A comparative analysis method of Parkinson's disease gait based on time series characteristic curves

Abstract

The present invention provides a method for comparative analysis of Parkinson's disease gait based on time-series characteristic curves. The method comprises acquiring video data containing gait information, wherein the gait information includes gait information of a user before and after taking medication; performing posture recognition on a human body in the video data to extract position information of key body parts; calculating gait features based on the extracted key point information, and constructing a time-series curve based on the changes of these features over time; performing data processing on the extracted gait features using a digital filter; integrating local frequency and amplitude into a distance metric using a frequency-amplitude enhanced dynamic time warping method to quantify differences in gait motion; obtaining a curve difference distance; and calculating a metric based on the cumulative frequency difference and amplitude difference in the optimal warping path. The present invention exhibits significant advantages in terms of accuracy, time sensitivity, robustness, adaptability, comprehensive evaluation capability, quantitative analysis, and clinical application potential.

Description

Technical Field

[0001] The present invention relates to the technical field of human posture recognition, and in particular to a gait comparative analysis method for Parkinson's disease based on a time series characteristic curve. Background Art

[0002] Parkinson's disease (PD) is one of the most common movement disorders and seriously affects patients' quality of life. Among the typical symptoms of PD, patients have a variety of gait disorders, which often lead to falls, reduced quality of life, and shortened lifespan. The degree of gait abnormality is an important indicator of a patient's health status. By evaluating a patient's gait movements, doctors can understand the severity of the disease, track its progression, and monitor the effectiveness of treatment. Therefore, accurately identifying and evaluating gait is crucial for evaluating medication and treatment for PD patients.

[0003] Currently, there is a lack of comparative studies on the gait of Parkinson's patients before and after medication, which is crucial for evaluating the effectiveness of drug treatment and adjusting treatment plans; many time series comparison methods are currently used in various fields and do not have specificity; in the time series curve comparison method, only the difference between the two curves can be obtained, and the degree of change cannot be quantified. Summary of the Invention

[0004] In order to solve the various problems existing in the prior art, the purpose of the present invention is to provide a Parkinson's disease gait comparative analysis method based on time series characteristic curves, which shows significant advantages in accuracy, time sensitivity, robustness, adaptability, comprehensive evaluation ability, quantitative analysis and clinical application potential.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] A method for comparative analysis of Parkinson's disease gait based on time series characteristic curves, the method comprising the following steps:

[0007] acquiring video data including gait information, the gait information including gait information of the user before and after taking the drug;

[0008] Perform posture recognition on human bodies in video data and extract the location information of key body parts;

[0009] Based on the extracted key point information, gait features are calculated, and the changes of these features over time are constructed into a time series curve;

[0010] The extracted gait features are processed using digital filters;

[0011] A frequency-amplitude-enhanced dynamic time warping method is used to integrate local frequency and amplitude into the distance metric to quantify the differences in gait motion.

[0012] Get the curve difference distance;

[0013] The metrics are calculated based on the cumulative frequency differences and amplitude differences in the optimally regularized paths.

[0014] According to a Parkinson's disease gait comparative analysis method based on time series characteristic curves provided by the present invention, when performing human posture recognition, the OpenPose framework is used to detect skeletal key points from video data obtained in different time periods, wherein a BODY25 model is constructed using the OpenPose framework, and a series of K key point coordinates are obtained from the video data.

[0015] According to the present invention, a comparative analysis method for Parkinson's disease gait based on time series characteristic curves is provided, which defines gait features that can be directly calculated from the coordinates of key points in each frame to estimate the patient's cadence, including:

[0016] Leg length ratio difference R legs , is defined as the difference between the length ratios of the left and right legs, where leg length is determined using keypoints detected using the OpenPose framework and is measured as the distance between the hip and ankle keypoints of each leg, expressed as formula (1):

[0017]

[0018] Among them, P9 represents the coordinates of the right hip key point; P 11 represents the coordinates of the key point of the right ankle; P 12 Represents the coordinates of the left hip key point; P 14 Represents the coordinates of the left ankle keypoint.

