An analysis method based on leg training data and a training chair

Abstract

The present invention discloses an analysis method and a training chair based on leg training data, and relates to the technical field of leg training. The method comprises: obtaining change characteristics from the multidimensional dynamic data through a time series analysis method; establishing a dynamic model through the change characteristics, and predicting the natural movement pattern during leg training according to the dynamic model; simulating the predicted natural movement pattern, identifying movement errors and injury risks, and adjusting the natural movement pattern. The present invention generates a personalized movement pattern by analyzing the user's physical characteristics, and integrates it with the natural movement pattern, so that the training can be more in line with personal needs, and through simulation and real-time monitoring, it can timely discover and correct movement errors, set reasonable joint angles, speeds and energy consumption thresholds, thereby avoiding injuries caused by improper movements and ensuring safety during training.

Description

Technical Field

[0001] The invention relates to the technical field of leg training, and in particular to an analysis method based on leg training data and a training chair. Background Art

[0002] With the improvement of health awareness and the advancement of science and technology, exercise and fitness have become an indispensable part of modern life. Especially in terms of improving physical function, preventing diseases, and improving the quality of life, leg training has received widespread attention due to its important impact on cardiopulmonary function, muscle strength and balance ability. However, traditional leg training often relies on personal experience or coaching guidance, lacks scientificity and personalization, which to some extent limits the optimization of training effects.

[0003] The Chinese invention patent with publication number CN115981478A discloses a digital virtual human leg training device, including an information storage device, a video playback device, a character sampling device, a secondary collection device, a digital virtual human device and a data comparison processor, wherein the information storage device, the video playback device, the character sampling device, the secondary collection device, the digital virtual human device and the data comparison processor are connected using a data cable. The digital virtual human leg training device processes an existing video, creates key points at the leg connection positions of the characters in the video, and records the motion trajectories of the key points when the characters in the video are active, and then matches and imitates the leg key points and motion trajectories of the digital virtual human model, so that the leg movements of the digital virtual human are trained, achieving the effect of training the legs of the digital virtual human to make the digital virtual human more coordinated.

[0004] Most of the above-mentioned and similar leg training devices obtain static coordinate points and motion trajectories through analysis of image and video data. However, in the actual training process, since the training movements are highly dynamic and continuous, involving a variety of subtle changes and transitions, the leg training devices that use image and video data analysis cannot fully capture the subtle changes and transitions in dynamic movements, which will cause stiffness and unnaturalness when performing complex or fast movements. Summary of the invention

[0005] The purpose of the present invention is to provide an analysis method based on leg training data and a training chair to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: an analysis method based on leg training data, comprising:

[0007] Extracting change features: acquiring multidimensional dynamic data during leg training, and acquiring change features from the multidimensional dynamic data by using a time series analysis method;

[0008] Establishing a dynamic model: establishing a dynamic model based on the change characteristics, and predicting the natural movement pattern during leg training based on the dynamic model, specifically:

[0009] Divide the user's legs into a plurality of rigid body segments, and obtain the kinetic energy and potential energy corresponding to each rigid body segment, and simultaneously obtain the total kinetic energy and total potential energy during leg training according to the kinetic energy and potential energy corresponding to each rigid body segment;

[0010] According to the total kinetic energy and total potential energy during the leg training, the size of the Lagrangian quantity is obtained to determine the joint angle and joint torque during the leg training;

[0011] According to the maximum and minimum bending angles of the knee joint, determine the range of motion of the knee and hip joints during leg training;

[0012] Adjusting the movement pattern: simulating the predicted natural movement pattern, identifying movement errors and injury risks, and adjusting the natural movement pattern.

[0013] Furthermore, obtaining change characteristics from the multi-dimensional dynamic data includes:

[0014] Acquiring time series data: acquiring multidimensional dynamic data during leg training through sensors, preprocessing the multidimensional dynamic data, and sorting the preprocessed multidimensional dynamic data in chronological order to acquire the time series data;

[0015] Obtaining a fused feature vector: according to the time series data, obtaining a hidden state vector through a long short-term memory network model, obtaining a convolution vector through a convolutional neural network model, and simultaneously fusing the hidden state vector and the convolution vector to obtain the fused feature vector.

