Underwater treadmill training speed optimization method and device based on key point recognition

Through key point recognition and optical flow algorithms, the motion parameters are tracked in real time, combined with the support vector regression model, the speed of underwater treadmills is intelligently adjusted, which solves the problem of neglecting the characteristics of underwater environment and individual differences in the existing technology, and improves training efficiency and safety .

CN119068544AInactive Publication Date: 2024-12-03GUANGZHOU J&J SANITARY WARE

Patent Information

Application Number
CN202411005200.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The training speed adjustment mechanism of existing underwater treadmills fails to fully consider the underwater environment and individual runners, resulting in mismatch in training speeds, reducing efficiency and increasing the risk of sports injury.

Method used

Using a key point recognition method, the runner's movement key points are tracked in real time through the human body key point detection algorithm and optical flow algorithm, calculate the stride, stride frequency and knee bending angle, and input it into the pre-trained support vector regression model, intelligent Adjust the treadmill speed.

Benefits of technology

This has achieved personalized and dynamic optimization of runners' training speed on underwater treadmills, improving training efficiency and safety, and reducing the risk of sports injury.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068544A_ABST
    Figure CN119068544A_ABST
Patent Text Reader

Abstract

The invention discloses a training speed optimization method and device of an underwater treadmill based on key point recognition. The method comprises the following steps: recording a video segment of a preset duration when a runner moves on the underwater treadmill in a current time period; in the first frame image of the video segment, identifying and positioning motion key points of the runner through a human body key point detection algorithm; tracking the position movement of a motion key point for each subsequent frame of image of the video segment through an optical flow algorithm; calculating the motion parameters of the runner in the current time period according to the position movement change of the motion key points in each frame of image of the video segment, wherein the motion parameters comprise stride, stride frequency and knee joint bending angle; performing feature extraction on the motion parameters to obtain motion parameter features; and inputting the motion parameter characteristics into a pre-trained underwater training speed optimization model to obtain the underwater training speed of the underwater treadmill, which is most suitable for the runner at present. According to the invention, the training efficiency and safety of a runner on the underwater treadmill can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a method and a device for optimizing the training speed of an underwater treadmill based on key point recognition. Background Art

[0002] For the current application of underwater treadmills, the current training speed adjustment mechanism is mostly based on preset training plans or heart rate monitoring of runners, lacking accurate analysis and feedback of the runners' real-time motion status. Current methods often rely on the heart rate belt or built-in heart rate sensor worn by the runner to adjust the speed or inclination of the treadmill by monitoring the heart rate to maintain training within the target heart rate range. However, this heart rate-based adjustment mechanism does not fully consider the impact of the underwater environment on athletic performance, such as the changes in gait and energy consumption caused by water resistance and buoyancy, and the dynamic characteristics of these factors as the runner's motion state changes.

[0003] Therefore, the existing technical solutions fail to fully take into account the particularities of the underwater environment and the individual differences of runners, which may result in the training speed adjustment mechanism of the underwater treadmill not matching the actual movement state of the runner, which not only reduces the training efficiency, but also may increase the risk of sports injuries due to overtraining, improper exercise intensity or delayed feedback. Summary of the invention

[0004] The embodiment of the present invention provides a method and device for optimizing the training speed of an underwater treadmill based on key point recognition, which can improve the training efficiency and safety of runners on the underwater treadmill and effectively reduce the risk of sports injuries.

[0005] An embodiment of the present invention provides a method for optimizing the training speed of an underwater treadmill based on key point recognition, comprising:

[0006] Record a video segment of a preset length when the runner is exercising on the underwater treadmill during the current time period;

[0007] In the first frame image of the video segment, the movement key points of the runner are identified and located by using a human key point detection algorithm; the movement key points include ankle key points, hip key points and knee key points;

[0008] Tracking the position movement of the motion key point for each subsequent frame of the video segment through an optical flow algorithm;

[0009] Calculating the motion parameters of the runner in the current time period according to the position movement changes of the motion key points in each frame image of the video segment: the motion parameters include: stride, step frequency and knee joint bending angle;

[0010] Extracting features of the motion parameters to obtain motion parameter features;

[0011] The motion parameter characteristics are input into a pre-trained underwater training speed optimization model to obtain the underwater training speed of the underwater treadmill that is currently most suitable for the runner.

[0012] As an improvement of the above solution, the step of calculating the motion parameters of the runner in the current time period according to the position movement changes of the motion key points in each frame image of the video segment includes:

[0013] Calculating the stride of the runner in each frame of the video segment according to the positions of the key points of the ankles of the left and right feet of the runner in each frame of the video segment, and calculating the average of the stride of the runner in all the frames of the video segment to obtain the stride of the runner in the current time period;

[0014] Calculating the knee bending angle of the runner in each frame of the video segment according to the positions of the ankle key point of the left or right foot, the knee key point of the left or right foot, and the hip key point of the runner in each frame of the video segment, and calculating the average value of the knee bending angle of the runner in all the frames of the video segment to obtain the knee bending angle of the runner in the current time period;

[0015] The number of gait cycles of the runner in the video segment is calculated according to the longitudinal coordinates of the ankle key points of the left or right foot of the runner in all frame images of the video segment, and the step frequency of the runner in the current time period is calculated according to the number of gait cycles of the runner in the video segment and the preset duration.