[0019] Vertical angle of the body It is defined as the angle between the midpoints of the two ankles and a line passing through the neck and parallel to the y-axis, expressed as formula (2):

[0020]

[0021] Among them, x1 represents the horizontal coordinate of the neck key point; x 11,14 represents the horizontal coordinate of the midpoint between the right ankle key point and the left ankle key point; P1 represents the coordinate of the neck key point; P 11,14 represents the coordinates of the midpoints of the left and right ankles;

[0022] Horizontal angle of the ankle It is defined as the angle between the line between the two ankles and the x-axis, expressed as formula (3):

[0023]

[0024] Among them, the horizontal coordinate of the right ankle key point is expressed as x11 ;y 11 Indicates the vertical coordinate of the right ankle key point; x 14 Indicates the horizontal coordinate of the left ankle key point; y 14 Indicates the vertical coordinate of the left ankle keypoint.

[0025] According to the Parkinson's disease gait comparative analysis method based on time series characteristic curves provided by the present invention, the digital filter is a Savitzky-Golay filter.

[0026] According to a Parkinson's disease gait comparative analysis method based on time series characteristic curve provided by the present invention, when using Savitzky-Golay filter, let x k is the gait feature of the kth frame, y k is the smoothed data, and the smoothed curve is expressed as formula (4):

[0027]

[0028] in, is the smoothing value of the first frame data, H is the normalization factor, l is the half width of the smoothing window, x k+i is the data point of the k+i frame in the window, h i It is the smoothing coefficient obtained by fitting a polynomial using the least squares method, and is used to best represent the trend of the data within the window.

[0029] According to the present invention, a method for comparative analysis of Parkinson's disease gait based on time series characteristic curves is provided. The method uses a dynamic time warping method with frequency-amplitude enhancement to integrate local frequency and amplitude into a distance metric to quantify differences in gait motion, including:

[0030] Given two time series S1 and S2, with lengths of m and n respectively, represented as S1 = (u1, u2, ..., u m ) and S2=(v1,v2,…,v n );

[0031] u in S1 i and S2's v j A fixed-size window ω is selected from the neighborhood of each point in and a subsequence is defined: U i =(u i-ω / 2 ,…,u i ,…,u i+ω / 2 ) and (v j-ω / 2 ,…,v j ,…,v j+ω / 2 );

[0032] Computing local frequency and amplitude involves finding peaks in the windowed subsequence to calculate the average period and frequency;

[0033] where the amplitude is determined by calculating the difference between the average and minimum of these peaks, and calculating the local frequency and amplitude based on the local frequency and amplitude involves finding the peak in the windowed subsequence to calculate the average period and frequency;

[0034] According to the local frequency and amplitude, the difference between gait sequences is measured by the distance function, which is defined as formula (5):

[0035]

[0036] Among them, V j The local frequency and amplitude are f(v j ) and A(v j );f(u i ) and A(u i ) is U i The local frequency and amplitude of ; and λ1 and λ2 are weighting parameters that adjust the effect of frequency and amplitude on distance.

[0037] According to a comparative analysis method for Parkinson's disease gait based on time series characteristic curves provided by the present invention, in order to align two time series, an n×m matrix is constructed, which is a cumulative distance matrix calculated by dynamic programming, and the optimal aligned warping path is obtained by backtracking according to the following recursive function, which is expressed as formula (6):

[0038] D(i,j)=d(u i ,v j )+min(D(i-1,j),D(i,j-1),D(i-1,j-1)), (6)

[0039] According to a comparative analysis method for Parkinson's disease gait based on a time series characteristic curve provided by the present invention, the regularized path needs to meet boundary conditions, continuity conditions, and monotonicity conditions. Therefore, the starting point and end point of the normalized path can only be the two endpoints of the matrix diagonal, and the step length can only be two adjacent matrix elements. The regularized path is expressed as formula (7):

[0040] W=(w1,w2,...,w k ,...,w K ),max(m,n)≤K≤m+n-1 (7)

[0041] Among them, w k is the position (i, j) of the kth observation in the regularized path; then, by minimizing the cost of the regularized path, the DTW distance to the optimal aligned regularized path is obtained, which is expressed as formula (8):

[0042]

[0043] in, is the set of all possible regular paths, and K is the length of the regular path. The distance is divided by K to eliminate the influence of regular paths of different lengths.

