Comprehensive walking function evaluation method and device for old people

Abstract

The invention discloses a comprehensive walking function assessment method for old people, which comprises the following steps: S1, collecting motion posture data from an individual motion process, and processing to obtain motion posture feature data; s2, collecting video data in an individual movement process, and processing to obtain a time attitude sequence; s3, plantar pressure data are collected in the individual movement process, and plantar pressure feature data are obtained through processing; and S4, inputting the processed data into a GRU model to carry out walking function type identification. According to the method, the index data, the video data and the foot pressure data during individual walking are fused, so that comprehensive evaluation on the individual walking function is achieved, and the accuracy, universality and efficiency of individual motion evaluation are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of human behavior and motion assessment, and in particular to a method and device for assessing the comprehensive walking function of the elderly. Background Art

[0002] Traditional walking assessment for the elderly mainly relies on manual observation and scale scoring, such as the Timed Up and Go Test (TUGT) and the 6-minute walk test (6MWT).

[0003] With the rapid development of sensor technology, the Internet of Things, and artificial intelligence, an increasing number of intelligent assessment systems have been developed to more accurately evaluate the walking function of older adults. Gait analysis systems based on pressure sensors can detect the distribution of plantar pressure during walking and, by analyzing this data, assess walking stability and balance. Furthermore, wearable devices such as smartwatches and smart shoes can monitor parameters such as walking speed and stride length, providing new tools for assessing walking function.

[0004] However, existing evaluation methods have the following shortcomings:

[0005] 1. Subjectivity and limitations: Traditional methods often rely on manually defined rules and feature design, which can be subject to the subjectivity and limitations of expert knowledge. Different experts may have different understandings and definitions, which affects the consistency and accuracy of evaluation results.

[0006] 2. Poor versatility and adaptability: Traditional methods often require a flat and wide site, and may not be universal and adaptable for application scenarios in narrow spaces.

[0007] 3. Lack of generalization: Traditional methods often struggle to adapt to individual differences and variations. Because they are typically based on fixed evaluation methods, they may not be well adapted to individual variations in movement patterns and feature expression abilities, resulting in poor individualization and generalization of evaluation results.

[0008] 4. High labor costs: Traditional methods typically require a lot of manual effort, including method teaching, measurement, and result interpretation. This consumes a lot of manpower and time, and makes large-scale evaluation difficult.

[0009] The above shortcomings limit the accuracy, versatility and efficiency of traditional methods for assessing walking function in the elderly.

[0010] In view of this, it is necessary to improve the motion evaluation method in the prior art to solve the above problems.

[0011] Therefore, those skilled in the art are committed to developing a comprehensive walking function assessment method and portable device for the elderly. Summary of the Invention

[0012] To solve the above problems, the present invention provides a method and a portable device for evaluating the comprehensive walking function of the elderly.

[0013] A comprehensive walking function assessment method for the elderly includes the following steps:

[0014] S1: Collect motion posture data from the individual during motion, and process the motion posture data to obtain motion posture feature data;

[0015] S2: Collect video data from the individual's movement process and process the video data to obtain a temporal posture sequence;

[0016] S3: Collect plantar pressure data from the individual during exercise, and process the plantar pressure index data to obtain plantar pressure characteristic data;

[0017] S4: The motion posture feature data obtained in step S1, the time posture sequence obtained in step S2, and the plantar pressure feature data obtained in step S3 are input into the trained GRU model to identify the type of walking function.

[0018] Furthermore, step S1 may specifically include:

[0019] S11: using a wearable device with an integrated accelerometer and gyroscope to collect motion posture data of the individual during motion, wherein the motion posture data includes the linear acceleration of the individual on the three coordinate axes of X, Y, and Z and the angular velocity of the individual around the X, Y, and Z axes;

[0020] S12: Preprocessing the motion posture data;

[0021] S13: Extracting features from the pre-processed motion posture data to obtain motion posture feature data.

[0022] Furthermore, step S12 may specifically include:

[0023] S121: performing zero bias calibration on the motion posture data, collecting data for a certain period of time in a stationary state, calculating a zero bias value, and subtracting the zero bias value from the motion posture data to eliminate zero bias error;

[0024] S122: Filter the motion posture data using a low-pass filter to remove high-frequency noise interference, with the cutoff frequency of the low-pass filter set to 25 Hz;

[0025] S123: Convert the filtered motion posture data into a digital signal through an A / D converter and resample it. The resampling frequency is set to 150 Hz.