[0016] As an improvement of the above solution, the calculation formula of the runner's stride in each frame of image is:

[0017]

[0018] Among them, x ankle,right and g ankle,right (t) respectively represent the x-coordinate and y-coordinate of the key point of the right ankle of the runner in the t-th frame image of the video segment; ankle,left (t) and y akle,left (t) respectively represent the x-coordinate and y-coordinate of the key point of the left ankle of the runner in the t-th frame image of the video segment;

[0019] The calculation formula for the knee bending angle of the runner in each frame of the image is:

[0020]

[0021] Among them, p hip(t), p knnee (t) and p ankle (t) represents the coordinate position vector of the ankle key point of the left or right foot, the knee key point of the left or right foot and the hip key point of the runner in the t-th frame image of the video segment; · represents the vector dot product; ||·|| represents the Euclidean norm of the vector;

[0022] The indicator function δ(i) for detecting the gait cycle of the runner in the video segment is:

[0023]

[0024] Among them, g ankle (i) represents the y coordinate of the left or right ankle key point of the runner in the i-th frame image of the video segment; τ y is a threshold used to distinguish the beginning and end of a gait cycle.

[0025] As an improvement of the above solution, the motion parameter characteristics are input into a pre-trained underwater training speed optimization model to obtain the underwater training speed of the underwater treadmill that is currently most suitable for the runner, including:

[0026] The motion parameter characteristics are input into a pre-trained underwater training speed optimization model based on a support vector regression algorithm to obtain the underwater training speed V that is currently most suitable for the runner on the underwater treadmill. opt ; The algorithm formula of the underwater training speed optimization model is:

[0027]

[0028] Wherein, f is the motion parameter characteristic of the runner; α i and is the Lagrange multiplier determined by the training process; K(f,x i ) is the result of the Gaussian RBF kernel function, which measures the current motion parameter feature f of the runner and each sample x in the training set. i similarity; b is a bias term, which is determined by the training process; i is an index variable used to traverse each sample in the training data set; n is the total number of samples in the training data set, and each sample in the training data set consists of a feature vector of motion parameter features and a corresponding target value of the ideal underwater training speed.

[0029] As an improvement of the above scheme, the training process of the underwater training speed optimization model based on the support vector regression algorithm includes:

[0030] Acquire a training data set; each sample in the training data set consists of a feature vector of a motion parameter feature f and a corresponding target value of an ideal underwater training speed;

[0031] The support vector regression model for constructing the underwater training speed optimization model is:

[0032]

[0033] The Gaussian RBF kernel function is selected as the kernel function of the support vector regression model:

[0034] K(f,x i )=exp(-γ||xx i || 2 ), where ||xx i || is the motion parameter feature f and each sample x in the training set i Euclidean distance in the original feature space; γ is a hyperparameter that controls the width of the kernel function and affects the complexity and fitting ability of the model;

[0035] The following optimization problem is solved for the support vector regression model of the underwater training speed optimization model by a quadratic programming method based on the Lagrange multiplier method, so as to obtain the optimal model parameters of the support vector regression model of the underwater training speed optimization model. The optimization problem includes:

[0036]

[0037] Among them, w is the weight vector, which determines the direction and weight of the feature vector of the motion parameter feature f in the feature space; b is the bias term; ∈ is the allowed prediction error threshold, which means that as long as the error between the predicted value and the true value does not exceed ∈, there will be no penalty; ∈ i and are slack variables, representing the positive and negative prediction errors that exceed ∈, respectively. They are used to quantify the extent to which the prediction error of each training sample exceeds ∈; C is a regularization parameter used to balance model complexity and training error. A larger C value will more strictly punish the training error, while a smaller C value is more inclined to tolerate errors to avoid overfitting; n is the total number of samples in the training data set; y i is the target value of the ideal underwater training speed of the i-th training sample; φ(x i ) is the feature mapping function, which transforms the original feature vector x i Mapping to a higher dimensional space makes it linearly separable in that space.

[0038] Another embodiment of the present invention provides a training speed optimization device for an underwater treadmill based on key point recognition, comprising:

[0039] A recording module, used to record a video segment of a preset duration when the runner is exercising on the underwater treadmill during the current time period;

[0040] A key point recognition module, used to recognize and locate the movement key points of the runner in the first frame image of the video segment by using a human key point detection algorithm; the movement key points include ankle key points, hip key points and knee key points;

[0041] A key point tracking module, used for tracking the position movement of the motion key point for each subsequent frame image of the video segment through an optical flow algorithm;

[0042] A calculation module, for calculating the motion parameters of the runner in the current time period according to the position movement change of the motion key points in each frame image of the video segment: the motion parameters include: stride, step frequency and knee bending angle;

[0043] A feature extraction module, used for extracting features of the motion parameters to obtain motion parameter features;

[0044] The optimization module is used to input the motion parameter characteristics into a pre-trained underwater training speed optimization model to obtain the underwater training speed that is currently most suitable for the runner on the underwater treadmill.

[0045] As an improvement of the above solution, the calculation module is specifically used for:

[0046] Calculating the stride of the runner in each frame of the video segment according to the positions of the key points of the ankles of the left and right feet of the runner in each frame of the video segment, and calculating the average of the stride of the runner in all the frames of the video segment to obtain the stride of the runner in the current time period;

[0047] Calculating the knee bending angle of the runner in each frame of the video segment according to the positions of the ankle key point of the left or right foot, the knee key point of the left or right foot, and the hip key point of the runner in each frame of the video segment, and calculating the average value of the knee bending angle of the runner in all the frames of the video segment to obtain the knee bending angle of the runner in the current time period;

[0048] The number of gait cycles of the runner in the video segment is calculated according to the longitudinal coordinates of the ankle key points of the left or right foot of the runner in all frame images of the video segment, and the step frequency of the runner in the current time period is calculated according to the number of gait cycles of the runner in the video segment and the preset duration.