[0016] Furthermore, obtaining the fused feature vector includes:

[0017] M1: Obtain hidden state vector: Use the time series data as the input of the long short-term memory network model and output the hidden state vector, specifically:

[0018] ;

[0019] in: is the hidden state vector output at time step t, is the leg training data information output at time step t, is the memory unit corresponding to the leg motion state at time step t, is the weight matrix, is the hidden state vector output at time step t-1, is the time series data input at time step t, is the bias vector, is the Sigmoid activation function;

[0020] M2: Obtain convolution vector: Use the time series data as the input of the convolutional neural network model and output the convolution vector, specifically:

[0021] ;

[0022] in: is the convolution vector, is the activation function, is the convolution kernel, For time series data, is the bias term;

[0023] M3: Obtain a splicing vector: fuse the hidden state vector and the convolution vector to form a splicing vector, and normalize the splicing vector, specifically:

[0024] ;

[0025] in: is the normalized concatenation vector, is the concatenation vector, is the minimum value in the concatenated vector, is the maximum value in the concatenated vector;

[0026] M4: Obtaining a fused feature vector: extracting features from the normalized concatenated vector, and the extracted features are the fused feature vector.

[0027] Furthermore, when performing leg training, the calculation formulas for the total kinetic energy and total potential energy are specifically:

[0028] ;

[0029] in: is the total kinetic energy during leg training, is the total potential energy during leg training, is the mass of the ith rigid body segment, is the acceleration due to gravity, is the position of the ith rigid body segment, is the moment of inertia of the ith rigid body segment, for The inversion of is the angular velocity of the ith rigid body segment, is the velocity of the ith rigid body segment.

[0030] Furthermore, the natural movement pattern is adjusted, including:

[0031] Determine personalized exercise mode: through physical characteristics, obtain the relationship between muscle strength and exercise ability, the relationship between height and step length, and the relationship between weight and exercise energy consumption, and determine the personalized exercise mode;

[0032] Obtaining a fused motion pattern: fusing the personalized motion pattern with the natural motion pattern to obtain the fused motion pattern, specifically:

[0033] ;

[0034] in: is the motion position after fusion at time t, is the joint angle after fusion at time t, is the weight coefficient, is the motion position in the personalized motion mode at time t, is the joint angle in the personalized motion mode at time t, is the motion position in the natural motion mode at time t, is the joint angle in the natural motion mode at time t;

[0035] Identifying movement errors and injury risks: simulating the fused movement pattern to identify movement errors and injury risks;

[0036] Adjusting weight coefficients: adjusting the weight coefficients in the fused motion model according to the motion error and injury risk.

[0037] Furthermore, in the process of determining the motion error and injury risk, the joint angle threshold, speed threshold and energy consumption threshold are set, and the fusion joint angle, fusion speed and fusion energy consumption to be adjusted are determined by comparing the fusion joint angle with the joint angle threshold, the fusion speed with the speed threshold, and the fusion speed with the energy consumption threshold, specifically:

[0038] When the fusion joint angle exceeds the joint angle threshold range, the fusion joint angle needs to be adjusted, otherwise, it does not need to be adjusted;

[0039] When the fusion speed exceeds the speed threshold range, the fusion speed needs to be adjusted, otherwise, no adjustment is required;

[0040] When the fusion energy consumption exceeds the energy consumption threshold range, the fusion energy consumption needs to be adjusted; otherwise, no adjustment is required.

[0041] Furthermore, adjusting the weight coefficient in the fusion motion mode includes:

[0042] N1: Based on the motion errors and injury risks, an objective function is established, specifically:

[0043] ;

[0044] in: is the objective function, is the weight of the motion error, Scoring for motion errors, is the weight of the risk of harm, To score the risk of harm;

[0045] N2: Obtain the gradient of the weight coefficient through the objective function, specifically:

[0046] ;

[0047] in: is the gradient of the objective function with respect to the weight coefficient, Add the weight factor The objective function after is the objective function, is a positive number;

[0048] N3: According to the gradient size of the weight coefficient, the size of the weight coefficient is adjusted by an adjustment formula, and the adjustment formula is specifically:

[0049] ;

[0050] in: is the weight coefficient at the k+1th iteration, is the weight coefficient at the kth iteration, is the learning rate, is the gradient of the objective function with respect to the weight coefficient at the kth iteration.