[0026] Furthermore, the motion posture feature data may include:

[0027] The mean and standard deviation of acceleration reflect the average intensity and fluctuation of movement;

[0028] The peak and valley values ​​of angular velocity reflect the maximum rotation speed and reverse rotation speed of the movement;

[0029] The correlation coefficient between acceleration and angular velocity is used to analyze the degree of correlation between the two;

[0030] The total acceleration energy of the movement, obtained by integrating the square of the acceleration over a period of time, measures the overall energy expenditure of the movement;

[0031] The rate of change of angular velocity reflects the speed of change of motion rotation.

[0032] Furthermore, step S2 may specifically include:

[0033] S21: Collect video data from the individual during exercise using a mobile device with video recording function. The collected movements are divided into standard movements and test movements.

[0034] S22: Preprocessing the video data, including performing standardization conversion on the acquired video data and normalizing the video parameters to keep the format, resolution, and frame rate of the video data consistent;

[0035] S23: Performing in-depth analysis on the pre-processed video data to obtain a temporal posture sequence, including:

[0036] Use the lightweight multimodal perception network Mediapipe Holistic model to obtain the coordinate sequence of key points of multiple parts of the human body.

[0037] Normalize and standardize the coordinate sequences of key points of multiple parts of the human body to generate multi-dimensional time posture sequences.

[0038] The multi-dimensional time posture sequence is input into the optimized LSTM model to obtain the time posture sequence.

[0039] Furthermore, step S23 may include: performing noise preprocessing on the coordinate sequence of key points of multiple parts of the human body using adaptive median filtering or Wiener filtering.

[0040] Furthermore, in step S23, the normalization and standardization processing method for the coordinate sequence of key points of multiple parts of the human body is specifically as follows:

[0041] The center point coordinates (x it ,y it) and take the average value to get the average center point coordinates of each part (mid x ,mid y ), and on this basis, the translation transformation is calculated as follows:

[0042] x′ it =x it -mid x

[0043] y′ it =y it -mid y

[0044] All coordinates are scaled normalized according to the width and height of the video image. The formula is:

[0045]

[0046] Among them, width represents the width of the video image, and height represents the height of the video image.

[0047] Further, step S3 includes:

[0048] S31: collecting plantar pressure data;

[0049] S32: Preprocessing of plantar pressure data:

[0050] S33: performing waveform positioning on the pre-processed plantar pressure data to determine the characteristic points of each walking cycle;

[0051] S34: Extracting plantar pressure characteristic data of each walking cycle.

[0052] Further, step S32 includes:

[0053] S321: using a signal amplifier to pre-amplify the plantar pressure data, and using a low-pass filter to filter the amplified signal;

[0054] S322: Convert the plantar pressure data into a digital signal, use an adaptive filtering algorithm to remove power frequency interference and baseline drift noise, use a wavelet transform denoising method to process the data, and resample at a resampling frequency of 500 Hz.

[0055] On the other hand, the present invention also provides a portable device for evaluating the comprehensive walking function of the elderly, which uses the above-mentioned evaluation method, including:

[0056] Wearable devices: including accelerometers and gyroscopes, used to collect motion posture data;

[0057] Mobile terminal: has video recording function, used to collect video data;

[0058] Plantar pressure measurement equipment: with plantar pressure measurement function, used to collect individual plantar pressure data during walking;

[0059] Index data processing module: capable of preprocessing the motion posture data collected by the wearable device and extracting features;

[0060] Video processing module: used to perform standardized conversion on the video data collected by the mobile terminal and normalize the video parameters;

[0061] Computer vision module: used to identify key points in the processed video and generate time posture sequences;

[0062] Plantar pressure data processing module: used to pre-process the plantar pressure data collected by the plantar pressure measurement device and extract features;

[0063] Deep learning module: used to process motion posture feature data, time posture sequence and plantar pressure feature data, and input them into the trained GRU model to identify the type of walking function.

