and voice guidance, which can identify abnormal gait features in real time during the walking process of a user, and provide clear voice prompts to help the patient correct the wrong pace and posture in time, understand the body activity ability, so as to realize intelligent, adaptive and closed-loop rehabilitation training support in the daily home environment, accelerate rehabilitation, return to society and improve life happiness. SUMMARY

[0007] The purpose of the present application is to provide a gait monitoring and muscle strength monitoring fusion algorithm and a rehabilitation walking aid using the same.

[0008] In order to achieve the above purpose, the technical scheme adopted by the present application is: 1. Accelerometer (IMU) / Gyroscope (Gyro) x 2; Installation position: middle part of the front two side columns (close to knee height), Function: used for monitoring the overall movement direction, travel acceleration and speed change of the patient, reflecting the walking rhythm and posture deviation.

[0009] Data type: three-axis acceleration (Ax, Ay, Az) + three-axis angular velocity (Gx, Gy, Gz); 2. Accelerometer (IMU) / Gyroscope (Gyro) x 2; Installation position: integrated with the same module as the acceleration sensor (combined IMU), also located in the middle part of the rear two side columns; Function: real-time acquisition of the rotation angle and inclination of the patient's trunk and upper body, judgment of whether the body is forward or laterally inclined.

[0010] Data type: three-axis acceleration (Ax, Ay, Az) + three-axis angular velocity (Gx, Gy, Gz), 1, 2 combined as a 6-axis sensor.

[0011] 3. Pressure sensor x 6; Installation position: Two under the handle: detect the symmetrical force condition of the two hands; Four at the bottom of the four columns: judge the ground contact pressure of the left and right front and rear columns, analyze the body center of gravity deviation and standing stability of the patient; Function: Judge whether the force of the two hands is balanced, and assist in analyzing whether hemiplegia exists; Detect whether the gait rhythm and step frequency are consistent; Judge the time distribution of the support period and the weight bearing difference between the left and right feet.

[0012] One of the data training purposes: Step frequency training, the step frequency should be maintained within a reasonable range (such as 0.8~1.5Hz); Step length training, the step length should reach a certain minimum value (such as ≥ 40cm); Body posture training, the upper body posture should be maintained upright, with an offset angle ≤ 10°.

[0013] These raw data are preprocessed by algorithms and converted into the gait features we need.

[0014] II. Core "feature values" extracted from raw sensor data.

[0015] Feature name, meaning and calculation method.

[0016] Step frequency (Hz), the number of steps per second. Calculated from the periodic fluctuations in pressure or acceleration, typical range 0.8~1.5Hz.

[0017] Step length (m), the distance between two steps. Calculated from the displacement obtained by integrating acceleration or the interval between left and right foot contact. The target value is generally >0.4m.

[0018] Body pitch angle, Pitch angle. Integrated from the angular velocity of the gyroscope or directly calculated from the attitude (usually using Gx data).

[0019] Right pressure difference (N), the pressure difference on the handle or column; Gait rhythm variation, the fluctuation of continuous step frequency or step length variation (such as standard deviation), to determine whether it is stable. Used for sliding window anomaly detection.

[0020] Ground contact duration, the time each foot is in contact with the ground, from the pressure sensor rhythm. Abnormal cases show that one side has too short a time. (Note that this sampling value can only be collected by a standard four-wheel walker, as the front-wheel walker cannot collect it. The reason is as follows: the front wheel part does not sense vertical pressure during walking, and the rear foot is generally a non-slip pad, which may only be in contact with the ground for part of the time. Therefore, the left and right foot support information may be masked in the rhythm of such a sliding walker. If you force to sample this kind of walker, you cannot obtain the user's walking pattern.) Now let's explain the above sampling rules (the rationality of extracting the above "feature values", and further discuss the rationality of the above sampling values): Although the sensors are not installed on the feet, the sensors installed on the walker (such as pressure sensors, inertial sensors) can still indirectly infer the ground contact duration of each foot of the user. This relies on the mechanical rhythm characteristics of the "human-walker coupled gait".

[0021] Key principle: "human-walker" coupled gait.

[0022] When walking with a walker, a fixed rhythm pattern of "walker support + alternating feet forward" is usually presented. Since the walker bears part of the body weight load and moves synchronously with the person in the gait, the pressure sensors installed on the walker (such as 4 pressure sensors on the legs) can perceive the up-and-down pressure pattern applied by the person. Thus, the "foot contact duration" of each foot is inferred, and the technical path is as follows: 1. Assumption: walking rhythm is synchronous and alternating: Most walker users adopt a "three-point gait" or "four-point gait", and the pattern is as follows: Walker moves forward → one foot steps out (support + contact) → the other foot follows (support + contact).

[0023] In this pattern: When the body center of gravity shifts to one side, the load (leg pressure) on the walker will change significantly, which can be perceived by the sensors: which foot is supporting, and which foot is just off the ground.

[0024] 2. Pressure / inertia signal changes are mapped to foot rhythm; When the left foot is supporting and the right foot is stepping: Body center of gravity shifts to the left → corresponding pressure sensor on the walker will feel a relatively stable and sustained pressure; When the right foot hits the ground: Redistribution of body center of gravity → the other side sensor also perceives an increase in pressure.

[0025] Therefore, if there are four pressure sensors (one on each leg), the rhythm of the front and back pressure / left and right pressure changes can be used to infer which side the person is contacting and how long the contact lasts.

[0026] Therefore, to summarize, the inference of "left and right foot contact duration imbalance" in the system can be done as follows: Pressure sensors are installed under each leg; By continuously monitoring the rhythm of the person pushing the walker and stepping; Analyze the interval between the pressure rise and fall times; Map to left and right gait rhythm; At this point, the sensor behavior reflects the force feedback of the walker as an "extension of the human body" and forms a coupling with the two-foot rhythm.

[0027] 3. Algorithm analysis method example For each step, collect the following in a continuous time period: Which pressure sensors are rising / dropping continuously; Does the acceleration sensor show that the body inertia is shifting to one side? Obtained through training or rule analysis: Duration of left foot touching the ground; Duration of right foot touching the ground.

[0028] If the time on one side is significantly shorter, it may indicate the presence of: Insufficient muscle strength Unable to provide stable support for extended periods. Abnormal gait such as hemiplegia / habitual weight-avoidance.

[0029] Which foot steps first, and the alternating rhythm of pressure or acceleration from the walker to the left and right. The duration of contact between each foot and the ground varies, with one side of the walker leg bearing more pressure for a longer period. Does the system exhibit hemiplegic behavior, with a significantly shorter ground contact time on one side, or an excessively steep force change curve (instantaneous support)?