[0051] A training chair based on leg training data, comprising:

[0052] Analysis module: obtaining multi-dimensional dynamic data during leg training, and establishing a dynamic model based on the multi-dimensional dynamic data, obtaining a natural movement pattern during leg training, and obtaining a personalized movement pattern based on body characteristics, and fusing the natural movement pattern with the personalized movement pattern to determine a fused movement pattern during leg training;

[0053] Movement module: setting the operation parameters of the training chair according to the joint angles and movement positions corresponding to the fusion movement mode;

[0054] Control module: exchanges information with the analysis module and the motion module.

[0055] Furthermore, the analysis module includes:

[0056] Extraction unit: obtaining a hidden state vector and a convolution vector through the multi-dimensional dynamic data, and fusing the hidden state vector and the convolution vector to obtain a fused feature vector;

[0057] Model unit: determining the natural movement pattern during leg training through the fused feature vector, and determining the personalized movement pattern during leg training according to the body characteristics, and fusing the natural movement pattern with the personalized movement pattern to obtain the fused movement pattern during leg training;

[0058] An adjustment unit simulates the fused motion pattern, determines motion errors and injury risks during the simulation, and adjusts the weight coefficients in the fused motion pattern according to the motion errors and injury risks.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] First, the present invention generates personalized movement patterns by analyzing the user's physical characteristics and integrates them with natural movement patterns, so that training can better meet personal needs. Through simulation and real-time monitoring, movement errors can be discovered and corrected in a timely manner, and reasonable joint angles, speeds and energy consumption thresholds can be set, thereby avoiding injuries caused by improper movements and ensuring safety during training.

[0061] Second, the present invention can provide a comprehensive training evaluation by collecting and analyzing the multi-dimensional dynamic data of the user, and adjust and set the training mode and parameters based on biomechanics and sports science, further ensuring the scientificity and effectiveness of the training;

[0062] Thirdly, the present invention dynamically adjusts the training plan according to the real-time data and feedback of the user, and can also monitor the user's exercise status in real time and provide instant feedback, thereby ensuring the maximization of the training effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 Schematic diagram of the analytical method of the present invention;

[0064] Figure 2 It is a schematic diagram of the process of predicting leg training through natural movement patterns according to the present invention;

[0065] Figure 3 It is a schematic diagram of the process of obtaining change characteristics through multi-dimensional dynamic data of the present invention;

[0066] Figure 4 is a system block diagram of the training chair of the present invention;

[0067] Figure 5 It is a schematic structural diagram of the training chair of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] With the development of science and technology and the improvement of people's health awareness, personalized and scientific sports training has become an important trend in the field of modern fitness and rehabilitation. In the actual training process, since the training movements are highly dynamic and continuous, there are many subtle changes and transitions involved. Therefore, the leg training equipment analyzed by image video data cannot fully capture the subtle changes and transitions in dynamic movements, which will cause stiffness and unnatural phenomena when performing complex or fast movements. The technical solution of the present application obtains the multi-dimensional dynamic data of the user, extracts the comprehensive feature vector from it, generates the corresponding natural movement pattern and personalized movement pattern, and fuses the fused movement pattern to obtain the corresponding movement position and joint angle, so that the training can be more in line with personal needs, and through simulation and real-time monitoring, it can timely discover and correct movement errors, set reasonable joint angles, speeds and energy consumption thresholds, further avoid injuries caused by improper movements, and ensure safety during training.

[0070] Example 1

[0071] refer to Figure 1-Figure 3 This embodiment provides an analysis method based on leg training data, which includes the following steps:

[0072] Step S1: extracting change features. That is, obtaining multi-dimensional dynamic data of the user during leg training, the multi-dimensional dynamic data including but not limited to video data, accelerometer data, gyroscope data and electromyography data. At the same time, by using a time series analysis method, extracting corresponding change features from the obtained multi-dimensional dynamic data. Specifically as follows:

[0073] Step S1.1: Obtaining time series data. During the user's leg training, multiple sensors are set to collect multi-dimensional dynamic data of the user's leg training. Specifically:

[0074] Image video data: Use a camera to record a video of the user’s leg training to capture the user’s movement posture and motion trajectory.

[0075] Accelerometer data: The accelerometer installed on the leg records the acceleration changes of the user's leg during leg training.

[0076] Gyroscope data: The gyroscope installed on the leg records the angle changes of the user's legs during leg training.

[0077] Electromyography data: Electrodes attached to the muscles record changes in the electrical signals of the muscles during leg training.