[0064] Technical Effects

[0065] This invention integrates individual walking metrics, video data, and foot pressure data to achieve a comprehensive assessment of an individual's walking function. It uses computer vision technology to analyze video data and performs comprehensive calculations based on metrics and foot pressure data to obtain accurate and reliable walking function assessment results. This invention fully adapts to individual differences in movement patterns and characteristic expressions, effectively improving the accuracy, versatility, and efficiency of individual movement assessment.

[0066] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of a method for evaluating comprehensive walking function of the elderly according to a preferred embodiment of the present invention;

[0068] Figure 2 This is a GRU model structure diagram of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following describes several preferred embodiments of the present invention with reference to the accompanying drawings to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0070] In the drawings, components with identical structures are denoted by the same reference numerals, and components with similar structures or functions are denoted by similar reference numerals. The size and thickness of each component shown in the drawings are arbitrary and are not limited by the present invention. For clarity, the thickness of components in some places in the drawings is appropriately exaggerated.

[0071] Example 1

[0072] Figure 1 A method for evaluating the comprehensive walking function of the elderly is disclosed, comprising the following steps:

[0073] S1, collecting motion posture data from the individual during motion, and processing the motion posture data to obtain motion posture feature data, specifically including:

[0074] S11, using a wearable device with an integrated accelerometer and gyroscope to collect motion posture data of an individual during motion, wherein the accelerometer is used to obtain the linear acceleration of the individual on the three coordinate axes of X, Y, and Z, and the gyroscope is used to obtain the angular velocity of the individual around the X, Y, and Z axes;

[0075] S12, pre-processing the motion posture data, including the following steps:

[0076] S121, performing zero bias calibration on the motion posture data, calculating a zero bias value by collecting data for a certain period of time in a stationary state, and subtracting the zero bias value from subsequent motion posture data to eliminate zero bias errors;

[0077] S122, using a low-pass filter to filter the motion posture data to remove high-frequency noise interference. Setting the cutoff frequency of the low-pass filter to 25 Hz can achieve a better filtering effect;

[0078] S123, converting the filtered motion posture data into a digital signal through an A / D converter and resampling it. Setting the resampling frequency to 150 Hz can achieve better processing results;

[0079] S13, extracting features from the pre-processed motion posture data to obtain motion posture feature data, specifically including:

[0080] S131, extracting features from the pre-processed motion posture data, including but not limited to:

[0081] a) The mean and standard deviation of acceleration, reflecting the average intensity and fluctuation of movement;

[0082] b) The peak and valley values ​​of angular velocity, reflecting the maximum rotational speed and reverse rotational speed of the movement;

[0083] c) Correlation coefficient between acceleration and angular velocity, used to analyze the degree of correlation between the two;

[0084] d) Total acceleration energy of the movement, obtained by integrating the square of the acceleration over a period of time, which measures the overall energy expenditure of the movement;

[0085] e) The rate of change of angular velocity reflects the speed of change of motion rotation.

[0086] S2, collects video data from the individual's movement process and processes the video data to obtain a temporal posture sequence, specifically including:

[0087] S21, using a mobile terminal with video recording function to collect video data from the individual during the movement process, and the collected movements are divided into standard movements and test movements;

[0088] S22, preprocessing the video data, including performing standardization conversion on the acquired video data and normalizing the video parameters to keep the format, resolution, and frame rate of the video data consistent;

[0089] The collection of video data may be completed by different devices, which often result in differences in resolution and frame rate, which is not conducive to the study of video data. Therefore, after obtaining the video data, it is necessary to perform a standardized conversion on it so that the format, resolution and frame rate of the video data remain consistent. The collected video data may differ depending on the different smart terminals. In order to better analyze and use the data, the parameters including the video format must be normalized. Commonly used video data file types include AVI format, MOV format, MPEG format, WEBM format, etc., and their different types are mainly caused by different video encoding methods. The MPEG format is further divided into MPEG-1, MPEG-2, and MPEG-4. In order to facilitate data processing, the action video file type saved after conversion in the present invention is MPEG-4, with a frame rate of 30 frames per second and a resolution of 340×256, that is, a frame width of 340 and a frame height of 256.