[0030] For example: If repeated recordings are continued, the following will be found: The pressure duration when the left foot is supporting the weight is always short, and the inertial data shows rapid switching during this period → It can be determined that the short duration of the left foot's contact with the ground indicates an anomaly.

[0031] The human and the walking aid form a coupled system, with the walking aid synchronously supporting the body's center of gravity and rhythm. The walking aid senses changes in load, reflecting changes in the force exerted on it by the body, indirectly indicating gait rhythm. Pressure sensor signals can identify rhythmic characteristics, determining which foot is supporting the weight and for how long. Superimposing multiple steps can infer abnormal patterns, identifying uneven force application between the left and right feet and inconsistent ground contact times.

[0032] Third, recommended sampling frequency Accelerometer / Gyroscope: 50~100Hz (standard IMU supports this by default); Pressure sensor: 10~30Hz; Recommended overall sampling processing window: 1 second as the basic unit.

[0033] For three common training goals of patients in the rehabilitation phase: cadence, stride length, and body posture control, the following algorithm is used for detection and improvement: 1. Rule-based decision; This is the simplest and most direct algorithm—a threshold range is set in advance, and when the sensor data exceeds the range, the patient's gait is considered abnormal, and a voice feedback prompt is given.

[0034] Implementation logic: Set a "reasonable range" for each training objective; Step frequency should be between 0.8-1.5 Hz (0.8-1.5 steps per second); Step length should be greater than 0.4 meters (avoid dragging feet); Body pitch should be within ±10° (avoid leaning forward / backward); The system collects a frame of sensor data (from IMU, pressure sensors, etc.) and parses out: Current step frequency value; Current step length value; Current posture angle; Left-right handle / ground pressure difference; In turn, determine whether each indicator is out of range: Step length too small → prompt: "Please take larger steps"; Step frequency too fast / slow → prompt: "Please pay attention to the step frequency rhythm"; Body leaning forward → prompt: "Please keep your body upright"; Uneven force → prompt: "Uneven force, please balance the force"; If all indicators are normal, remain silent or positive encouragement.

[0035] The advantages of this algorithm are: Simple and efficient, low deployment cost; Can be run offline, no model training required; Parameters are easy to adjust manually.

[0036] 1.1 Rule judgment: Each row of data is labeled according to the rules, for example: freq<0.8 → slow step frequency → label = 1; stride<0.4 → small step length → label = 2; abs(pitch)>10 → body tilt → label = 3; |pL - pR|>10 → unstable center of gravity → label = 4; The rest → normal gait → label = 0; Explanation of data meaning: Freq, (Frequency) Data meaning: Step frequency (unit: Hz or steps / second); Judgment standard: <0.8 steps / second will be considered slow step frequency; Physiological significance: Reflects walking rhythm, normal adult step frequency is about 1.8-2.5 steps / second, less than 0.8 may indicate difficulty in movement or cautious gait; Stride; Data meaning: Stride length (unit: meters / step); Judgment criteria: A stride of less than 0.4 meters is considered too small; Physiological significance: It reflects the distance covered in a single step. A normal stride length is about 0.6-0.8 meters. A reduced stride length may indicate insufficient muscle strength or limited joint mobility. Pitch; Data meaning: Angle of body tilt forward or backward (unit: degrees); Judgment criteria: An absolute value >10° is considered a body tilt; Physiological significance: Positive values ​​indicate forward tilting, negative values ​​indicate backward tilting, and excessive tilting may indicate balance disorders or abnormal posture; pL / pR, (Pressure Left / Right); Data meaning: Difference in pressure distribution between the left and right feet (unit: percentage or pressure value); Judgment criteria: A difference of more than 10 on the left and right sides is considered to indicate instability in the center of gravity; Physiological significance: It reflects the symmetry of body weight distribution; excessive differences are often seen in unilateral pain or neurological diseases. Label, classification logic:; The system will check in priority order 1 to 4 (marking as long as any condition is met, without marking repeatedly); Only when there are absolutely no abnormalities will it be marked as 0 (normal gait).

[0037] 2. Sliding window + anomaly detection (unsupervised or semi-supervised); Even without labeled data, one can determine whether a certain gait segment is abnormal by using "behavioral continuity + statistical stability".

[0038] Implementation logic: Sensor data is collected in real time and recorded once every 1 second; Construct a sliding time window (e.g., one window every 5 seconds); Calculate statistical characteristics within the window: Step frequency mean and variance; average stride length; The maximum fluctuation in body tilt angle; Stability (rate of change) of left and right pressure; Set tolerance range (dynamic or fixed): If the variance is too large, it indicates that the step frequency is unstable; If the average stride length remains consistently small, it indicates a tendency to slack off. If multiple windows malfunction consecutively, a notification will be displayed. Anomaly detection methods can be: Statistical threshold (e.g. 2 standard deviations) Lightweight unsupervised method (e.g. Isolation Forest / One-Class SVM) System only focus on "is abnormal" rather than "type of abnormality", can be used for early warning or dynamic monitoring.

[0039] Advantages: No need for manual labeling of data Can detect trend-type and fluctuation-type abnormalities Can adapt to individual baseline gait (self-adaptive adjustment)

[0040] 2.1 Use Isolation Forest to detect "is abnormal" Output results: 1: normal sample -1: identified as an abnormal sample 3, simple statistical learning algorithm (KNN, decision tree) Based on historical data, use supervised learning to train the model, so that the model can automatically identify which feature combination corresponds to "normal gait" or various "abnormal gait types".

[0041] Implementation logic: Collect a batch of real patient gait data and manually label the gait state: Normal / small stride / unstable stride frequency / body forward inclination / unstable center of gravity, etc. Extract features for each frame of data: Stride frequency (Hz) Stride length (m) Body inclination (pitch / roll) Left and right pressure distribution (N) Gait rhythm (two-foot landing interval, by checking the periodicity of left and right hand pressure / center of gravity movement to determine the left and right foot landing time points) Send features to classification model training: Such as KNN, decision tree The model learns to "judge which type of error from the features" Real-time monitoring stage: System collects current data → feature extraction → input model → output gait category → broadcast voice prompt Model continuous optimization: As the amount of data collected increases, it can be continuously trained to adapt to more diverse patient behaviors.

[0042] Advantages: Can identify errors caused by complex feature combinations; Supports fine classification (multiple gait errors); Adaptable to individual differences (reflected in training data).