[0049] As an improvement of the above solution, the calculation formula of the runner's stride in each frame of image is:

[0050]

[0051] Among them, xankle,right and g ankle,right (t) respectively represent the x-coordinate and y-coordinate of the key point of the right ankle of the runner in the t-th frame image of the video segment; ankle,lefi (t) and y ankle,left (t) respectively represent the x-coordinate and y-coordinate of the key point of the left ankle of the runner in the t-th frame image of the video segment;

[0052] The calculation formula for the knee bending angle of the runner in each frame of the image is:

[0053]

[0054] Among them, p hip (t), p knee (t) and p ankle (t) represents the coordinate position vector of the ankle key point of the left or right foot, the knee key point of the left or right foot and the hip key point of the runner in the t-th frame image of the video segment; ˙ represents the vector dot product; ||·|| represents the Euclidean norm of the vector;

[0055] The indicator function δ(i) for detecting the gait cycle of the runner in the video segment is:

[0056]

[0057] Among them, y ankle (i) represents the y coordinate of the left or right ankle key point of the runner in the i-th frame image of the video segment; τ y is a threshold used to distinguish the beginning and end of a gait cycle.

[0058] As an improvement of the above solution, the optimization module is specifically used for:

[0059] The motion parameter characteristics are input into a pre-trained underwater training speed optimization model based on a support vector regression algorithm to obtain the underwater training speed V that is currently most suitable for the runner on the underwater treadmill. opt ; The algorithm formula of the underwater training speed optimization model is:

[0060]

[0061] Wherein, f is the motion parameter characteristic of the runner; α i and is the Lagrange multiplier determined by the training process; K(f, x i ) is the result of the Gaussian RBF kernel function, which measures the current motion parameter feature f of the runner and each sample X in the training set. isimilarity; b is a bias term, which is determined by the training process; i is an index variable used to traverse each sample in the training data set; n is the total number of samples in the training data set, and each sample in the training data set consists of a feature vector of motion parameter features and a corresponding target value of the ideal underwater training speed.

[0062] As an improvement of the above solution, the device further includes a training module, and the training module is specifically used for:

[0063] Acquire a training data set; each sample in the training data set consists of a feature vector of a motion parameter feature f and a corresponding target value of an ideal underwater training speed;

[0064] The support vector regression model for constructing the underwater training speed optimization model is:

[0065]

[0066] The Gaussian RBF kernel function is selected as the kernel function of the support vector regression model:

[0067] K(f,x i )=exp(-γ||xx i || 2 ), where ||xx i || is the motion parameter feature f and each sample X in the training set i Euclidean distance in the original feature space; γ is a hyperparameter that controls the width of the kernel function and affects the complexity and fitting ability of the model;

[0068] The following optimization problem is solved for the support vector regression model of the underwater training speed optimization model by a quadratic programming method based on the Lagrange multiplier method, so as to obtain the optimal model parameters of the support vector regression model of the underwater training speed optimization model. The optimization problem includes:

[0069]

[0070] Among them, w is the weight vector, which determines the direction and weight of the feature vector of the motion parameter feature f in the feature space; b is the bias term; ∈ is the allowed prediction error threshold, which means that as long as the error between the predicted value and the true value does not exceed ∈, there will be no penalty; ∈ i and are slack variables, representing the positive and negative prediction errors that exceed ∈, respectively. They are used to quantify the extent to which the prediction error of each training sample exceeds ∈; C is a regularization parameter used to balance model complexity and training error. A larger C value will more strictly punish the training error, while a smaller C value is more inclined to tolerate errors to avoid overfitting; n is the total number of samples in the training data set; yi is the target value of the ideal underwater training speed of the i-th training sample; φ(x i ) is the feature mapping function, which transforms the original feature vector x i Mapping to a higher dimensional space makes it linearly separable in that space.

[0071] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0072] The embodiment of the present invention first uses the human key point detection algorithm in the first frame of the video to accurately locate the key movement parts of the runner, including the ankle, hip and knee. Subsequently, the optical flow algorithm is used to track the continuous displacement of these key points in subsequent video frames in real time, so as to obtain the dynamic information of the runner's movement. Based on the movement changes of the key points, the key movement parameters of the runner, such as the stride, step frequency and knee bending angle, are calculated, and then these parameters are converted into feature vectors that can be recognized by the model through feature extraction. Finally, the extracted motion parameter features are input into the pre-trained underwater training speed optimization model, which is based on the machine learning algorithm and can intelligently predict and adjust the speed of the treadmill according to the current movement state of the runner and the characteristics of the underwater environment to achieve the best training effect. From the above analysis, it can be seen that the embodiment of the present invention solves the problem that the training speed adjustment mechanism in the prior art ignores the underwater environment characteristics and individual training differences by capturing and analyzing the real-time movement state of the runner in the underwater environment, realizes the personalization and dynamic optimization of the training speed of the runner on the underwater treadmill, significantly improves the training efficiency and safety of the runner on the underwater treadmill, and effectively reduces the risk of sports injuries. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a flow chart of a method for optimizing the training speed of an underwater treadmill based on key point recognition provided by one embodiment of the present invention;

[0074] Figure 2 It is a structural schematic diagram of a training speed optimization device for an underwater treadmill based on key point recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0075] 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.

[0076] See also Figure 1, is a flow chart of a method for optimizing the training speed of an underwater treadmill based on key point recognition provided by an embodiment of the present invention. The method for optimizing the training speed of an underwater treadmill based on key point recognition comprises steps S10 to S15:

[0077] S10, recording a video segment of a preset length of time when the runner is exercising on the underwater treadmill in the current time period;

[0078] S11, in the first frame image of the video segment, identifying and locating the movement key points of the runner by using a human key point detection algorithm; the movement key points include ankle key points, hip key points and knee key points;

[0079] S12, tracking the position movement of the motion key point for each subsequent frame of the video segment through an optical flow algorithm;

[0080] S13, calculating the motion parameters of the runner in the current time period according to the position movement change of the motion key points in each frame image of the video segment, wherein the motion parameters include: stride length, step frequency and knee joint bending angle;

[0081] S14, extracting features from the motion parameters to obtain motion parameter features;

[0082] S15, inputting the motion parameter characteristics into a pre-trained underwater training speed optimization model to obtain the underwater training speed of the underwater treadmill that is currently most suitable for the runner.