[0090] S23, performing in-depth analysis on the pre-processed video data to obtain a temporal posture sequence.

[0091] The lightweight multimodal perception network Mediapipe Holistic model is used to obtain the coordinate sequence of key points in multiple parts of the human body. The model can accurately identify the face and body joints, focus on the key points of the lower limb joints, and output a comprehensive key point coordinate sequence.

[0092] The coordinate sequences of key points of multiple parts of the human body are normalized and standardized to generate a multi-dimensional temporal posture sequence. The multi-dimensional temporal posture sequence is input into the optimized LSTM (long short-term memory artificial neural network) model for detailed type recognition to obtain the temporal posture sequence. The optimization method of the LSTM model can adopt the existing commonly used optimization methods.

[0093] In step S23, the identified coordinate sequence of multiple key points of the human body can also be subjected to noise preprocessing to obtain better results. Due to the influence of objective factors such as changes in ambient light and jitter of the shooting equipment during the data acquisition process, the identified coordinates of multiple key points may contain abnormal values. The present invention uses adaptive median filtering or Wiener filtering to process the discrete coordinate values ​​of the coordinates of multiple key points. These filtering technologies can adaptively adjust the filtering parameters according to the statistical characteristics of the data, effectively filter out the coordinates of zero values ​​or abnormal points, and better retain the high-frequency detail information of the data, thereby more reliably detecting the bone point data.

[0094] In step S23, due to differences in body proportions between individuals, different viewing angles and distances of the shooting equipment, the generated skeletal point data are inconsistent when representing the size of the human body. In order to eliminate the impact of these differences on subsequent analysis, the original data needs to be normalized and standardized. If the standardization operation is not performed, similar human body movements may produce different output results due to differences in data scale. The specific steps for normalizing and standardizing the coordinate sequence of key points of multiple parts of the human body are as follows:

[0095] By calculating the center point coordinates (x it ,y it ) and take the average value to get the average center point coordinates of each part (mid x ,mid y ). On this basis, the translation transformation calculation is performed, and the formula is as follows:

[0096] x′ it =x it -mid x

[0097] y′ it =y it -mid y

[0098] At the same time, precise alignment is performed on the spatial scale. All coordinates are normalized according to the width and height of the video image so that the posture to be measured and the standard posture are precisely aligned in space. The formula is as follows:

[0099]

[0100] Among them, width represents the width of the video image, and height represents the height of the video image.

[0101] S3, collecting plantar pressure data from the individual during exercise, and processing the plantar pressure index data to obtain plantar pressure characteristic data, as follows:

[0102] S31, select a thin film pressure sensor to collect plantar pressure data.

[0103] S32, pre-processing the collected plantar pressure data, the detailed steps are as follows:

[0104] S321 uses a signal amplifier to pre-amplify the plantar pressure data, enhancing the signal strength to a level suitable for subsequent processing. Simultaneously, a low-pass filter is used to filter the amplified signal, removing high-frequency noise and retaining the true pressure change information.

[0105] At step S322, the plantar pressure data is converted into a digital signal using an A / D converter. Adaptive filtering algorithms are used to remove noise from the data, such as power frequency interference and baseline drift. Wavelet transform denoising can also be used to address possible interference, such as electromyographic artifacts. After processing, the data is resampled at a resampling frequency of 500 Hz to make the data more regular in the temporal dimension, facilitating subsequent analysis.

[0106] S33 , performing waveform positioning on the pre-processed plantar pressure data to determine the characteristic points of each walking cycle.

[0107] By analyzing the pressure curve, we identify pressure peaks, which serve as key markers for a walking cycle. Based on the locations of these peaks, we extract pressure data for each complete walking cycle. Furthermore, we further segment the pressure changes within each walking cycle, pinpointing key moments in time when pressure in different areas of the foot begins to rise, peak, and decline, enabling more precise analysis of pressure patterns during walking.

[0108] S34, extracting plantar pressure characteristic data of each walking cycle.

[0109] 17 features are extracted for each walking cycle, including but not limited to:

[0110] a) Peak heel pressure: reflects the maximum pressure on the heel during walking.

[0111] b) Forefoot pressure peak: reflects the force applied to the forefoot during walking.