[0078] In this embodiment, the acquired multidimensional dynamic data is preprocessed, that is, denoising, filtering, time synchronization and space calibration are performed. Specifically, wavelet denoising or Kalman filtering removes noise in the multidimensional dynamic data, retains useful signal frequency bands through bandpass filters, and removes high-frequency noise and low-frequency drift. It is worth noting that during the acquisition of the multidimensional dynamic data, the time synchronization of its data acquisition needs to be maintained, and the position of the sensor needs to be calibrated to ensure the consistency and accuracy of the data. At the same time, after the multidimensional dynamic data is acquired, it needs to be arranged in order according to the time sequence to form time series data.

[0079] In the specific implementation process, each piece of data contains 100 time steps, and each time step contains 4 features, namely acceleration, angular velocity, joint angle and electromyography signal.

[0080] Step S1.2: Obtain fused feature vector. That is, establish a long short-term memory network model and a convolutional neural network model, and according to the time series data formed in step S1.1, obtain the hidden state vector through the long short-term memory network model, and obtain the convolution vector through the convolutional neural network model. At the same time, fuse the hidden state vector and the convolution vector to obtain a splicing vector. At the same time, normalize the splicing vector to ensure that the hidden state vector and the convolution vector are in the same dimension, and determine the fused feature vector from the splicing vector through the feature selection method. The details are as follows:

[0081] Step M1: Obtain hidden state vector. That is, establish a long short-term memory network model, and use the time series data formed in step S1.1 as the input of the long short-term memory network model, and output the hidden state vector. The hidden state vector is specifically:

[0082] ;

[0083] in: is the hidden state vector output at time step t, is the leg training data information output at time step t, is the memory unit corresponding to the leg motion state at time step t, is the weight matrix, is the hidden state vector output at time step t-1, is the time series data input at time step t, is the bias vector, is the Sigmoid activation function;

[0084] Step M2: Obtain the convolution vector. That is, establish a convolution neural network model, and use the time series data formed in step S1.1 as the input of the convolution neural network model, and output the convolution vector. The convolution vector is specifically:

[0085] ;

[0086] in: is the convolution vector, is the activation function, is the convolution kernel, For time series data, Bias term;

[0087] Step M3: Get the concatenated vector. That is, the hidden state vector obtained in step M1 Convolution vector obtained in step M2 Fusion to form a splicing vector It is worth noting that to ensure that the hidden state vector and the convolution vector can be in the same dimension, so in this embodiment, the concatenation vector Normalization is performed, specifically:

[0088] ;

[0089] in: is the normalized concatenation vector, is the concatenation vector, is the minimum value in the concatenated vector, is the maximum value in the concatenated vector.

[0090] Step M4: Obtain fusion feature vector. That is, the normalized concatenated vector obtained from step M3 Specifically, in this embodiment, the principal component analysis algorithm or the recursive feature elimination algorithm can be used to extract features from the normalized splicing vector Extract features from .

[0091] In the process of specific implementation, the following data are set:

[0092] Acceleration: [0.5, 0.8, 1.2, 0.9, 0.6].

[0093] Angular velocity: [0.2, 0.4, 0.6, 0.5, 0.3].

[0094] Joint angles: [30°,40°,50°,45°,35°].

[0095] EMG signal: [0.1, 0.2, 0.3, 0.25, 0.15].

[0096] According to the above data, the following fusion feature vector can be obtained, specifically:

[0097] [0.8,0.25,0.4,0.15,40,5.77,0.2,0.07].

[0098] Step S2: Establish a dynamic model. That is, establish a dynamic model based on the dynamic model, and use the fused feature vector obtained in step M4 as the input of the dynamic model, and output the natural movement pattern during leg training. In other words, the natural movement pattern during leg training is simulated and predicted through the dynamic model. The details are as follows:

[0099] Step S2.1: Divide the user's legs into multiple rigid body segments (such as thighs and calves), and determine the mass and moment of inertia of each rigid body segment, so as to obtain the total kinetic energy and total potential energy of the user during leg training, specifically:

[0100] ;

[0101] in: Total kinetic energy during leg training, is the total potential energy during leg training, is the mass of the ith rigid body segment, is the acceleration due to gravity, is the position of the ith rigid body segment, is the moment of inertia of the ith rigid body segment, for The inversion of is the angular velocity of the ith rigid body segment, is the velocity of the i-th rigid body segment.