[0043] Further, the above algorithm is integrated with the stand-walk test (TUG).

[0044] TUG process test: the patient stands up from a sitting position, walks 3 meters, turns back, and sits back in the chair; the entire process time is measured.

[0045] Purpose: to judge the patient's motor coordination, balance, and functional independence, widely used in rehabilitation and geriatric assessment.

[0046] 1, preset rule judgment algorithm + TUG: Application method: Set the time / height / angle threshold of the key nodes of the TUG test: Rising time <3s; Acceleration Z-axis peak value > certain value (indicating the team stands up); Walking step frequency is between 0.8-1.2 Hz; Turnover time <2s; Total time <13.5s (or patient individual baseline value); Effect: Real-time monitoring of whether the standing is too slow, the walking is too slow, and the turning tension; Voice prompts such as: "Please stand up more decisively", "Please speed up the button", "Pay attention to keep the center of gravity"; Can be set to restore the stage classification target (such as stage one target: TUG time <20s, stage two <16s).

[0047] 2, KNN / decision tree classification model + TUG Application method: Use multiple patient TUG test data (sensor training sequence features + TUG completion time) to classify the model; Features include: standing acceleration, step frequency, walking speed, turning angular speed, pressure center of gravity offset, etc. Training goal: divide patients into A (normal), B (mild), C (severe disability), etc.

[0048] Effect: After the TUG test is completed, the current patient gait state is automatically classified; Voice prompt: "The current stability is good, please keep training"; The subsequent training intensity and feedback frequency can be automatically adjusted accordingly.

[0049] 3, sliding window anomaly detection + TUG Application mode: The standing, walking, deployment and other stages are divided into sliding windows; Monitor whether there are sudden outliers in each stage, such as sudden increase in center of gravity forward inclination at the beginning; Use the mean, standard deviation, and change rate within the window to detect sudden changes; Effect: Accurately locate the stage where the patient has an abnormality: Such as "too fast to stand up and lose balance"; "Twist process deviates to the right"; Voice prompt: "Please stand up smoothly", "Slow down and stand up when turning"; Form a phased atlas of patient abnormal risks. Algorithm type Purpose in TUG test Input features Output results Threshold determination Fast solution of abnormality Step frequency, time, acceleration, angle Real-time voice feedback KNN / decision tree Staged or overall evaluation of rehabilitation level Multi-dimensional sensor statistical features Category (normal / mild / severe disability) Sliding window detection Fine detection of abnormal behavior occurrence point Sensor sequence features within a continuous time window Abnormal point prompt and suggestion Further, add a gait feature synchronous extraction module to solve the problem of lack of specific stage switching recognition algorithm and unclear automatic recognition method of each stage boundary point.

[0050] Traditional decomposition and confirmation of each stage of continuous action is recognized by optical sensing or image algorithm (photographing or photography), which can easily identify the start-end / starting point time of each of the six stages in TUG test. However, if the data of the walker sensor is used for judgment, a set of algorithms for dividing the start-end / starting point time of each of the six stages in TUG test according to the sensor data should be proposed based on the physiological characteristics of the user, who is often a post-stroke or weak elderly person walking.

[0051] 1. Gait feature synchronous extraction module algorithm principle The gait synchronous extraction module is the core of the TUG test data acquisition system, and its main function is to realize the time-synchronous acquisition of multiple sensors and the real-time extraction of feature parameters in the six continuous stages of TUG test. This module adopts a time-based synchronization mechanism to ensure that the data from multiple sensors such as 6-axis gyroscope and pressure sensor can be accurately corresponded in time, thereby providing a reliable data basis for subsequent fusion analysis.

[0052] 1.1 Stage boundary recognition algorithm The first technical problem to be solved is how to automatically identify the boundary points of each phase in the TUG test. The algorithm uses a multi-sensor fusion method for phase segmentation. The identification of the starting phase requires a combination of increasing segment sensor pressure values and a sudden upward change in Z-axis acceleration. The walking phase module confirms by detecting the rapid change pattern of the sensor pressure and the persistence of the forward acceleration feature. The identification of the recovery phase relies on the significant change in the gyroscope angular velocity in the Y-axis direction and the redistribution of the pressure distribution. The sitting back phase is determined by the sudden increase in pressure and the appearance of progressive directional acceleration.

[0053] The multi-feature boundary fusion recognition method can effectively avoid misjudgment by a single sensor signal, improving the accuracy and robustness of phase segmentation. The system sets corresponding time windows and amplitude thresholds to remove noise and other disturbances, ensuring the reliability of boundary recognition.

[0054] 1.2 Synchronous feature strategy extraction After determining the boundary phases, the module uses a fixed time window sliding sampling strategy for feature extraction. The time window is set to 200 milliseconds, which can capture sufficient signal details and ensure real-time processing efficiency. Within each time window, the system calculates the corresponding feature parameters according to the current boundary TUG phase type.

[0055] In the starting phase, the module focuses on extracting the change in body posture angle, including the complete change from the reclining angle in the sitting position to the upright angle in the static state. At the same time, the growth rate of the interval pressure and the time to reach the highest point are calculated, which can reflect the user's anatomical strength output characteristics and motion control ability. The center of gravity transfer process is obtained by analyzing the changes in the four bottom sensor pressure data, describing the user's process from relying on chair support to the highest center of gravity adjustment.

[0056] In the walking phase, the algorithm focuses on the regularity and symmetry features of the gait. Step frequency stability is quantified by analyzing the cycle coefficient of variation of the gait detected by the pressure sensor. Step frequency stability is evaluated by comparing the alternating change pattern of the extreme left and right pressure sensor signals. The body's ability to maintain posture is unstable by calculating the standard deviation of the forward inclination angle during walking.

[0057] The features of the recovery phase reduce the weakening of balance ability. The speed of the recovery action is smooth by analyzing the frequency domain characteristics of the gyroscope angular signal, and the smooth recovery action should start with a rapid high-frequency control component. The center of gravity stability is calculated by calculating the shortest time of the pressure center offset during the recovery process.

[0058] 1.3 Feature management construction and storage mechanism Each TUG stage corresponds to a feature processing set, which contains the feature data of all time windows in this stage and the corresponding time information. This organization follows the subsequent statistical analysis and pattern recognition processing.

[0059] The module also implements the standardization processing function of feature data, which maps different dimensional feature parameters to a unified numerical range, creating conditions for subsequent fusion analysis. At the same time, a data management mechanism based on cyclic topology is established, which can support real-time data updating and provide historical data access functions.