[0083] The embodiment of the present invention first uses the human key point detection algorithm in the first frame of the video to accurately locate the key movement parts of the runner, including the ankle, hip and knee. Subsequently, the optical flow algorithm is used to track the continuous displacement of these key points in subsequent video frames in real time, so as to obtain the dynamic information of the runner's movement. Based on the movement changes of the key points, the key movement parameters of the runner, such as the stride, step frequency and knee bending angle, are calculated, and then these parameters are converted into feature vectors that can be recognized by the model through feature extraction. Finally, the extracted motion parameter features are input into the pre-trained underwater training speed optimization model, which is based on the machine learning algorithm and can intelligently predict and adjust the speed of the treadmill according to the current movement state of the runner and the characteristics of the underwater environment to achieve the best training effect. From the above analysis, it can be seen that the embodiment of the present invention solves the problem that the training speed adjustment mechanism in the prior art ignores the underwater environment characteristics and individual training differences by capturing and analyzing the real-time movement state of the runner in the underwater environment, realizes the personalization and dynamic optimization of the training speed of the runner on the underwater treadmill, significantly improves the training efficiency and safety of the runner on the underwater treadmill, and effectively reduces the risk of sports injuries.

[0084] It is understandable that after obtaining the underwater training speed that is currently most suitable for the runner of the underwater treadmill, the process of adjusting the training speed of the underwater treadmill can be: the computing device (such as the controller of the underwater treadmill, or the mobile phone or computer connected to the underwater treadmill for communication) calculates the underwater training speed that is most suitable for the runner through the pre-trained underwater training speed optimization model, and it generates a speed adjustment instruction. This instruction is usually a digital signal, indicating the specific numerical value of the target speed. The speed instruction is sent to the control system of the underwater treadmill through the communication interface. After receiving the speed instruction from the computing device, the control system parses the target speed information in the instruction. The control system of the underwater treadmill converts the target speed information into a specific control command for the motor. This may involve adjusting the voltage, current or frequency of the motor to change the speed of the treadmill belt.

[0085] It should be noted that the human key point detection algorithm can be OpenPose, HRNet (High-ResolutionNet) or Simple Baseline. The optical flow algorithm is used to estimate the pixel motion between consecutive image frames and is an important tool for tracking the position changes of objects and key points. The optical flow algorithm can be the Lucas-Kanade algorithm or the Horn-Schunck algorithm, etc., which are not specifically limited here. The feature extraction algorithm can refer to various existing feature extraction algorithms.

[0086] The preset duration may be 1-3 seconds, that is, each 1-3 seconds of video is recorded to evaluate the current underwater motion condition of the runner.

[0087] In one embodiment, the step of calculating the motion parameters of the runner in the current time period according to the position movement changes of the motion key points in each frame image of the video segment includes:

[0088] Calculating the stride of the runner in each frame of the video segment according to the positions of the key points of the ankles of the left and right feet of the runner in each frame of the video segment, and calculating the average of the stride of the runner in all the frames of the video segment to obtain the stride of the runner in the current time period;

[0089] Calculating the knee bending angle of the runner in each frame of the video segment according to the positions of the ankle key point of the left or right foot, the knee key point of the left or right foot, and the hip key point of the runner in each frame of the video segment, and calculating the average value of the knee bending angle of the runner in all the frames of the video segment to obtain the knee bending angle of the runner in the current time period;

[0090] The number of gait cycles of the runner in the video segment is calculated according to the longitudinal coordinates of the ankle key points of the left or right foot of the runner in all frame images of the video segment, and the step frequency of the runner in the current time period is calculated according to the number of gait cycles of the runner in the video segment and the preset duration.

[0091] Specifically, the calculation formula for the runner's stride in each frame of the image is:

[0092]

[0093] Among them, x ankle,right and ankle,right (t) respectively represent the x-coordinate and y-coordinate of the key point of the right ankle of the runner in the t-th frame image of the video segment; ankle,left (t) and y ankle,left (t) respectively represent the x-coordinate and y-coordinate of the key point of the left ankle of the runner in the t-th frame image of the video segment;

[0094] The calculation formula for the knee bending angle of the runner in each frame of the image is:

[0095]

[0096] Among them, p hip (t), p knee (t) and p ankle (t) represents the coordinate position vector of the ankle key point of the left or right foot, the knee key point of the left or right foot and the hip key point of the runner in the t-th frame image of the video segment; · represents the vector dot product; ||·|| represents the Euclidean norm of the vector;

[0097] The indicator function δ(i) for detecting the gait cycle of the runner in the video segment is:

[0098]

[0099] Among them, y ankle (i) represents the y coordinate of the left or right ankle key point of the runner in the i-th frame image of the video segment; τ y is a threshold used to distinguish the beginning and end of a gait cycle.

[0100] In this embodiment, by accurately capturing the runner's key movement points, including ankles, knees and hips, from the video data, and then analyzing the runner's movement parameters in real time, including stride, knee flexion angle and step frequency, personalized and intelligent management of underwater running training can be achieved, thereby ensuring that the adjustment of training speed can closely fit the runner's ability and needs.