[0102] Step S2.2: According to the total kinetic energy and total potential energy obtained in step S2.1, the size of the Lagrangian is determined, and according to the size of the Lagrangian, the generalized coordinates and generalized forces are determined, specifically:

[0103] ;

[0104] in: is the Lagrangian, is the total kinetic energy during leg training, is the total potential energy during leg training, are generalized coordinates, It is a general force.

[0105] In this embodiment, the generalized coordinates is the joint angle during leg training, generalized force This is the joint torque during leg training.

[0106] Step S2.3: The maximum bending angle and the minimum bending angle of the knee joint are used to determine the range of motion of the knee joint and the hip joint during the user's leg training.

[0107] In the specific implementation process, the user's legs are divided into two rigid body segments, the thigh and the calf. That is, the mass and moment of inertia data corresponding to the two rigid body segments are:

[0108] Thigh: 5kg, 0.1kg.m 2 .

[0109] Calf: 3kg, 0.05kg.m 2 .

[0110] Furthermore, the following data can also be obtained:

[0111] Initial position: The position of the first rigid body segment is [0,0,0] T , the position of the second rigid body segment is [0,0.5,0] T .

[0112] Initial velocity: The velocity of the first rigid body segment is [0,0,0] T , the velocity of the second rigid body segment is [0,0,0] T .

[0113] Initial angle: The first joint angle is 0°, and the second joint angle is 0°.

[0114] Initial angular velocity: The first angular velocity is 0°, and the second angular velocity is 0°.

[0115] The maximum flexion angle of the knee joint is: 120°.

[0116] The minimum flexion angle of the knee joint is: 0°.

[0117] Furthermore, at the 10th time step t, the following data can be obtained:

[0118] Acceleration: [0.5, 0.8, 1.2, 0.9, 0.6].

[0119] Angular velocity: [0.2, 0.4, 0.6, 0.5, 0.3].

[0120] Joint angles: [30°,40°,50°,45°,35°].

[0121] EMG signal: [0.1, 0.2, 0.3, 0.25, 0.15].

[0122] According to the above data, we can know that:

[0123] The kinetic energy and potential energy of the thigh are 67.98 and 0 respectively.

[0124] The kinetic energy and potential energy of the calf are 1.89875 and 14.715 respectively.

[0125] That is to say, when the user performs leg training at the 10th time step t, his total kinetic energy is 69.87875 and his total potential energy is 14.715.

[0126] Step S3: Adjust the movement mode. According to the user's physical characteristics (such as weight, height and muscle strength), determine the corresponding personalized movement mode, and fuse the obtained personalized movement mode with the natural movement mode predicted in step S2 to obtain a fused movement mode. Simulate the fused movement mode to identify movement errors and injury risks, and adjust the fused movement mode. The details are as follows:

[0127] Step S3.1: Determine the personalized exercise mode. That is, according to the user's weight, height and muscle strength, determine the relationship between muscle strength and exercise ability, the relationship between height and step length, and the relationship between weight and exercise energy consumption. At the same time, according to the relationship between muscle strength and exercise ability, the relationship between height and step length, and the relationship between weight and exercise energy consumption, obtain the user's personalized exercise trajectory.

[0128] In this embodiment, the relationship between muscle strength and athletic ability is specifically as follows:

[0129] ;

[0130] in: To build muscle strength during leg training, is the muscle strength constant per unit body weight, is the user's weight.

[0131] Furthermore, the relationship between height and stride length is as follows:

[0132] ;

[0133] in: For leg training, is the step length constant per unit height, is the user's height.

[0134] To further explain, the relationship between body weight and exercise energy consumption is as follows:

[0135] ;

[0136] in: For the energy consumption during leg training, is the energy consumption constant per unit weight and distance, is the user's weight, is the user's walking distance.

[0137] Step S3.2: Obtaining a fused motion pattern. That is, the personalized motion pattern determined in step S3.1 and the natural motion pattern predicted in step S2 are fused to obtain a fused motion pattern. Specifically:

[0138]

[0139] in: is the motion position after fusion at time t, is the joint angle after fusion at time t, is the weight coefficient, is the motion position in the personalized motion mode at time t, is the joint angle in the personalized motion mode at time t, is the motion position in the natural motion mode at time t, is the joint angle in the natural motion mode at time t.