[0060] Further, the fusion judgment module algorithm architecture is added to avoid making conclusions on a single indicator, but to comprehensively use the user's muscle strength, gait performance, and performance in each action stage, and then give a comprehensive evaluation, simulating human experts.

[0061] The fusion judgment module assumes the core task of converting multi-source measurement data into a unified functional evaluation result. This module uses a hierarchical fusion technical architecture, through three levels of feature-level fusion, decision-level fusion, and intermediate-level fusion, to achieve a comprehensive evaluation of the user's functional state.

[0062] 2.1 Multi-level fusion framework design In reality, a person's functional state is complex. His muscle strength may be good, but his balance ability is poor; or his muscle strength is poor, but his coordination is good. Looking at any one indicator alone may lead to a biased conclusion.

[0063] The design concept of the hierarchical architecture is to decompose the complex multi-dimensional decision-making problem into several relatively simple sub-problems, and then gradually build the final judgment result through hierarchical means. This design not only ensures the transparency and interpretability of the decision-making process, but also improves the system's adaptability to different types of input data.

[0064] The module works in three levels, similar to the thinking process of a doctor when diagnosing: The first layer of thinking: first look at each specific performance. For example, how is the body control when standing up, and how is the balance when walking? Give each small performance a score. The algorithm idea is as follows: The first feature level fusion mainly processes feature data from different sensors in the same TUG stage. For example, in the starting stage, multiple features such as posture angle change, pressure growth curve, and center of gravity change trend need to be integrated as the comprehensive performance score of this stage. This layer uses the weighted average method, and the weight setting is based on the correlation strength of each feature to the functional evaluation goal of this stage.

[0065] Second layer thinking: Look at the performance of each stage as a whole. For example, although standing up is a bit slow, walking is very stable, and adjustment is very coordinated, the overall TUG performance may still be good. The algorithm is as follows: The second layer of decision fusion is responsible for the evaluation of the six TUG stages and the results of real-time gait monitoring. Each data source has its specific evaluation focus and scope of application, and the decision fusion needs to determine a reasonable weight distribution according to the reliability and complementarity of each data source. This layer also needs to handle the conflicts and contradictions that may exist between different data sources, and improve the reliability of the fusion results through consistency testing and outlier detection.

[0066] Third layer thinking: Put all the information together and weigh it. The results of the strength test, the results of the gait monitoring, and the results of the TUG test each have a certain reference value, but the importance may be different. The algorithm is as follows: The third layer of trend level fusion considers the relationship between historical data and current data. The results of a single test may be affected by multiple temporary factors, while the trend changes of multiple tests can more accurately reflect the real changes in the user's functional status. The intermediate level establishes a dynamic weight adjustment mechanism through fusion, enabling the system to correct the current evaluation results based on historical performance data.

[0067] 2.2 Adaptive weight distribution algorithm Weight distribution is a key technical approach to fusion judgment, directly affecting the accuracy of the final evaluation results. The module uses an adaptive weight distribution algorithm that can dynamically adjust the importance of each evaluation dimension based on the individual characteristics and current state of the user.

[0068] The algorithm first establishes a basic weight configuration, which is largely determined based on the statistical analysis results of clinical data. For general rehabilitation patients, muscle strength evaluation accounts for 40%, gait stability accounts for 35%, and TUG comprehensive performance accounts for 25%. However, different types of user groups have different functional impairment characteristics, and the weight distribution needs to be adjusted accordingly.

[0069] For stroke patients with hemiplegia, since their main problem is unilateral muscle weakness and left-right imbalance, the algorithm will increase the weight of muscle strength to 45% and increase the attention to pressure indicators. For older users, since their main risk is falling and balance problems, the algorithm will reduce the weight of muscle strength to 35% and increase the weight of gait stability to 40%.

[0070] In addition to static adjustment based on user type, the algorithm also implements a dynamic adjustment mechanism based on the current state. When detecting a decrease in the user's gait stability, the system will automatically increase the weight of muscle strength-related indicators to more sensitively monitor muscle strength improvement. When the user's gait stability is found to be decreasing, the system will correspondingly increase the weight of stability-related indicators.

[0071] 2.3 Abnormality resolution and decision rules To realize the conversion from qualitative judgment to quantitative evaluation, the module establishes an abnormality degree quantification system. Each type of gait abnormality is mapped to a continuous numerical interval from 0 to 1, where 0 represents complete normality and 1 represents severe abnormality. This quantification method enables the comparison and integration of different types and degrees of abnormality within a unified framework.

[0072] The calculation formula for stride abnormality degree is the ratio of the deviation of actual stride frequency and standard stride frequency to the allowed deviation range. Stride abnormality degree is determined by comparing the actual stride with the minimum required stride. The posture abnormality degree is calculated according to the relationship between the actual forward angle and the allowed forward angle range. The pressure degree is quantified by the ratio of the left-right pressure difference to the total pressure.

[0073] The system strictly follows the strategy specified in claim 8 to integrate the decision logic. When the muscle strength rating evaluation result is level 3 or below, and gait monitoring finds stride frequency or stride length abnormalities, the system determines that "function is significantly decreased". When the muscle strength rating evaluation result is level 4 to 5, but gait finds posture or pressure uneven function abnormalities, the system determines that "local function is abnormal".

[0074] For cases that do not meet the above specific conditions, the system uses a grading judgment method based on fusion scores. By setting different score threshold intervals, the user's functional status is divided into excellent, good, abnormal, moderate abnormality obvious, and decreased, etc. This grading method not only ensures the monitoring of the judgment, but also provides a clear reference for the user's functional status.

[0075] 2.4 Result output and feedback mechanism The output of the fusion judgment module not only includes the final functional status classification result, but also provides detailed analysis process and measures suggestions. The output result includes intermediate scores at each level, final fusion scores, functional status classification, and personalized rehabilitation suggestions.

[0076] Personalized suggestions are generated based on the functional status classification results and specific abnormal patterns. For users with "obvious functional decline", the system will suggest increasing the frequency of basic muscle strength training, focusing on the normalization of stride frequency and stride length, and seeking professional medical personnel for evaluation. For users with "local functional abnormalities", the system will give corresponding improvement suggestions according to the specific abnormal type, such as strengthening balance training or improving body posture, etc.

[0077] The system will adjust the importance of each indicator according to the specific situation. For example, for stroke patients, the system may pay more attention to energy consumption and left-right price; for the situation, the system may pay more attention to balance and stability.