[0101] In one embodiment, the step of inputting the motion parameter characteristics into a pre-trained underwater training speed optimization model to obtain the underwater training speed of the underwater treadmill that is currently most suitable for the runner includes:

[0102] The motion parameter characteristics are input into a pre-trained underwater training speed optimization model based on a support vector regression algorithm to obtain the underwater training speed V that is currently most suitable for the runner on the underwater treadmill. opt ; The algorithm formula of the underwater training speed optimization model is:

[0103]

[0104] Wherein, f is the motion parameter characteristic of the runner; α i and is the Lagrange multiplier determined by the training process; K(f,x i ) is the result of the Gaussian RBF kernel function, which measures the current motion parameter feature f of the runner and each sample x in the training set. i similarity; b is a bias term, which is determined by the training process; i is an index variable used to traverse each sample in the training data set; n is the total number of samples in the training data set, and each sample in the training data set consists of a feature vector of motion parameter features and a corresponding target value of the ideal underwater training speed.

[0105] In the present embodiment, the motion parameter features extracted based on computer vision technology, such as stride, step frequency and knee bending angle, are used as inputs, and a pre-trained support vector regression (SVR) model is used to predict and optimize the training speed of the underwater treadmill. The SVR model utilizes the Gaussian radial basis function (RBF) kernel, which can handle nonlinear relationships and predict the most suitable underwater training speed by calculating the similarity between the current motion state features of the runner and the sample features in the historical training data set. The Lagrange multipliers and bias terms in the model are obtained by optimizing the training process to ensure that the model can accurately map the motion parameter features to the ideal training speed. Therefore, the personalization and dynamic optimization of underwater treadmill training can be achieved, ensuring that the training speed can respond to the changes in the runner's motion state in real time, thereby improving the efficiency and safety of training. By utilizing a machine learning algorithm, particularly support vector regression, this embodiment can intelligently adjust the speed of the underwater treadmill according to the current motion parameters of the runner to match the runner's optimal motion state, avoid overtraining and sports injuries, and maximize the training benefit.

[0106] As an improvement of the above scheme, the training process of the underwater training speed optimization model based on the support vector regression algorithm includes:

[0107] Acquire a training data set; each sample in the training data set consists of a feature vector of a motion parameter feature f and a corresponding target value of an ideal underwater training speed;

[0108] The support vector regression model for constructing the underwater training speed optimization model is:

[0109]

[0110] The Gaussian RBF kernel function is selected as the kernel function of the support vector regression model:

[0111] K(f,x i )=exp(-γ||xx i || 2 ), where ||xx i || is the motion parameter feature f and each sample x in the training set i Euclidean distance in the original feature space; γ is a hyperparameter that controls the width of the kernel function and affects the complexity and fitting ability of the model;

[0112] The following optimization problem is solved for the support vector regression model of the underwater training speed optimization model by a quadratic programming method based on the Lagrange multiplier method, so as to obtain the optimal model parameters of the support vector regression model of the underwater training speed optimization model. The optimization problem includes:

[0113]

[0114] Among them, ω is the weight vector, which determines the direction and weight of the feature vector of the motion parameter feature f in the feature space; b is the bias term; ∈ is the allowed prediction error threshold, which means that as long as the error between the predicted value and the true value does not exceed ∈, there will be no penalty; ∈i and are slack variables, representing the positive and negative prediction errors that exceed ∈, respectively. They are used to quantify the extent to which the prediction error of each training sample exceeds ∈; C is a regularization parameter used to balance model complexity and training error. A larger C value will more strictly punish the training error, while a smaller C value is more inclined to tolerate errors to avoid overfitting; n is the total number of samples in the training data set; y i is the target value of the ideal underwater training speed of the i-th training sample; φ(x i ) is the feature mapping function, which transforms the original feature vector x i Mapping to a higher dimensional space makes it linearly separable in that space.

[0115] In the present embodiment, the motion parameter features are first extracted from the collected training data set, and each sample not only contains the vector representation of these features, but also associates an ideal underwater training speed target value. In order to build the model, the Gaussian radial basis function (RBF) kernel is selected, which can effectively handle nonlinear relationships and control the complexity and generalization ability of the model by adjusting the hyperparameter γ. In the model training process, the Lagrange multiplier method is used to solve the quadratic programming problem to find the optimal weight vector αα and bias term b, and the slack variables ξ and ξ* are introduced to quantify the degree to which the prediction error exceeds the allowable range, ensuring that the model can accurately predict and avoid overfitting. Parameter C, as a regularization term, helps balance the complexity of the model with the training error, ensuring that the model reaches a good compromise between the generalization ability and the performance on the training set. Therefore, the present embodiment creates a highly accurate and robust prediction model by utilizing support vector regression, which can predict the ideal speed of the athlete's underwater training according to the motion parameter characteristics, thereby guiding the formulation of personalized and efficient swimming training plans, and improving training effects and competitive performance.

[0116] See also Figure 2 , is a training speed optimization device for an underwater treadmill based on key point recognition provided by an embodiment of the present invention. The training speed optimization device for an underwater treadmill based on key point recognition comprises:

[0117] A recording module 10, for recording a video segment of a preset duration when the runner is exercising on the underwater treadmill during a current time period;

[0118] A key point recognition module 11 is used to recognize and locate the movement key points of the runner in the first frame image of the video segment by using a human key point detection algorithm; the movement key points include ankle key points, hip key points and knee key points;

[0119] A key point tracking module 12, configured to track the position movement of the motion key point for each subsequent frame of the video segment using an optical flow algorithm;

[0120] A calculation module 13, for calculating the motion parameters of the runner in the current time period according to the position movement change of the motion key points in each frame image of the video segment: the motion parameters include: stride, step frequency and knee bending angle;

[0121] A feature extraction module 14 is used to extract features of the motion parameters to obtain motion parameter features;

[0122] The optimization module 15 is used to input the motion parameter characteristics into a pre-trained underwater training speed optimization model to obtain the underwater training speed that is currently most suitable for the runner on the underwater treadmill.