[0140] Step S3.3: Identify movement errors and injury risks. That is, simulate the fused movement pattern obtained after fusion in step S3.2 to determine movement errors and injury risks. Specifically, according to the user's physiological characteristics, movement type and safety standards, the joint angle threshold, speed threshold and energy consumption threshold are set respectively.

[0141] Furthermore, according to the set joint angle threshold, it is determined whether the fused joint angle in the fused motion mode obtained after fusion exceeds the threshold range. In other words, when the fused joint angle exceeds the joint angle threshold range, the fused joint angle needs to be adjusted. Otherwise, the fused joint angle does not need to be adjusted.

[0142] Further, according to the set speed threshold, specifically including the speed threshold and the acceleration threshold, it is determined whether the fusion speed and the fusion acceleration in the fusion motion mode obtained after fusion exceed the threshold range. That is to say, when the fusion speed exceeds the speed threshold range, the fusion speed needs to be adjusted. Otherwise, the fusion speed does not need to be adjusted. Similarly, when the fusion acceleration exceeds the acceleration threshold range, the fusion acceleration needs to be adjusted. Otherwise, the fusion acceleration does not need to be adjusted.

[0143] Furthermore, according to the set energy consumption threshold, it is determined whether the fusion energy consumption in the fusion motion mode obtained after fusion exceeds the threshold range. In other words, when the fusion energy consumption exceeds the energy consumption threshold range, the fusion energy consumption needs to be adjusted. Otherwise, the fusion energy consumption does not need to be adjusted.

[0144] Step S3.4: Adjust the weight coefficient. That is, adjust the weight coefficient of the fusion movement pattern obtained in step S3.2 according to the movement error and injury risk identified in step S3.3. That is, by adjusting the weight coefficient The size of the movement errors and injury risks identified are adjusted / avoided. The details are as follows:

[0145] Step N1: According to the movement errors and injury risks identified in step S3.3, determine the objective function, specifically:

[0146] ;

[0147] in: is the objective function, is the weight of the motion error, Scoring for motion errors, is the weight of the risk of harm, Score the risk of injury.

[0148] Step N2: Use the objective function determined in step N1 , get the weight coefficient The gradient size is:

[0149] ;

[0150] in: is the gradient of the objective function with respect to the weight coefficient, Add the weight factor The objective function after is the objective function, Is a positive number.

[0151] It is worth noting that positive Usually set to 10 -6 or 10 -8 In this embodiment, the positive number Set to 10 -6 .

[0152] Step N3: According to the gradient of the objective function with respect to the weight coefficient obtained in step N2 The size of the weight coefficient The size of is adjusted so that it moves in the direction that can reduce the objective function. In this embodiment, the adjustment formula is specifically:

[0153] ;

[0154] in: is the weight coefficient at the k+1th iteration, is the weight coefficient at the kth iteration, is the learning rate, is the gradient of the objective function with respect to the weight coefficient at the kth iteration.

[0155] Furthermore, through this adjustment method, the weight coefficient can be continuously adjusted. The size of is adjusted until the corresponding number of iterations or the objective function is completed. Convergence is performed.

[0156] In the specific implementation process, the initial weight coefficient is set to 0.5, the learning rate is set to 0.01, the number of iterations is set to 100, the weight of motion error is set to 0.7, and the weight of injury risk is set to 0.3. According to the data:

[0157] The objective function is:

[0158] ;

[0159] The gradient of the objective function with respect to the weight coefficient is:

[0160] ;

[0161] The weight coefficients for the first iteration are:

[0162] ;

[0163] That is to say, according to the weight coefficient and adjustment formula of the first iteration, the corresponding weight coefficient of each subsequent iteration can be obtained until 100 iterations or the objective function is reached. Convergence is performed.

[0164] Example 2

[0165] refer to Figure 4 and Figure 5 , this embodiment provides a training chair based on leg training data, and the training chair includes an analysis module, a motion module and a control module. In this embodiment, the analysis module is used to obtain the fusion motion pattern when performing leg training, and the motion module is used to set the operating parameters of the training chair according to the joint angles and motion positions corresponding to the obtained fusion motion pattern, so that the user can directly perform leg training. Furthermore, the analysis module and the motion module are electrically connected through the control module. It is worth noting that in the process of obtaining the fusion motion pattern through the analysis module, the obtained fusion motion pattern needs to be simulated, and in the process of simulation, the motion errors and injury risks corresponding to the fusion motion pattern are identified, and at the same time, according to the motion errors and injury risks, the fusion motion pattern is further adjusted to obtain the best fusion motion pattern.