[0078] The final decision will follow some clear rules. For example, if the walking strength is relatively low (3 or below), and there are obvious problems with gait (abnormal step frequency or step length), it will be determined as "obvious functional decline". If the walking strength is not bad (4-5), but there are problems with posture or balance, it will be determined as "local functional abnormality".

[0079] The benefits of the design are both scientific and practical. The science is not in the head making such decisions, but based on multi-dimensional monitoring data; the practicality is that it can give specific and persuasive recommendations.

[0080] That is, the system may simply tell you "your strength training is good, but you need to strengthen your balance", or "your overall coordination is good, but you need to enhance your pre-strength". Such suggestions are more effective than saying "need to exercise" to correct.

[0081] The above two modules are the key to making the rehabilitation walker "intelligent". The gait feature synchronous extraction module can assist in observing and recording fusion, and the judgment module can think and judge like a human expert. The combination of the two can provide personalized and professional rehabilitation guidance for users. BRIEF DESCRIPTION OF DRAWINGS

[0082] Figure 1 is the overall algorithm flowchart; Figure 2 is the gait detection algorithm flowchart; Figure 3 is the feature value checking logic in embodiment 1; Figure 4 is the voice prompt logic in embodiment 1; Figure 5 is one of the KNN model gait charts in embodiment 1; Figure 6 is the second KNN model gait chart in embodiment 1; Figure 7 is the TUG test and muscle strength evaluation fusion algorithm flowchart in embodiment 2 Figure 8 is the visualization of the six stages of standing, starting, walking, turning, returning, and sitting back in embodiment 2 as a time bar chart; Figure 9 is the total time of TUG test and abnormality detection result in embodiment 2; Figure 10 is the system diagram of embodiment 10. DETAILED DESCRIPTION

[0083] The application will be further described below in conjunction with the embodiments of the application and the accompanying drawings. Embodiment one

[0084] AsFigure 1 , 2 .

[0085] Complete code for "rule-based judgment + gait data + voice feedback" import numpy as np import pyttsx3 import time # Initialize the speech engine engine = pyttsx3.init() engine.setProperty('rate', 160) # Set speech rate engine.setProperty('volume', 1.0) def speak(text): print("[Prompt Voice]:", text) engine.say(text) engine.runAndWait() # Example: Analog sensor acquisition function def simulate_sensor_data(): # Returns the cadence, stride length, body lean angle, and left and right hand stress from a single sample. step_freq = np.random.normal(1.2, 0.2) # Hz stride_len = np.random.normal(0.38, 0.08) # m pitch_angle = np.random.normal(5, 5) # deg pressure_L = np.random.randint(10, 30) # N pressure_R = np.random.randint(10, 30) # N return step_freq, stride_len, pitch_angle, pressure_L, pressure_R # Threshold setting STEP_FREQ_RANGE = (0.8, 1.5) # Hz STRIDE_LEN_MIN = 0.4 # m MAX_PITCH_ANGLE = 10 # deg MAX_PRESSURE_DIFF = 10 # N # Main loop def monitor_loop(duration_sec=30, interval=1.0): t0 = time.time() while time.time() - t0 <duration_sec: # Simulated Data Acquisition freq, stride, pitch, pL, pR = simulate_sensor_data() print(f"Frequency = {freq:.2f}Hz, stride = {stride:.2f}m, pitch = {pitch:.1f}°, L = {pL}N, R = {pR}N") # Step frequency judgment if freq <STEP_FREQ_RANGE[0]: speak("Your pace is too slow, please increase your speed.") elif freq>STEP_FREQ_RANGE[1]: speak("Your cadence is too fast, please slow down a bit") # Stride Length Judgment if stride <STRIDE_LEN_MIN: speak("Please try to take bigger steps") # Attitude Judgment if abs(pitch)>MAX_PITCH_ANGLE: speak("Please keep your body upright") # Center of gravity judgment if abs(pL - pR)>MAX_PRESSURE_DIFF: speak("The force applied on both sides is uneven; please apply force in a balanced manner.") time.sleep(interval) if __name__ == "__main__": print("Starting the gait monitoring and voice guidance system...") monitor_loop(duration_sec=30) # Run for 30 seconds for demonstration.

[0086] The sensor value judgment logic and voice prompt logic in the process are as follows Figure 3 , 4 .

[0087] Then a set of simulated sensor data (such as step frequency, step length, posture angle, pressure value, data omitted) is generated, and the KNN model is applied for demonstration calculation: Figure 6 Visualize the distribution of different categories of samples in the two-dimensional feature space and the classification boundary.

[0088] Accuracy: 75%; Good performance in accurately identifying "normal gait" (label = 0); Insufficient identification of abnormal gait (too few samples); Figure 5 Color background area: represents the "gait category area" automatically divided by the model according to the training results: Green: normal gait (label = 0) Red: slow step frequency (label = 1) Blue: small step length (label = 2) Orange: excessive forward body inclination (label = 3) Purple: unstable center of gravity (label = 4). Example Two

[0089] As Figure 7 , the above algorithm in Example 1 is integrated with the Timed Up and Go test (TUG).

[0090] Figure 8 , 9 The six consecutive stages of the "Timed Up and Go" test (TUG, Timed Up and Go) and their corresponding time distribution during typical execution are shown. The graph takes the time axis as the horizontal axis, and breaks down the entire test process into key action segments such as "standing up", "accelerating start", "walking forward", "turning around", "returning", and "sitting back", and represents the duration of each stage in the form of colored rectangular blocks.

[0091] In the illustrated data, the total duration of TUG is about 12.5 seconds, and the time consumption of each stage is as follows: Standing up (2.0 seconds): reflects the patient's reaction and lower limb strength during the process from sitting to standing; Accelerating start (1.5 seconds): the stability and control of the patient's walking after standing up; Walk forward (3.0 seconds): the rhythm and coordination of the patient during the 3-meter walk; Turn around (1.2 seconds): the balance of the patient when turning around; Return (3.0 seconds): the persistence of the return gait; Sit back (1.8 seconds): the process of finally sitting back in the chair, reflecting the end control ability.

[0092] If we consider the ability of the human body to complete the above actions, it is actually the output of the lower limb muscle strength, the auxiliary adjustment of the upper limb muscle strength to balance the body, and the transmission and adjustment of the core (abdominal muscle) strength to the whole body and limbs. That is, the duration of each stage can reflect the user's ability to perform each stage (process). Therefore, the above "stand-up and walk timing test" is converted into a preliminary screening of muscle strength.