[0123] The embodiment of the present invention first uses the human key point detection algorithm in the first frame of the video to accurately locate the key movement parts of the runner, including the ankle, hip and knee. Subsequently, the optical flow algorithm is used to track the continuous displacement of these key points in subsequent video frames in real time, so as to obtain the dynamic information of the runner's movement. Based on the movement changes of the key points, the key movement parameters of the runner, such as the stride, step frequency and knee bending angle, are calculated, and then these parameters are converted into feature vectors that can be recognized by the model through feature extraction. Finally, the extracted motion parameter features are input into the pre-trained underwater training speed optimization model, which is based on the machine learning algorithm and can intelligently predict and adjust the speed of the treadmill according to the current movement state of the runner and the characteristics of the underwater environment to achieve the best training effect. From the above analysis, it can be seen that the embodiment of the present invention solves the problem that the training speed adjustment mechanism in the prior art ignores the underwater environment characteristics and individual training differences by capturing and analyzing the real-time movement state of the runner in the underwater environment, realizes the personalization and dynamic optimization of the training speed of the runner on the underwater treadmill, significantly improves the training efficiency and safety of the runner on the underwater treadmill, and effectively reduces the risk of sports injuries.

[0124] As an improvement of the above solution, the calculation module is specifically used for:

[0125] Calculating the stride of the runner in each frame of the video segment according to the positions of the key points of the ankles of the left and right feet of the runner in each frame of the video segment, and calculating the average of the stride of the runner in all the frames of the video segment to obtain the stride of the runner in the current time period;

[0126] Calculating the knee bending angle of the runner in each frame of the video segment according to the positions of the ankle key point of the left or right foot, the knee key point of the left or right foot, and the hip key point of the runner in each frame of the video segment, and calculating the average value of the knee bending angle of the runner in all the frames of the video segment to obtain the knee bending angle of the runner in the current time period;

[0127] The number of gait cycles of the runner in the video segment is calculated according to the longitudinal coordinates of the ankle key points of the left or right foot of the runner in all frame images of the video segment, and the step frequency of the runner in the current time period is calculated according to the number of gait cycles of the runner in the video segment and the preset duration.

[0128] As an improvement of the above solution, the calculation formula of the runner's stride in each frame of image is:

[0129]

[0130] Among them, x ankle,rightand ankle,right (t) respectively represent the x-coordinate and y-coordinate of the key point of the right ankle of the runner in the t-th frame image of the video segment; ankle,left (t) and y ankle,left (t) respectively represent the x-coordinate and y-coordinate of the key point of the left ankle of the runner in the t-th frame image of the video segment;

[0131] The calculation formula for the knee bending angle of the runner in each frame of the image is:

[0132]

[0133] Among them, p hip (t), p knee (t) and p ankle (t) represents the coordinate position vector of the left or right ankle key point, the left or right knee key point and the hip key point of the runner in the t-th frame image of the video segment; represents the vector dot product; represents the Euclidean norm of the vector;

[0134] The indicator function δ(i) for detecting the gait cycle of the runner in the video segment is:

[0135]

[0136] Among them, y ankle (i) represents the y coordinate of the left or right ankle key point of the runner in the i-th frame image of the video segment; τ y is a threshold used to distinguish the beginning and end of a gait cycle.

[0137] As an improvement of the above solution, the optimization module is specifically used for:

[0138] The motion parameter characteristics are input into a pre-trained underwater training speed optimization model based on a support vector regression algorithm to obtain the underwater training speed V that is currently most suitable for the runner on the underwater treadmill. opt ; The algorithm formula of the underwater training speed optimization model is:

[0139]

[0140] Wherein, f is the motion parameter characteristic of the runner; α i and is the Lagrange multiplier determined by the training process; K(f-xii) is the result of the Gaussian RBF kernel function, which measures the runner's current motion parameter feature f and each sample X in the training set. isimilarity; b is a bias term, which is determined by the training process; i is an index variable used to traverse each sample in the training data set; n is the total number of samples in the training data set, and each sample in the training data set consists of a feature vector of motion parameter features and a corresponding target value of the ideal underwater training speed.

[0141] As an improvement of the above solution, the device further includes a training module, and the training module is specifically used for:

[0142] Acquire a training data set; each sample in the training data set consists of a feature vector of a motion parameter feature f and a corresponding target value of an ideal underwater training speed;

[0143] The support vector regression model for constructing the underwater training speed optimization model is:

[0144]

[0145] The Gaussian RBF kernel function is selected as the kernel function of the support vector regression model: K(f, x i )exp(-γ||xx i || 2 ), where ||xx i || is the motion parameter feature f and each sample x in the training set i Euclidean distance in the original feature space; γ is a hyperparameter that controls the width of the kernel function and affects the complexity and fitting ability of the model;

[0146] The following optimization problem is solved for the support vector regression model of the underwater training speed optimization model by a quadratic programming method based on the Lagrange multiplier method, so as to obtain the optimal model parameters of the support vector regression model of the underwater training speed optimization model. The optimization problem includes:

[0147]

[0148] Among them, w is the weight vector, which determines the direction and weight of the feature vector of the motion parameter feature f in the feature space; b is the bias term; ∈ is the allowed prediction error threshold, which means that as long as the error between the predicted value and the true value does not exceed ∈, there will be no penalty; ∈ i and are slack variables, representing the positive and negative prediction errors that exceed ∈, respectively. They are used to quantify the extent to which the prediction error of each training sample exceeds ∈; C is a regularization parameter used to balance model complexity and training error. A larger C value will more strictly punish the training error, while a smaller C value is more inclined to tolerate errors to avoid overfitting; n is the total number of samples in the training data set; y iis the target value of the ideal underwater training speed of the i-th training sample; φ(x i ) is the feature mapping function, which transforms the original feature vector x i Mapping to a higher dimensional space makes it linearly separable in that space.