[0044] According to the Parkinson's disease gait comparative analysis method based on time series characteristic curve provided by the present invention, the optimal matching path W found by DTW * For each pair of regular points (u i ,v j ), M is defined as formula (9):

[0045]

[0046] in, and Represents u i and v j For the qth feature of , a positive value of the gait comparison metric M indicates an improvement in gait motion.

[0047] According to the present invention, a method for comparative analysis of Parkinson's disease gait based on time series characteristic curves further performs:

[0048] Select multiple gait datasets with different characteristics and sources, which contain gait information of different groups of people;

[0049] For each gait dataset, it is divided into a training set and a test set, where the training set is used for the training process of the algorithm or model, and the test set is used to evaluate the performance of the algorithm or model on unseen data;

[0050] Conduct multiple repeated experiments and cross-validation to obtain the average performance of the algorithm model on different data sets, thereby more accurately evaluating its generalization ability;

[0051] Analyze and compare the evaluation results of the algorithm model on different data sets.

[0052] It can be seen that compared with the prior art, the method provided by the present invention has the following beneficial effects:

[0053] 1. The present invention can obtain the comparative differences in the gait curves of Parkinson's patients before and after medication and set a measurement parameter to quantify the gait changes through the cumulative frequency difference and amplitude difference in the best matching path, so as to accurately evaluate the gait disorders of PD patients before and after medication.

[0054] 2. The present invention designs a frequency-amplitude-sensitive distance measurement method to evaluate the improvement or deterioration of gait by combining local frequency and amplitude features.

[0055] 3. By extracting the position information of key body parts (such as joints) and constructing a timing curve, the present invention can more accurately capture the dynamic changes of gait and better reflect the subtle differences in gait than traditional overall profile analysis, especially the subtle changes under the influence of drugs.

[0056] 4. This method utilizes time-series characteristic curves to analyze how gait characteristics change over time, which is crucial for understanding the dynamics of drug effects on gait. Time series analysis allows for observing specific changes in gait at different time points, enabling a more comprehensive assessment of drug effects.

[0057] 5. The application of digital filter (Savitzky-Golay) helps to reduce the impact of noise and interference on gait feature extraction and improve the robustness of the data.

[0058] 6. The present invention adopts a frequency-amplitude-enhanced DTW method, which can effectively handle the scaling and offset problems of different gait sequences on the time axis, and can more accurately compare the gait differences of different individuals or the same individual under different conditions, even if these gait sequences are not completely aligned in time.

[0059] 7. This invention integrates local frequency and amplitude differences into the distance metric, accounting for not only gait speed variations but also gait amplitude variations. This comprehensive assessment method can more comprehensively reflect the complexity and diversity of gait, providing richer information for drug efficacy assessment.

[0060] 8. By calculating curve difference distances and metrics, this method can quantify gait differences into specific numerical values, facilitating statistical analysis and comparison. This helps researchers more objectively assess the effects of drugs on gait and the differences in effects between different drugs.

[0061] 9. By monitoring and analyzing the changes in patients' gait before and after taking medication, the present invention can provide doctors with important clinical information and guide the formulation and adjustment of treatment plans.

[0062] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 The present invention is a flowchart of an embodiment of a method for comparative analysis of Parkinson's disease gait based on a time series characteristic curve.

[0064] Figure 2 This is a schematic diagram of time series data on leg ratio differences in an embodiment of a Parkinson's disease gait comparative analysis method based on a time series characteristic curve of the present invention.

[0065] Figure 3This is a schematic diagram of vertical angle time series data in an embodiment of a Parkinson's disease gait comparative analysis method based on a time series characteristic curve of the present invention.

[0066] Figure 4 This is a schematic diagram of time series data on the horizontal angle of the ankle in an embodiment of a method for comparative analysis of Parkinson's disease gait based on a time series characteristic curve of the present invention.

[0067] Figure 5 This is a schematic diagram of an original image curve that has not been filtered by a Savitzky-Golay filter in an embodiment of a Parkinson's disease gait comparative analysis method based on a time series characteristic curve of the present invention.

[0068] Figure 6 It is a schematic diagram of an image curve processed by Savitzky-Golay filter in an embodiment of a method for comparative analysis of Parkinson's disease gait based on a time series characteristic curve of the present invention.