[0166] In this embodiment, the analysis module includes an extraction unit, a model unit and an adjustment unit. Specifically, the extraction unit is used to establish a long short-term memory network model and a convolutional neural network model through multi-dimensional dynamic data (such as joint angle, muscle strength, speed and acceleration, energy consumption), so as to obtain a hidden state vector and a convolution vector. At the same time, the hidden state vector and the convolution vector are fused to obtain a fused feature vector.

[0167] Furthermore, the model unit is used to establish a dynamic model through the fusion feature vector obtained in the extraction unit, and determine the natural movement mode when performing leg training through the dynamic model. At the same time, the personalized movement mode when performing leg training is determined according to the body characteristics. It is worth noting that in the present embodiment, the model unit fuses the natural movement mode with the personalized movement mode to obtain the fused movement mode when performing leg training, thereby obtaining the corresponding joint angle and movement position under the fused movement mode. In other words, the obtained joint angle and movement position are sent to the motion module through the control module, and the training chair in the motion module can perform corresponding leg training according to the data information.

[0168] Furthermore, the adjustment unit is used to simulate the fusion movement pattern determined by the model unit, and in the process of simulation, identify the movement errors and injury risks corresponding to the fusion movement pattern. By analyzing and identifying the movement errors and injury risks, the weight coefficients in the fusion movement pattern are adjusted. In other words, by adjusting the weight coefficients, the fusion movement pattern is optimized and improved, thereby adjusting the corresponding joint angles and movement positions, and then during the actual training of the user, the operation of the training chair can better meet the training and body load requirements.

[0169] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is limited by the attached embodiments and their equivalents.

Claims

1. An analysis method based on leg training data, characterized in that: Included are: Extracting change features: acquiring multidimensional dynamic data during leg training, and acquiring change features from the multidimensional dynamic data by using a time series analysis method; Establishing a dynamic model: establishing a dynamic model based on the change characteristics, and predicting the natural movement pattern during leg training based on the dynamic model, specifically: Divide the user's legs into a plurality of rigid body segments, and obtain the kinetic energy and potential energy corresponding to each rigid body segment, and simultaneously obtain the total kinetic energy and total potential energy during leg training according to the kinetic energy and potential energy corresponding to each rigid body segment; According to the total kinetic energy and total potential energy during the leg training, the size of the Lagrangian quantity is obtained to determine the joint angle and joint torque during the leg training; According to the maximum and minimum bending angles of the knee joint, determine the range of motion of the knee and hip joints during leg training; Adjusting the movement pattern: simulating the predicted natural movement pattern, identifying movement errors and injury risks, and adjusting the natural movement pattern; Obtaining change characteristics from the multi-dimensional dynamic data, including: Acquire time series data, acquire multi-dimensional dynamic data during leg training through sensors, pre-process the multi-dimensional dynamic data, and sort the pre-processed multi-dimensional dynamic data in chronological order to acquire the time series data; Obtain a fused feature vector, obtain a hidden state vector through a long short-term memory network model according to the time series data, obtain a convolution vector through a convolutional neural network model, and fuse the hidden state vector and the convolution vector to obtain the fused feature vector; Obtaining the fused feature vector includes: M1: Obtain hidden state vector: Use the time series data as the input of the long short-term memory network model and output the hidden state vector, specifically: ； in, is the hidden state vector output at time step t, is the leg training data information output at time step t, is the memory unit corresponding to the leg motion state at time step t, is the weight matrix, is the hidden state vector output at time step t-1, is the time series data input at time step t, is the bias vector, is the Sigmoid activation function; M2: Obtain convolution vector: Use the time series data as the input of the convolutional neural network model and output the convolution vector, specifically: ； in, is the convolution vector, is the activation function, is the convolution kernel, For time series data, is the bias term; M3: Obtain a splicing vector: fuse the hidden state vector and the convolution vector to form a splicing vector, and normalize the splicing vector, specifically: ； in: is the normalized concatenation vector, is the concatenation vector, is the minimum value in the concatenated vector, is the maximum value in the concatenated vector; M4: Obtaining a fused feature vector: extracting features from the normalized concatenated vector, and the extracted features are the fused feature vector.