[0149] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0150] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A training speed optimization method for an underwater treadmill based on key point recognition, characterized in that: include: Record a video segment of a preset length when the runner is exercising on the underwater treadmill during the current time period; In the first frame image of the video segment, the movement key points of the runner are identified and located by using a human key point detection algorithm; the movement key points include ankle key points, hip key points and knee key points; Tracking the position movement of the motion key point for each subsequent frame of the video segment through an optical flow algorithm; Calculating the motion parameters of the runner in the current time period according to the position movement changes of the motion key points in each frame image of the video segment: the motion parameters include: stride, step frequency and knee joint bending angle; Extracting features of the motion parameters to obtain motion parameter features; The motion parameter characteristics are input into a pre-trained underwater training speed optimization model to obtain the underwater training speed of the underwater treadmill that is currently most suitable for the runner.

2. The training speed optimization method of an underwater treadmill based on key point recognition as claimed in claim 1, characterized in that: The step of calculating the motion parameters of the runner in the current time period according to the position movement changes of the motion key points in each frame image of the video segment includes: Calculating the stride of the runner in each frame of the video segment according to the positions of the key points of the ankles of the left and right feet of the runner in each frame of the video segment, and calculating the average of the stride of the runner in all the frames of the video segment to obtain the stride of the runner in the current time period; Calculating the knee bending angle of the runner in each frame of the video segment according to the positions of the ankle key point of the left or right foot, the knee key point of the left or right foot, and the hip key point of the runner in each frame of the video segment, and calculating the average value of the knee bending angle of the runner in all the frames of the video segment to obtain the knee bending angle of the runner in the current time period; The number of gait cycles of the runner in the video segment is calculated according to the longitudinal coordinates of the ankle key points of the left or right foot of the runner in all frame images of the video segment, and the step frequency of the runner in the current time period is calculated according to the number of gait cycles of the runner in the video segment and the preset duration.

3. The training speed optimization method of an underwater treadmill based on key point recognition as claimed in claim 2, characterized in that: The formula for calculating the stride of the runner in each frame of the image is: Among them, x ankle,right and ankle,right (t) respectively represent the x-coordinate and y-coordinate of the key point of the right ankle of the runner in the t-th frame image of the video segment; ankle,left (t) and y ankle,left (t) respectively represent the x-coordinate and y-coordinate of the key point of the left ankle of the runner in the t-th frame image of the video segment; The calculation formula for the knee bending angle of the runner in each frame of the image is: Among them, p hip (t), p knee (t) and p ankle (t) represents the coordinate position vector of the ankle key point of the left or right foot, the knee key point of the left or right foot and the hip key point of the runner in the t-th frame image of the video segment; ˙ represents the vector dot product; ||·|| represents the Euclidean norm of the vector; The indicator function δ(i) for detecting the gait cycle of the runner in the video segment is: Among them, y ankle (i) represents the y coordinate of the left or right ankle key point of the runner in the i-th frame image of the video segment; τ y is a threshold used to distinguish the beginning and end of a gait cycle.

4. The training speed optimization method of an underwater treadmill based on key point recognition as claimed in claim 1, characterized in that: The step of inputting the motion parameter characteristics into a pre-trained underwater training speed optimization model to obtain the underwater training speed of the underwater treadmill that is currently most suitable for the runner comprises: The motion parameter characteristics are input into a pre-trained underwater training speed optimization model based on a support vector regression algorithm to obtain the underwater training speed V that is currently most suitable for the runner on the underwater treadmill. opt ; The algorithm formula of the underwater training speed optimization model is: Wherein, f is the motion parameter characteristic of the runner; α i and is the Lagrange multiplier determined by the training process; K(f, x i ) is the result of the Gaussian RBF kernel function, which measures the current motion parameter feature f of the runner and each sample x in the training set. i similarity; b is a bias term, which is determined by the training process; i is an index variable used to traverse each sample in the training data set; n is the total number of samples in the training data set, and each sample in the training data set consists of a feature vector of motion parameter features and a corresponding target value of the ideal underwater training speed.

5. The training speed optimization method of an underwater treadmill based on key point recognition as claimed in claim 4, characterized in that: The training process of the underwater training speed optimization model based on the support vector regression algorithm includes: Acquire a training data set; each sample in the training data set consists of a feature vector of a motion parameter feature f and a corresponding target value of an ideal underwater training speed; The support vector regression model for constructing the underwater training speed optimization model is: The Gaussian RBF kernel function is selected as the kernel function of the support vector regression model: K(f,x i )=exp(-γ||xx i || 2 ), where ||xx i || is the motion parameter feature f and each sample X in the training set i Euclidean distance in the original feature space; γ is a hyperparameter that controls the width of the kernel function and affects the complexity and fitting ability of the model; The following optimization problem is solved for the support vector regression model of the underwater training speed optimization model by a quadratic programming method based on the Lagrange multiplier method, so as to obtain the optimal model parameters of the support vector regression model of the underwater training speed optimization model. The optimization problem includes: Among them, ω is the weight vector, which determines the direction and weight of the feature vector of the motion parameter feature f in the feature space; b is the bias term; ∈ is the allowed prediction error threshold, which means that as long as the error between the predicted value and the true value does not exceed ∈, there will be no penalty; ∈ i and are slack variables, representing the positive and negative prediction errors that exceed ∈, respectively. They are used to quantify the extent to which the prediction error of each training sample exceeds ∈; C is a regularization parameter used to balance model complexity and training error. A larger C value will more strictly punish the training error, while a smaller C value is more inclined to tolerate errors to avoid overfitting; n is the total number of samples in the training data set; y i is the target value of the ideal underwater training speed of the i-th training sample; φ(x i ) is the feature mapping function, which transforms the original feature vector X i Mapping to a higher dimensional space makes it linearly separable in that space.