[0069] Figure 7 This is a schematic diagram of the amplitude of time series data on the difference in gait characteristics of a user before and after using a drug in an embodiment of a Parkinson's disease gait comparative analysis method based on a time series characteristic curve of the present invention.

[0070] Figure 8 It is a frequency diagram of time series data on the difference in gait characteristics of a user before and after using a drug in an embodiment of a Parkinson's disease gait comparative analysis method based on a time series characteristic curve of the present invention.

[0071] Figure 9 1 is a schematic diagram of the distance matrix of FAE-DTW in an embodiment of a method for comparative analysis of Parkinson's disease gait based on time series characteristic curves of the present invention.

[0072] Figure 10 1 is a schematic diagram of a distance matrix of a gait comparison metric M on a path in an embodiment of a method for comparative analysis of Parkinson's disease gait based on a time series characteristic curve of the present invention. DETAILED DESCRIPTION

[0073] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0074] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0075] See also Figures 1 to 10 This embodiment provides a method for comparative analysis of Parkinson's disease gait based on a time series characteristic curve, the method comprising the following steps:

[0076] Step S1, obtaining video data containing gait information, where the gait information includes gait information of a user before and after taking a drug;

[0077] Step S2, performing posture recognition on the human body in the video data to extract the position information of key body parts;

[0078] Step S3, calculating gait features based on the extracted key point information, and constructing a time series curve of the changes of these features over time;

[0079] Step S4, performing data processing on the extracted gait features using a digital filter;

[0080] Step S5, using a dynamic time warping method with frequency-amplitude enhancement to integrate local frequency and amplitude into the distance metric to quantify the difference in gait motion;

[0081] Step S6, obtaining the curve difference distance;

[0082] Step S7: Calculate a metric based on the cumulative frequency difference and amplitude difference in the optimal regularized path.

[0083] In the above step S2, when performing human posture recognition, the OpenPose framework is used to detect skeletal key points from the walking videos of Parkinson's disease patients captured at different time periods before and after drug treatment.

[0084] OpenPose is an open-source library based on convolutional neural networks and supervised learning. It can detect key points on the face, torso, and limbs, and is suitable for both single-person and multi-person pose recognition scenarios. The OpenPose model is trained on a large-scale dataset of human skeletal key points to ensure the accuracy of key point detection. This example uses the OpenPose framework to construct a BODY25 model, which extracts a series of 25 key point coordinates from video data.

[0085] Because it is challenging to directly obtain the motion characteristics of cadence from the skeleton key points of the BODY25 model, this embodiment defines three gait features that can be directly calculated from the coordinates of the key points in each frame to estimate the patient's cadence. These gait features are defined as follows:

[0086] 1. Leg length ratio difference R legs : Leg length ratio difference is defined as the difference between the length ratios of the left and right legs, which is calculated by calculating the ratio of the lengths of the left and right legs and then finding the difference between these ratios. Leg length is determined using the key points detected using the OpenPose framework. Specifically, leg length is measured as the distance between the hip and ankle key points of each leg, expressed as formula (1):

[0087]

[0088] Among them, P9 represents the coordinates of the right hip key point; P 11 represents the coordinates of the key point of the right ankle; P 12 Represents the coordinates of the left hip key point; P 14 Represents the coordinates of the left ankle keypoint.

[0089] 2. Vertical angle of the body It is defined as the angle between the midpoints of the two ankles and a line passing through the neck and parallel to the y-axis, expressed as formula (2):

[0090]

[0091] Among them, x1 represents the horizontal coordinate of the neck key point; x 11,14 represents the horizontal coordinate of the midpoint between the right ankle key point and the left ankle key point; P1 represents the coordinate of the neck key point; P 11,14 represents the coordinates of the midpoints of the left and right ankles;

[0092] 3. Horizontal angle of the ankle It is defined as the angle between the line between the two ankles and the x-axis, expressed as formula (3):

[0093]

[0094] Among them, the horizontal coordinate of the right ankle key point is expressed as x 11 ;y 11 Indicates the vertical coordinate of the right ankle key point; x 14 Indicates the horizontal coordinate of the left ankle key point; y 14 Indicates the vertical coordinate of the left ankle keypoint.