[0112] c) Heel-to-forefoot peak pressure ratio: used to assess the shift of center of gravity during walking.

[0113] d) Average plantar pressure: indicates the average pressure on the entire plantar during a walking cycle.

[0114] e) Pressure standard deviation: measures the degree of pressure fluctuation during the walking cycle.

[0115] f) Pressure rise time: the time from when the sole of the foot touches the ground to when the pressure reaches its peak, reflecting the force applied at the start of walking.

[0116] g) Pressure drop time: The time from the peak pressure to the foot leaving the ground, reflecting the cushioning condition at the end of walking.

[0117] h) Heel pressure duration ratio: the ratio of the heel pressure duration to the entire walking cycle time.

[0118] i) Forefoot pressure duration ratio: the ratio of forefoot pressure duration to the entire walking cycle time.

[0119] j) Movement distance of plantar pressure center of gravity: the distance that the plantar pressure center of gravity moves in the horizontal direction during one walking cycle.

[0120] k) Pressure center of gravity moving speed: the ratio of the pressure center of gravity moving distance to the walking cycle time.

[0121] l) Left and right foot pressure symmetry index: compares the pressure difference in corresponding areas of the left and right feet during the same walking cycle.

[0122] m) Pressure change rate between adjacent walking cycles: measures the degree of change in pressure data between two adjacent walking cycles.

[0123] n) Plantar pressure distribution uniformity index: evaluates the uniformity of pressure distribution in different areas of the sole of the foot.

[0124] o) Pressure peak occurrence time deviation: the deviation between the actual pressure peak occurrence time and the theoretical normal time.

[0125] p) Shape characteristics of the pressure curve: such as the slope and curvature of the curve, which reflect the trend and mode of pressure change.

[0126] q) Pressure change trend in a specific area: For example, whether the pressure in the arch area increases, decreases, or remains stable during the walking cycle.

[0127] S4, input the motion posture feature data obtained in step S1, the time posture sequence obtained in step S2, and the plantar pressure feature data obtained in step S3 into the trained GRU (gated recurrent unit) model to perform walking function type recognition. The training method can be implemented using the commonly used training methods in this field. Figure 2The structure of the GRU model in this embodiment is shown.

[0128] Example 2

[0129] The present invention also discloses a portable comprehensive elderly walking function assessment device, which adopts the above-mentioned assessment method and includes:

[0130] Wearable device, integrated with raw physiological signal acquisition module, including accelerometer and gyroscope, for collecting motion posture data;

[0131] The mobile terminal has a video recording function for collecting video data;

[0132] Plantar pressure measurement equipment, with plantar pressure measurement function, used to collect individual plantar pressure data during walking;

[0133] An indicator data processing module, capable of preprocessing the motion posture data collected by the wearable device and extracting features;

[0134] A video processing module, configured to perform standardized conversion on the video data collected by the mobile terminal and normalize the video parameters;

[0135] Computer vision module, used to identify key points in the processed video and generate temporal pose sequences;

[0136] The plantar pressure data processing module is capable of preprocessing the plantar pressure data collected by the plantar pressure measurement device and extracting features;

[0137] The deep learning module is used to process motion posture feature data, time posture sequence and plantar pressure feature data, and input them into the GRU model to identify the type of walking function.

[0138] As a further improvement of the present invention, the computer vision module selects a lightweight multimodal perception network Mediapipe Holistic model, which includes: a facial key point detection module, a body posture estimation module and a lower limb motion capture module.

[0139] The facial landmark detection module performs in-depth analysis of captured facial images, accurately detecting and locating the face area and identifying up to 468 facial landmarks. By outputting the precise coordinates of these landmarks, facial features such as eyebrow shape, eye contour, nose structure, and mouth shape can be meticulously delineated, providing rich and accurate data for subsequent applications such as facial expression analysis and identity recognition.

[0140] The body posture estimation module accurately infers the overall posture of the human body. It outputs the 3D coordinates of 33 major joints and body parts, including the head, neck, shoulders, arms, and waist. By tracking and analyzing these joint coordinates in real time, it can accurately capture various human movements and posture changes, such as standing, raising an arm, and turning, with excellent performance in capturing body posture details and accuracy.