[0093] The duration of each stage = a "time dimension index" of the user's functional execution ability TUG ability assessment and abnormal performance in each stage Examples of clinical implications: If the standing time is > 4 seconds, it usually indicates a decline in balance or insufficient knee joint strength; If the turning speed is too fast or too slow, it may reflect a decline in coordination or disorientation; If the return segment is significantly slower than the forward segment, it indicates a decrease in physical fitness or attention concentration; If there is repeated adjustment or slow motion in the sit-back stage, it indicates a decrease in center of gravity control and end judgment.

[0094] TUG recommended time for each stage and voice strategy linkage table: The time used in each stage of the TUG test (especially the "standing" and "accelerating start" stages) can indirectly reflect the lower limb muscle strength and control ability of the user. The slower the standing stage, the more likely it is that the muscle strength is insufficient. Simple, easy to use, suitable for families, dynamic reflection of function. At the same time, it is affected by coordination, joint mobility, pain, etc., and is non-specific.

[0095] If a patient's "standing time exceeds 3.5 seconds", possible reasons include: Insufficient lower limb muscle strength (common); Balance or proprioceptive disorders; Joint mobility is limited or painful; Cognitive hesitation or slow reaction; Note: prolonged time ≠ muscle weakness, but in most cases it is indeed related to weak muscle strength, especially after excluding other disorders.

[0096] Therefore, the functional muscle strength assessment module is added to the rehabilitation walker: such as "standing time > 3.5s → judging muscle strength weak level 1"; "<1.5s → normal"; Trend analysis module: record TUG time curve for 7 consecutive days, such as continuous shortening, which reflects muscle strength improvement; Combine sensor pressure data: if the pressure sensor shows that the standing force value is low and the time is long, it is more supportive of the "muscle strength deficiency" conclusion.

[0097] Therefore, the above modules of the walker (functional muscle strength assessment module, trend analysis module) can be used as daily household screening, and can be used as a simplified muscle strength assessment method.

[0098] "Automatic muscle strength level score based on TUG time" logic or model: 1, Model design idea: By analyzing the time of each segment of the patient in the six stages of TUG, especially the three "high muscle strength related stages" of standing, accelerating starting and sitting back, the lower limb muscle strength is judged (1-5 levels): automatic scoring of lower limb muscle strength level. This scoring system is particularly suitable for real-time assessment and voice feedback modules in rehabilitation walker devices.

[0099] Muscle strength level assessment table: 2. Input variables Get the following time (unit: seconds) from the sensor or recording system: t_rise: standing time (standing to standing) t_start: starting time (after standing to the first step) t_sit: sit back time (turn back to completely sit down) Scoring logic: def evaluate_leg_strength(t_rise, t_start, t_sit): # Weight distribution: 50% for standing, 20% for starting, and 30% for sitting score = 0 # Standing score if t_rise<2.0: score += 5 elif t_rise<3.0: score += 4 elif t_rise<4.0: score += 3 elif t_rise<5.0: score += 2 else: score += 1 # Starting score if t_start < 1.5: score += 2 elif t_start<2.5: score += 1 else: score += 0 # Return to rating if t_sit < 2.0: score += 3 elif t_sit<3.0: score += 2 elif t_sit<4.0: score += 1 else: score += 0 # Total score is 10 points, mapped to levels 1-5 if score>= 9: level = 5 elif score>= 7: level = 4 elif score>= 5: level = 3 elif score>= 3: level = 2 else: level = 1 return level 4. Output Results For example: evaluate_leg_strength(t_rise=3.2, t_start=2.0, t_sit=2.8) # Return to Level 3 (Moderate Muscle Weakness) You can use the returned level as a basis for assessing the rehabilitation phase, and provide voice prompts such as: "This sitting up and sitting back movement indicates that your current lower limb muscle strength is at a moderate level. We recommend that you continue to strengthen your training." See the test procedure. Figure 8 .

[0100] 5. Suggestions for Further Development By combining AI-powered gait anomaly detection, the scoring can be further calibrated; The scoring weights can be dynamically adjusted based on the patient's age and disease course (e.g., the time standard can be relaxed for elderly patients). Example 3

[0101] Pseudocode for the gait feature synchronous extraction module.