[0095] In the above step S4, the digital filter selected in this embodiment is the Savitzky-Golay filter. In this embodiment, a video of a Parkinson's patient walking forward is selected for analysis. Figure 5 As shown, the time series data exhibits significant fluctuations, which increases the computational load. Therefore, data smoothing is necessary to remove noise while preserving the signal's key characteristics. The Savitzky-Golay filter is a digital filter designed to smooth curves, enhancing the accuracy of the data without changing the original signal's trend or width.

[0096] In this embodiment, when using the Savitzky-Golay filter, let x k is the gait feature of the kth frame, y k is the smoothed data, and the smoothed curve is expressed as formula (4):

[0097]

[0098] in, is the smoothing value of the first frame data, H is the normalization factor, l is the half width of the smoothing window, x k+i is the data point of the k+i frame in the window, h i It is the smoothing coefficient obtained by fitting a polynomial using the least squares method, and is used to best represent the trend of the data within the window.

[0099] In the above-mentioned step S5, conventional dynamic time warping (DTW) algorithms mainly focus on dynamically aligning two time series data points to find the shortest distance and the optimal matching path. However, these algorithms generally fail to take into account potential filtering features such as frequency and amplitude. In the case of a single PD patient before and after drug treatment, it is also aimed to quantify the changes in the gait characteristic curve to identify gait improvement and deterioration. To address this limitation, the present embodiment introduces a novel frequency-amplitude enhanced dynamic time warping (FAE-DTW) method for quantifying differences in gait motion. By integrating local frequency and amplitude into the distance metric, FAE-DTW shows superior performance in aligning time series with significant harmonic characteristics.

[0100] Specifically, given two time series S1 and S2, with lengths of m and n respectively, expressed as S1 = (u1, u2, ..., u m ) and S2=(v1,v2,…,v n ); u in S1 i and S2's v j A fixed-size window ω is selected from the neighborhood of each point in and a subsequence is defined: U i =(u i-ω / 2 ,…,u i,…,u i+ω / 2 ) and (v j-ω / 2 ,…,v j ,…,v j+ω / 2 ). Calculating the local frequency and amplitude involves finding peaks in the windowed subsequence to calculate the average period and frequency. The amplitude is determined by calculating the difference between the average and minimum of these peaks. Calculating the local frequency and amplitude based on the local frequency and amplitude involves finding peaks in the windowed subsequence to calculate the average period and frequency. Based on the local frequency and amplitude, the difference between gait sequences is measured by a distance function, which is defined as formula (5):

[0101]

[0102] Among them, V j The local frequency and amplitude are f(v j ) and A(v j );f(u i ) and A(u i ) is U i The local frequency and amplitude of ; and λ1 and λ2 are weighting parameters that adjust the effect of frequency and amplitude on distance.

[0103] In this embodiment, in order to align two time series, an n×m matrix is constructed, which is the cumulative distance matrix calculated by dynamic programming, and the optimal aligned warping path is obtained by backtracking according to the following recursive function, which is expressed as formula (6):

[0104] D(i,j)=d(u i ,v j )+min(D(i-1,j),D(i,j-1),D(i-1,j-1)), (6)

[0105] In this embodiment, the regularized path needs to meet the boundary conditions, continuity conditions, and monotonicity conditions. Therefore, the starting point and end point of the normalized path can only be the two endpoints of the matrix diagonal, and the step size can only be two adjacent matrix elements. The regularized path is expressed as formula (7):

[0106] W=(w1,w2,...,w k ,...,w K ),max(m,n)≤K≤m+n-1 (7)

[0107] Among them, w k is the position (i, j) of the kth observation in the regularized path; then, by minimizing the cost of the regularized path, the DTW distance to the optimal aligned regularized path is obtained, which is expressed as formula (8):

[0108]

[0109] in, is the set of all possible regular paths, and K is the length of the regular path. The distance is divided by K to eliminate the influence of regular paths of different lengths.