[0141] The lower limb motion capture module is used to accurately capture and analyze lower limb movements. It can identify the position and motion state of lower limb joints, such as the hip, knee, and ankle. By continuously monitoring changes in the coordinates and angles of these joints, it can record detailed lower limb motion trajectories, such as stride length, cadence, leg lift height, and speed during walking. The lower limb motion data output by this module provides key support for applications such as sports training assessment, rehabilitation therapy monitoring, and lower limb motion simulation in human-computer interaction scenarios.

[0142] As a further improvement of the present invention, the lower limb motion capture module has a unique network architecture, which is mainly composed of two core parts: a lower limb coarse positioning network and a lower limb fine adjustment network.

[0143] Lower Limb Coarse Localization Network: Rapidly locates the lower limb region and keypoint probabilities. This network is specifically designed for preliminary detection of the human lower limbs and utilizes a deep convolutional neural network architecture. During data processing, it first applies a series of convolutional layers of varying sizes to comprehensively scan the input image or video frame. These convolutional layers act as sensitive "detectors," capturing diverse features related to the lower limbs from complex visual information. Subsequently, pooling layers filter and reduce the dimensionality of the extracted features, removing redundant information and improving feature representativeness and computational efficiency. After feature extraction and processing, the network utilizes a multi-layer perceptron to conduct in-depth feature analysis, achieving preliminary classification and location estimation of key parts of the lower limbs. It quickly delineates the approximate extent of the lower limbs in the input image and generates candidate frames for the lower limbs. Furthermore, for each candidate frame, the network generates a corresponding keypoint heatmap. These heatmaps act as "probability maps," visually displaying the probabilities of the locations of various lower limb keypoints (such as the hip, knee, and ankle) in the image using varying color and brightness distributions, laying the foundation for subsequent precise analysis.

[0144] Lower Limb Fine-Tuning Network: Accurately outputs the three-dimensional coordinates of key points of the lower limbs. The lower limb fine-tuning network is also built based on a deep convolutional neural network. Its core task is to finely calibrate and accurately regress the positions of key points of the lower limbs. The network takes as input the key point heat map generated by the lower limb coarse localization network and the original input image. During processing, the network uses convolutional layers and an advanced residual block structure to deeply mine and enhance the features of the input information. It carefully analyzes and repeatedly corrects the preliminary key point positions given by the lower limb coarse localization network, fully considering the complexity and diversity of human lower limb movements. Through continuous optimization and adjustment, the network can accurately output the precise three-dimensional coordinates of each key point of the lower limbs. These coordinate data can accurately depict the specific posture and movement details of the human lower limbs in three-dimensional space, providing reliable data support for high-precision lower limb motion capture and in-depth motion analysis.

[0145] The preferred embodiments of the present invention have been described in detail above. It should be understood that numerous modifications and variations based on the concepts of the present invention are possible without inventive effort by those skilled in the art. Therefore, any technical solution that can be derived by one skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A method for evaluating the comprehensive walking function of the elderly, characterized by: The following steps are involved: S1: Collect motion posture data from the individual during motion, and process the motion posture data to obtain motion posture feature data; S2: Collect video data from the individual's movement process and process the video data to obtain a temporal posture sequence; S3: Collect plantar pressure data from the individual during exercise, and process the plantar pressure index data to obtain plantar pressure characteristic data; S4: The motion posture feature data obtained in step S1, the time posture sequence obtained in step S2, and the plantar pressure feature data obtained in step S3 are input into the trained GRU model to identify the type of walking function.

2. The method for evaluating comprehensive walking function of the elderly according to claim 1, wherein: The step S1 comprises: S11: using a wearable device with an integrated accelerometer and gyroscope to collect motion posture data of the individual during motion, wherein the motion posture data includes the linear acceleration of the individual on the three coordinate axes of X, Y, and Z and the angular velocity of the individual around the X, Y, and Z axes; S12: Preprocessing the motion posture data; S13: Extracting features from the pre-processed motion posture data to obtain motion posture feature data.