[0102] Algorithm: Gait feature synchronous extraction module; Input: sensor data stream (sensor_stream), TUG stage labels (stage_labels); Output: Synchronization features by stage; BEGIN Initialize the feature container features_by_stage = {} Time window size window_size = 200ms FOR each time window in sensor_stream / / 1. Get the current stage identifier current_stage = Get the current TUG stage (timestamp) / / 2. Extract the original features of the current window Acceleration data = Extract acceleration(sensor_stream, window_size) Gyroscope data = Extract gyroscope(sensor_stream, window_size) Stress data = Extract stress(sensor_stream, window_size) / / 3. Calculate specific features based on stage type SWITCH current_stage CASE "Standing-up Phase": Attitude change rate = Calculated forward tilt angle change rate (gyroscope data) Pressure growth curve = Calculation of pressure change gradient (pressure data) Center of gravity transfer trajectory = Calculation of changes in center of gravity position (pressure data) CASE "Walking Phase": Step frequency stability = Calculation of gait period consistency (stress data) Stride Symmetry = Calculate left-right gait symmetry (pressure data) Posture Maintenance = Calculate forward lean angle fluctuation range (gyroscope data) CASE "Turn Phase": Turn Smoothness = Calculate angular velocity smoothness (gyroscope data) Center of Gravity Stability = Calculate center of gravity excursion range (pressure data) Dynamic Balance = Calculate pressure distribution changes (pressure data) / / Similar processing for other phases... END SWITCH / / 4. Feature Normalization and Storage Normalized Features = Normalize current stage features features_by_stage[current_stage].add(Normalized Features, Timestamp) END FOR / / 5. Calculate Comprehensive Stage Feature Indicators FOR each stage in features_by_stage stage_summary = Calculate stage summary features(features_by_stage[stage]) features_by_stage[stage].summary = stage_summary END FOR RETURN features_by_stage END Sub-algorithm: Get Current TUG Phase Input: Current Timestamp timestamp Output: Phase Identifier stage_id BEGIN / / Identify phase boundaries based on sensor signal features IF Detect sudden handle pressure increase AND Z-axis acceleration peak THEN IF Previous State == "Sitting" THEN RETURN "Standing Phase" END IF END IF IF Detect periodic pressure changes AND forward acceleration persists THEN RETURN "Walking Phase" END IF IF significant angular velocity change detected AND rapid pressure distribution change THEN RETURN "turn phase" END IF / / Other phase identification logic... RETURN current_stage END Example Four Fusion judgment module pseudocode: Algorithm: Multi-level fusion judgment module Input: gait features gait_features, muscle level muscle_level, TUG phase features tug_features Output: comprehensive function assessment result assessment_result BEGIN / / First level: feature-level fusion FOR each phase in tug_features phase weight = get phase weight (phase type) phase score = 0 FOR each feature in tug_features[phase] feature weight = get feature weight (feature type, user type) feature abnormality = calculate abnormality (feature value, standard range) phase score += feature weight × (1 - feature abnormality) END FOR TUG comprehensive score[phase] = phase score × phase weight END FOR / / Second level: phase-level fusion TUG total score = SUM(TUG comprehensive score[all phases]) / / Calculate gait monitoring comprehensive score gait abnormality weight = {step frequency: 0.25, step length: 0.30, posture: 0.20, pressure: 0.25} gait comprehensive score = 1.0 FOR each abnormality type in gait_features IF there is an abnormality THEN Abnormal severity = calculate abnormality degree (gait_features[abnormal type]) Gait composite score = gait_abnormal_weight[abnormal type] × abnormal severity END IF END FOR / / Third layer: temporal level fusion Muscle base score = muscle_level / 5.0 / / Dynamic weight allocation Fusion weight = calculate adaptive weight(user type, muscle_level, historical stability) Final fusion score = muscle base score × fusion_weight.muscle + gait composite score × fusion_weight.gait + TUG total score × fusion_weight.TUG / / Functional status classification decision Functional status = execute classification decision tree(muscle_level, gait_features, final fusion score) RETURN { fusion score: final fusion score, functional status: functional status, layer scores: {muscle base score, gait composite score, TUG total score}, suggested measures: generate personalized recommendations(functional status, muscle_level) } END Sub-algorithm: execute classification decision tree Input: muscle level muscle_level, gait features gait_features, fusion score fusion_score Output: functional status classification BEGIN / / Implement the fusion strategy of claim 8 / / Strategy 1: muscle level ≤ 3 and presence of step frequency or stride length abnormalities IF muscle_level ≤ 3 THEN IF gait_features.step_frequency_abnormal OR gait_features.stride_length_abnormal THEN RETURN "significant functional decline" END IF END IF Strategy 2: Muscle strength grade 4-5 but with postural or stress abnormalities IF muscle_level ≥ 4 THEN IF gait_features.posture abnormality OR gait_features.uneven pressure THEN RETURN "localized dysfunction" END IF END IF / / Grading judgment based on fusion score If fusion_score ≥ 0.85 then RETURN "Excellent functionality" ELSE IF fusion_score ≥ 0.70 THEN RETURN "Functionally working" ELSE IF fusion_score ≥ 0.55 THEN RETURN "Mild Functional Impairment" ELSE IF fusion_score ≥ 0.40 THEN RETURN "Moderate functional impairment" ELSE RETURN "Significant decline in function" END IF END Sub-algorithm: Calculating the degree of anomaly Input: feature_value, normal_range Output: Abnormality degree [0,1] BEGIN Standard median = normal_range.median Permissible deviation = normal_range.standard deviation IF eigenvalue in normal_range THEN RETURN 0.0 / / Normal ELSE Deviation level = ABS(feature_value - standard median) / allowable deviation Abnormality level = MIN(1.0, Deviation level / 2.0) / / Limit to the range [0,1] RETURN degree of exception END IF END Sub-algorithm: Calculate adaptive weights Input: user type user_type, muscle level muscle_level, historical stability stability Output: weight distribution weights BEGIN / / Base weight configuration base_weights = {muscle: 0.4, gait: 0.35, TUG: 0.25} / / User type adjustment IF user_type == "stroke patient" THEN base_weights = {muscle: 0.45, gait: 0.35, TUG: 0.20} ELSE IF user_type == "elderly user" THEN base_weights = {muscle: 0.35, gait: 0.40, TUG: 0.25} END IF / / Dynamic adjustment based on current state IF muscle_level ≤ 2 THEN base_weights.muscle += 0.1 base_weights.gait -= 0.05 base_weights.TUG -= 0.05 END IF IF stability<0.5 THEN base_weights.gait += 0.1 base_weights.muscle -= 0.05 base_weights.TUG -= 0.05 END IF / / Weight normalization total = SUM(base_weights.all weights) FOR each weight in base_weights weight = weight / total END FOR RETURN base_weights END Example Five

[0103] As Figure 10 , a comparison of the three speech prompt threshold determination algorithms is made.

Claims

1. A gait monitoring algorithm for a traditional four-legged walker, a 6-axis gyroscope is installed in the middle of the walker's four legs, and pressure sensor data is installed on the handle and the bottom of the four legs, comprising: a sensor data acquisition module for acquiring raw sensor data related to the user's gait, including: acceleration sensor data to obtain three-axis acceleration information; gyroscope sensor data to obtain three-axis angular velocity information; pressure sensor data to obtain handle pressure and ground contact pressure information; characterized by extracting gait features from raw sensor data, including: a gait feature extraction module for extracting gait feature values from the raw sensor data, including: step frequency, calculated by analyzing the periodic fluctuations of pressure or acceleration to calculate the number of steps per second; stride length, calculated by integrating acceleration or the interval between left and right foot contact; body posture, calculated by integrating gyroscope angular velocity to calculate the body inclination angle; pressure symmetry, calculated by the difference between the handle or column pressure to balance the left and right forces; gait stability, determined by statistical analysis of consecutive step frequency or stride length changes to determine gait rhythm stability; abnormal gait judgment, based on the gait feature values to identify the user's gait state; ground contact duration, the time each foot continuously contacts the ground; using at least one of the following algorithms: preset threshold rule judgment algorithm, set the normal range threshold of each gait feature value, when the feature value exceeds the threshold, it is judged as abnormal; classification recognition algorithm based on K-nearest neighbor or decision tree, learn normal and abnormal gait patterns through training samples; statistical anomaly detection algorithm based on sliding time window, identify abnormal fluctuations by analyzing the statistical distribution of feature values within the time window; voice guidance feedback module, when the gait abnormality judgment module detects that the gait deviates from the normal standard, it real-time broadcasts the corresponding voice prompt information to the user, the voice prompt content is determined according to the detected abnormal type.