[0110] Specifically, since traditional dynamic time warping (DTW) can only provide gait differences with positive distance values, it cannot indicate whether the motor ability has improved after drug treatment. In order to further evaluate the improvement or deterioration of gait movement, this embodiment designs a novel metric that can be calculated based on the cumulative frequency difference and amplitude difference in the optimal warping path. This method not only measures the similarity between gait sequences, but also indicates whether the gait is improving through positive and negative values. The optimal matching path W found by DTW is * For each pair of regular points (u i ,v j ), M is defined as follows:

[0111]

[0112] in, and Represents u i and v j For the qth feature of , a positive value of the gait comparison metric M indicates an improvement in gait motion.

[0113] In this embodiment, the following is also performed:

[0114] Select multiple gait datasets with different characteristics and sources, which contain gait information of different groups of people;

[0115] For each gait dataset, it is divided into a training set and a test set, where the training set is used for the training process of the algorithm or model, and the test set is used to evaluate the performance of the algorithm or model on unseen data;

[0116] Conduct multiple repeated experiments and cross-validation to obtain the average performance of the algorithm model on different data sets, thereby more accurately evaluating its generalization ability;

[0117] Analyze and compare the evaluation results of the algorithm model on different data sets.

[0118] In summary, the present invention can compare the gait curves of Parkinson's patients before and after medication and set a metric parameter to quantify gait changes by using the cumulative frequency difference and amplitude difference in the best matching path, thereby accurately assessing gait impairment in PD patients before and after medication. This invention designs a frequency-amplitude-sensitive distance measurement method, combining local frequency and amplitude characteristics to assess gait improvement or deterioration.

[0119] Furthermore, the present invention can more accurately capture the dynamic changes of gait by extracting the position information of key body parts and constructing it into a timing curve. It can better reflect the subtle differences in gait than traditional overall profile analysis, especially the subtle changes under the influence of drugs.

[0120] Furthermore, the present invention utilizes time series characteristic curves to analyze how gait characteristics change over time, which is crucial for understanding the dynamics of drug effects on gait. Time series analysis allows for observing specific changes in gait at different time points, enabling a more comprehensive assessment of drug effects. The application of a digital filter (Savitzky-Golay) helps reduce the impact of noise and interference on gait feature extraction, improving data robustness.

[0121] Furthermore, the present invention adopts a frequency-amplitude-enhanced DTW method, which can effectively handle the scaling and offset problems of different gait sequences on the time axis, and can more accurately compare the gait differences of different individuals or the same individual under different conditions, even if these gait sequences are not completely aligned in time.

[0122] Furthermore, the present invention integrates local frequency and amplitude differences into the distance metric, accounting for not only gait speed variations but also gait amplitude variations. This comprehensive assessment method can more comprehensively reflect the complexity and diversity of gait, providing richer information for drug efficacy assessment.

[0123] Furthermore, by calculating curve difference distances and metrics, the present invention can quantify gait differences into specific numerical values, facilitating statistical analysis and comparison. This helps researchers more objectively assess the effects of drugs on gait, as well as the differences in effects between different drugs.

[0124] Furthermore, the present invention can provide doctors with important clinical information and guide the formulation and adjustment of treatment plans by monitoring and analyzing the changes in patients' gait before and after taking drugs.

[0125] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] The above embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and replacements made by technicians in this field on the basis of the present invention fall within the scope of protection required by the present invention.

Claims

1. A comparative analysis method for Parkinson's disease gait based on time series characteristic curves, characterized in that: The method comprises the following steps: acquiring video data including gait information, the gait information including gait information of the user before and after taking the drug; Perform posture recognition on human bodies in video data and extract the location information of key body parts; Based on the extracted key point information, gait features are calculated, and the changes of these features over time are constructed into a time series curve; The extracted gait features are processed using digital filters; The dynamic time warping method with frequency-amplitude enhancement is used to integrate local frequency and amplitude into the distance metric to quantify the difference in gait motion. Specifically, given two time series and , respectively, with lengths of and , expressed as and ;exist and A fixed size is selected from the neighborhood of each point in , and define the subsequence: and ;Calculating local frequency and amplitude involves finding peaks in the windowed subsequence to calculate the average period and frequency; where the amplitude is determined by calculating the difference between the average and minimum of these peaks, and calculating the local frequency and amplitude based on the local frequency and amplitude involves finding the peak in the windowed subsequence to calculate the average period and frequency; The difference between gait sequences is measured by the distance function based on the local frequency and amplitude. The distance function is defined as formula (5): (5) in, The local frequency and amplitude are and ; and yes The local frequency and amplitude of and It is a weighting parameter that adjusts the effect of frequency and amplitude on distance; Get the curve difference distance; The metrics are calculated based on the cumulative frequency differences and amplitude differences in the optimally regularized paths.