3. The method for evaluating the comprehensive walking function of the elderly according to claim 2, wherein: The step S12 includes: S121: performing zero bias calibration on the motion posture data, collecting data for a certain period of time in a stationary state, calculating a zero bias value, and subtracting the zero bias value from the motion posture data to eliminate zero bias error; S122: Filter the motion posture data using a low-pass filter to remove high-frequency noise interference, with the cutoff frequency of the low-pass filter set to 25 Hz; S123: Convert the filtered motion posture data into a digital signal through an A / D converter and resample it. The resampling frequency is set to 150 Hz.

4. The method for evaluating comprehensive walking function of the elderly according to claim 2, wherein: The motion posture feature data includes: The mean and standard deviation of acceleration reflect the average intensity and fluctuation of movement; The peak and valley values ​​of angular velocity reflect the maximum rotation speed and reverse rotation speed of the movement; The correlation coefficient between acceleration and angular velocity is used to analyze the degree of correlation between the two; The total acceleration energy of the movement, obtained by integrating the square of the acceleration over a period of time, measures the overall energy expenditure of the movement; The rate of change of angular velocity reflects the speed of change of motion rotation.

5. The method for evaluating comprehensive walking function of the elderly according to claim 1, wherein: The step S2 comprises: S21: Collect video data from the individual during exercise using a mobile device with video recording function. The collected movements are divided into standard movements and test movements. S22: Preprocessing the video data, including performing standardization conversion on the acquired video data and normalizing the video parameters to keep the format, resolution, and frame rate of the video data consistent; S23: Performing in-depth analysis on the pre-processed video data to obtain a temporal posture sequence, including: Use the lightweight multimodal perception network Mediapipe Holistic model to obtain the coordinate sequence of key points of multiple parts of the human body. Normalize and standardize the coordinate sequences of key points of multiple parts of the human body to generate multi-dimensional time posture sequences. The multi-dimensional time posture sequence is input into the optimized LSTM model to obtain the time posture sequence.

6. The method for evaluating comprehensive walking function of the elderly according to claim 5, wherein: The step S23 includes: performing noise preprocessing on the coordinate sequence of the key points of multiple parts of the human body using adaptive median filtering or Wiener filtering.

7. The method for evaluating comprehensive walking function of the elderly according to claim 5, wherein: The method for normalizing and standardizing the coordinate sequence of key points of multiple parts of the human body in step S23 is: The center point coordinates (x it ,y it ) and take the average value to get the average center point coordinates of each part (mid x ,mid y ), and on this basis, the translation transformation is calculated as follows: x′ it =x it -mid x y′ it =y it -mid y All coordinates are scaled normalized according to the width and height of the video image. The formula is: Among them, width represents the width of the video image, and height represents the height of the video image.

8. The method for evaluating comprehensive walking function of the elderly according to claim 1, wherein: The step S3 comprises: S31: collecting plantar pressure data; S32: Preprocessing of plantar pressure data: S33: performing waveform positioning on the pre-processed plantar pressure data to determine the characteristic points of each walking cycle; S34: Extracting plantar pressure characteristic data of each walking cycle.

9. The method for evaluating comprehensive walking function of the elderly according to claim 8, wherein: The step S32 includes: S321: using a signal amplifier to pre-amplify the plantar pressure data, and using a low-pass filter to filter the amplified signal; S322: Convert the plantar pressure data into a digital signal, use an adaptive filtering algorithm to remove power frequency interference and baseline drift noise, use a wavelet transform denoising method to process the data, and resample at a resampling frequency of 500 Hz.

10. A portable comprehensive walking function assessment device for the elderly, characterized by: Using the evaluation method according to claim 1, include: Wearable devices: including accelerometers and gyroscopes, used to collect motion posture data; Mobile terminal: has video recording function, used to collect video data; Plantar pressure measurement equipment: with plantar pressure measurement function, used to collect individual plantar pressure data during walking; Index data processing module: capable of preprocessing the motion posture data collected by the wearable device and extracting features; Video processing module: used to perform standardized conversion on the video data collected by the mobile terminal and normalize the video parameters; Computer vision module: used to identify key points in the processed video and generate time posture sequences; Plantar pressure data processing module: used to pre-process the plantar pressure data collected by the plantar pressure measurement device and extract features; Deep learning module: used to process motion posture feature data, time posture sequence and plantar pressure feature data, and input them into the trained GRU model to identify the type of walking function.

Images