2. The gait monitoring algorithm of claim 1, wherein, The method for obtaining the ground contact duration is as follows: by analyzing the load change rule of the walker's pressure sensor, based on the mechanical characteristics of the human-walker coupled gait, the left and right ground contact duration of the user's foot is inferred, specifically including: monitoring the continuous rising / falling time of the four pressure sensors; combined with the body inertia deflection direction displayed by the acceleration sensor; map the interval between pressure duration and falling time to left and right stable rhythm.

3. The gait monitoring algorithm of claim 1, wherein, The judgment rules of the preset threshold rule judgment algorithm include: step frequency threshold rule: step frequency should be maintained within 0.8-1.5Hz; stride length threshold rule: stride length should be greater than 0.4 meters; body posture threshold rule: body inclination angle should be controlled within ±10 degrees; pressure symmetry threshold rule: the left and right pressure difference should be less than the preset pressure tolerance value; when any rule is violated, the corresponding voice feedback prompt is triggered.

4. The gait monitoring algorithm of claim 1, wherein, The classification recognition algorithm includes the following steps: collect gait sample data from multiple users and perform manual annotation, the annotation categories include normal gait, step frequency abnormality, stride length abnormality, posture abnormality, and uneven pressure; use the labeled samples to train a K-nearest neighbor or decision tree classification model; input the real-time extracted gait feature values into the trained classification model, and output the gait state classification result; Generate corresponding voice guidance content according to the classification result.

5. The gait monitoring algorithm of claim 1, wherein, The statistical anomaly detection algorithm of the sliding time window comprises: Calculate the statistical characteristics of the gait feature values in the preset time window, including mean, variance, and change rate; Set a tolerance interval, and determine that it is abnormal when the statistical characteristics exceed the tolerance interval; An unsupervised learning method such as isolation forest or a support vector machine is used for anomaly detection; Trigger voice early warning when anomalies occur in continuous multiple time windows.

6. A gait monitoring and muscle strength monitoring fusion algorithm, characterized in that, The sensor layout and gait monitoring algorithm according to claim 1 further comprises: A TUG test data acquisition module for acquiring time segmentation data and sensor data during the user's execution of the standing-walking timing test, comprising: Standing time, recording the time from sitting to standing completely; Accelerated start-up time, recording the time from standing to starting walking; Walking forward time, recording the time of walking 3 meters forward; Turn time, recording the completion time of the turning action; Return time, recording the time of walking back; Sitting back time, recording the time from turning back to sitting completely; Synchronously acquiring acceleration, gyroscope, and pressure sensor data in each stage; A gait feature synchronous extraction module for synchronously extracting gait features in each stage of the TUG test, comprising: Phase recognition based on multi-sensor fusion: The starting phase is recognized by the combination of pressure value increment and sudden upward Z-axis speed; The walking phase is recognized by the rapid change pattern of pressure and the persistence of forward intensity; The confirmation phase relies on the significant change of gyroscope Y-axis angular velocity and the redistribution of pressure distribution; A sliding sampling strategy with a fixed time window of 200 milliseconds is used for feature extraction; A muscle strength grade evaluation module for calculating muscle strength grade scores based on the segmented time data of the TUG test, comprising: Score the standing time, the shorter the standing time, the higher the score; Score the accelerated start-up time to reflect the action start-up ability; Score the sitting back time to reflect the center of gravity control ability; Calculate the comprehensive score by using the weighted average method, and map the score to the muscle strength grade of 1-5; A fusion judgment module using a hierarchical fusion architecture: Feature level fusion: same processing of feature data from different sensors at TUG level; Decision level fusion: comprehensive evaluation of six TUG stages and real-time gait monitoring results; Fluctuation level fusion: considering the relationship between historical data and current data, a dynamic weight adjustment mechanism is established; An individualized feedback module for generating individualized voice guidance and training suggestions based on muscle strength grade evaluation results and gait anomaly detection results.

7. The gait monitoring and myodynamic monitoring fusion algorithm according to claim 6, characterized in that, The scoring rules of the muscle strength grade evaluation module are: Standing time less than 2 seconds is 5 points, 2-3 seconds is 4 points, 3-4 seconds is 3 points, 4-5 seconds is 2 points, and more than 5 seconds is 1 point; Accelerated start-up time less than 1.5 seconds is 2 points, 1.5-2.5 seconds is 1 point, and more than 2.5 seconds is 0 point; Sitting back time less than 2 seconds is 3 points, 2-3 seconds is 2 points, 3-4 seconds is 1 point, and more than 4 seconds is 0 point; Total score 9-10 corresponds to muscle strength grade 5, 7-8 corresponds to grade 4, 5-6 corresponds to grade 3, 3-4 corresponds to grade 2, and 0-2 corresponds to grade 1.

8. The gait monitoring and muscle strength monitoring fusion algorithm of claim 1, wherein, The fusion judgment module adopts the following fusion strategy: When the muscle strength grade evaluation result is level 3 and below, and gait monitoring finds abnormal step frequency or stride, it is determined that the function has decreased significantly; When the muscle strength grade evaluation result is level 4-5, but gait monitoring finds abnormal posture or uneven pressure, it is determined that the local function is abnormal; Combine the trend changes of multiple tests to judge the improvement or deterioration trend of function.

9. The gait monitoring and muscle strength monitoring fusion algorithm of claim 1, wherein, The TUG test data acquisition module also includes an abnormality detection function: Monitor whether the time of each stage exceeds the preset abnormal range; When the standing stage exceeds 3.5 seconds, prompt "please stand up decisively and keep the body weight stable"; When the walking stage exceeds 4.5 seconds, prompt "please try to take a larger step"; When the turning stage exceeds 2.5 seconds, prompt "please turn slowly and keep the body upright"; The individualized feedback module generates different training suggestions according to the muscle strength grade: When the muscle strength grade is 1-2, it is recommended to conduct basic muscle strength training and increase the training intensity; When the muscle strength grade is 3, it is recommended to maintain the current training intensity and focus on the action quality; When the muscle strength grade is 4-5, it is recommended to conduct maintenance training to prevent functional degradation; It also includes a trend analysis module, which is used to: Record the time data and muscle strength grade changes of continuous TUG tests; Analyze the trend of muscle strength improvement or degradation; When the continuous 7-day test shows that the time continues to shorten, it is determined that the muscle strength improvement trend; When the continuous 7-day test shows that the time continues to lengthen, it is determined that the function has a deterioration trend, and a warning prompt is issued.

10. A rehabilitation walker using a gait monitoring and muscle strength monitoring fusion algorithm, characterized by, The algorithms of claims 1-5 and claims 6-9 are used.

Images