2. The method according to claim 1, wherein: When performing human posture recognition, the OpenPose framework is used to detect skeletal key points from video data obtained at different time periods. The BODY25 model is constructed using the OpenPose framework to obtain a series of K key point coordinates from the video data.

3. The method according to claim 1, wherein: Define gait features that can be calculated directly from the keypoint coordinates of each frame to estimate the patient's cadence, including: Leg length ratio difference , is defined as the difference between the length ratios of the left and right legs, where leg length is determined using keypoints detected using the OpenPose framework and is measured as the distance between the hip and ankle keypoints of each leg, expressed as formula (1): (1) in, Represents the coordinates of the right hip key point; Represents the coordinates of the key points of the right ankle; Represents the coordinates of the left hip key point; Represents the coordinates of the key points of the left ankle; Vertical angle of the body , is defined as the angle between the midpoints of the two ankles and a line passing through the neck and parallel to the y-axis, expressed as formula (2): (2) in, Indicates the horizontal coordinates of the neck key points; Represents the horizontal coordinate of the midpoint between the right ankle keypoint and the left ankle keypoint; Represents the coordinates of the key points of the neck; represents the coordinates of the midpoints of the left and right ankles; Horizontal angle of the ankle , is defined as the angle formed by the line between the two ankles and the x-axis, expressed as formula (3): (3) Among them, the horizontal coordinate of the right ankle key point is expressed as ; Indicates the vertical coordinate of the right ankle key point; Represents the horizontal coordinate of the left ankle key point; Indicates the vertical coordinate of the left ankle keypoint.

4. The method according to claim 1, wherein: The digital filter is a Savitzky-Golay filter.

5. The method according to claim 4, characterized in that: When using the Savitzky-Golay filter, assume is the gait feature of the kth frame, is the smoothed data, and the smoothed curve is expressed as formula (4): ,(4) in, is the smoothed value of the frame data, is the normalization factor, is the half-width of the smoothing window, It is inside the window The data points of the frame, It is the smoothing coefficient obtained by fitting a polynomial using the least squares method, and is used to best represent the trend of the data within the window.

6. The method according to claim 1, wherein: To align two time series, construct a Matrix, which is the cumulative distance matrix calculated by dynamic programming, and the optimal aligned warping path is obtained by backtracking according to the following recursive function, expressed as formula (6): (6) 7. The method according to claim 6, characterized in that: The regularized path needs to meet the boundary conditions, continuity conditions and monotonicity conditions. Therefore, the starting point and end point of the normalized path can only be the two endpoints of the matrix diagonal, and the step size can only be two adjacent matrix elements. The regularized path is expressed as formula (7): (7) in, It is the first The location of the observation Then, by minimizing the cost of the regularized path, the DTW distance with the optimal aligned regularized path is obtained, which is expressed as formula (8): ,(8) in, is the set of all possible regular paths, and is the length of the regular path, and the distance is obtained by dividing To eliminate the influence of regular paths of different lengths.

8. The method according to claim 7, wherein: The best matching path found in DTW For each pair of regular points , The definition is expressed as formula (9): (9) in, and Respectively and No. Features, gait comparison metrics Positive values of indicate improvement in gait motion.

9. The method according to any one of claims 1 to 8, characterized in that Also execute: Select multiple gait datasets with different characteristics and sources, which contain gait information of different groups of people; For each gait dataset, it is divided into a training set and a test set, where the training set is used for the training process of the algorithm or model, and the test set is used to evaluate the performance of the algorithm or model on unseen data; Conduct multiple repeated experiments and cross-validation to obtain the average performance of the algorithm model on different data sets, thereby more accurately evaluating its generalization ability; Analyze and compare the evaluation results of the algorithm model on different data sets.

Images