2. The analysis method based on leg training data according to claim 1, characterized in that: When performing leg training, the calculation formulas for the total kinetic energy and total potential energy are specifically: ； in: is the total kinetic energy during leg training, is the total potential energy during leg training, is the mass of the ith rigid body segment, is the acceleration due to gravity, is the position of the ith rigid body segment, is the moment of inertia of the ith rigid body segment, for The inversion of is the angular velocity of the ith rigid body segment, is the velocity of the ith rigid body segment.

3. The analysis method based on leg training data according to claim 1, characterized in that: Adjusting the natural movement pattern includes: Determine personalized exercise mode: through physical characteristics, obtain the relationship between muscle strength and exercise ability, the relationship between height and step length, and the relationship between weight and exercise energy consumption, and determine the personalized exercise mode; Obtaining a fused motion pattern: fusing the personalized motion pattern with the natural motion pattern to obtain the fused motion pattern, specifically: ； in: is the motion position after fusion at time t, is the joint angle after fusion at time t, is the weight coefficient, is the motion position in the personalized motion mode at time t, is the joint angle in the personalized motion mode at time t, is the motion position in the natural motion mode at time t, is the joint angle in the natural motion mode at time t; Identifying movement errors and injury risks: simulating the fused movement pattern to identify movement errors and injury risks; Adjusting weight coefficients: adjusting the weight coefficients in the fused motion model according to the motion error and injury risk.

4. The analysis method based on leg training data according to claim 3, characterized in that: In the process of determining the motion error and injury risk, the joint angle threshold, speed threshold and energy consumption threshold are set, and the fusion joint angle, fusion speed and fusion energy consumption to be adjusted are determined by comparing the fusion joint angle with the joint angle threshold, the fusion speed with the speed threshold, and the fusion speed with the energy consumption threshold, specifically: When the fusion joint angle exceeds the joint angle threshold range, the fusion joint angle needs to be adjusted, otherwise, it does not need to be adjusted; When the fusion speed exceeds the speed threshold range, the fusion speed needs to be adjusted, otherwise, no adjustment is required; When the fusion energy consumption exceeds the energy consumption threshold range, the fusion energy consumption needs to be adjusted; otherwise, no adjustment is required.

5. The analysis method based on leg training data according to claim 3, characterized in that: Adjusting the weight coefficient in the fusion motion mode includes: N1: Based on the motion errors and injury risks, an objective function is established, specifically: ； in: is the objective function, is the weight of the motion error, Scoring for motion errors, is the weight of the risk of harm, To score the risk of harm; N2: Obtain the gradient of the weight coefficient through the objective function, specifically: ； in: is the gradient of the objective function with respect to the weight coefficient, Add the weight factor The objective function after is the objective function, is a positive number; N3: According to the gradient size of the weight coefficient, the size of the weight coefficient is adjusted by an adjustment formula, and the adjustment formula is specifically: ； in: is the weight coefficient at the k+1th iteration, is the weight coefficient at the kth iteration, is the learning rate, is the gradient of the objective function with respect to the weight coefficient at the kth iteration.

6. A training chair for implementing the leg training data analysis method according to any one of claims 1 to 5, characterized in that: Included are: Analysis module: obtaining multi-dimensional dynamic data during leg training, and establishing a dynamic model based on the multi-dimensional dynamic data, obtaining a natural movement pattern during leg training, and obtaining a personalized movement pattern based on body characteristics, and fusing the natural movement pattern with the personalized movement pattern to determine a fused movement pattern during leg training; Movement module: setting the operation parameters of the training chair according to the joint angles and movement positions corresponding to the fusion movement mode; Control module: to exchange information with the analysis module and the motion module; The analysis module includes: Extraction unit: obtaining a hidden state vector and a convolution vector through the multi-dimensional dynamic data, and fusing the hidden state vector and the convolution vector to obtain a fused feature vector; Model unit: determining the natural movement pattern during leg training through the fused feature vector, and determining the personalized movement pattern during leg training according to the body characteristics, and fusing the natural movement pattern with the personalized movement pattern to obtain the fused movement pattern during leg training; An adjustment unit simulates the fused motion pattern, determines motion errors and injury risks during the simulation, and adjusts the weight coefficients in the fused motion pattern according to the motion errors and injury risks.

Images