6. A training speed optimization device for an underwater treadmill based on key point recognition, characterized in that: include: A recording module, used to record a video segment of a preset duration when the runner is exercising on the underwater treadmill during the current time period; A key point recognition module, used to recognize and locate the movement key points of the runner in the first frame image of the video segment by using a human key point detection algorithm; the movement key points include ankle key points, hip key points and knee key points; A key point tracking module, used for tracking the position movement of the motion key point for each subsequent frame image of the video segment through an optical flow algorithm; A calculation module, for calculating the motion parameters of the runner in the current time period according to the position movement change of the motion key points in each frame image of the video segment: the motion parameters include: stride, step frequency and knee bending angle; A feature extraction module, used for extracting features of the motion parameters to obtain motion parameter features; The optimization module is used to input the motion parameter characteristics into a pre-trained underwater training speed optimization model to obtain the underwater training speed that is currently most suitable for the runner on the underwater treadmill.

7. The training speed optimization device for an underwater treadmill based on key point recognition as claimed in claim 6, characterized in that: The calculation module is specifically used for: Calculating the stride of the runner in each frame of the video segment according to the positions of the key points of the ankles of the left and right feet of the runner in each frame of the video segment, and calculating the average of the stride of the runner in all the frames of the video segment to obtain the stride of the runner in the current time period; Calculating the knee bending angle of the runner in each frame of the video segment according to the positions of the ankle key point of the left or right foot, the knee key point of the left or right foot, and the hip key point of the runner in each frame of the video segment, and calculating the average value of the knee bending angle of the runner in all the frames of the video segment to obtain the knee bending angle of the runner in the current time period; The number of gait cycles of the runner in the video segment is calculated according to the longitudinal coordinates of the ankle key points of the left or right foot of the runner in all frame images of the video segment, and the step frequency of the runner in the current time period is calculated according to the number of gait cycles of the runner in the video segment and the preset duration.

8. The training speed optimization device for an underwater treadmill based on key point recognition as claimed in claim 7, characterized in that: The formula for calculating the stride of the runner in each frame of the image is: Among them, x ankle,right and ankle,right (t) respectively represent the x-coordinate and y-coordinate of the key point of the right ankle of the runner in the t-th frame image of the video segment; ankle,left (t) and y ankle,left (t) respectively represent the x-coordinate and y-coordinate of the key point of the left ankle of the runner in the t-th frame image of the video segment; The calculation formula for the knee bending angle of the runner in each frame of the image is: Among them, p hip (t), p knee (t) and p ankle (t) represents the coordinate position vector of the ankle key point of the left or right foot, the knee key point of the left or right foot and the hip key point of the runner in the t-th frame image of the video segment; ˙ represents the vector dot product; ||·|| represents the Euclidean norm of the vector; The indicator function δ(i) for detecting the gait cycle of the runner in the video segment is: Among them, y ankle (i) represents the y coordinate of the left or right ankle key point of the runner in the i-th frame image of the video segment; τ y is a threshold used to distinguish the beginning and end of a gait cycle.

9. The training speed optimization device for an underwater treadmill based on key point recognition as claimed in claim 6, characterized in that: The optimization module is specifically used for: The motion parameter characteristics are input into a pre-trained underwater training speed optimization model based on a support vector regression algorithm to obtain the underwater training speed V that is currently most suitable for the runner on the underwater treadmill. opt ; The algorithm formula of the underwater training speed optimization model is: Wherein, f is the motion parameter characteristic of the runner; α i and is the Lagrange multiplier determined by the training process; K(f,x i ) is the result of the Gaussian RBF kernel function, which measures the current motion parameter feature f of the runner and each sample X in the training set. i similarity; b is a bias term, which is determined by the training process; i is an index variable used to traverse each sample in the training data set; n is the total number of samples in the training data set, and each sample in the training data set consists of a feature vector of motion parameter features and a corresponding target value of the ideal underwater training speed.

10. The training speed optimization device for an underwater treadmill based on key point recognition as claimed in claim 9, characterized in that: It also includes a training module, which is specifically used for: Acquire a training data set; each sample in the training data set consists of a feature vector of a motion parameter feature f and a corresponding target value of an ideal underwater training speed; The support vector regression model for constructing the underwater training speed optimization model is: The Gaussian RBF kernel function is selected as the kernel function of the support vector regression model: K(f,x i )=exp(-γ||xx i || 2 ), where ||xx i || is the motion parameter feature f and each sample X in the training set i Euclidean distance in the original feature space; γ is a hyperparameter that controls the width of the kernel function and affects the complexity and fitting ability of the model; The following optimization problem is solved for the support vector regression model of the underwater training speed optimization model by a quadratic programming method based on the Lagrange multiplier method, so as to obtain the optimal model parameters of the support vector regression model of the underwater training speed optimization model. The optimization problem includes: Among them, ω is the weight vector, which determines the direction and weight of the feature vector of the motion parameter feature f in the feature space; b is the bias term; ∈ is the allowed prediction error threshold, which means that as long as the error between the predicted value and the true value does not exceed ∈, there will be no penalty; ∈ i and are slack variables, representing the positive and negative prediction errors that exceed ∈, respectively. They are used to quantify the extent to which the prediction error of each training sample exceeds ∈; C is a regularization parameter used to balance model complexity and training error. A larger C value will more strictly punish the training error, while a smaller C value is more inclined to tolerate errors to avoid overfitting; n is the total number of samples in the training data set; y i is the target value of the ideal underwater training speed of the i-th training sample; φ(x i ) is the feature mapping function, which transforms the original feature vector x i Mapping to a higher dimensional space makes it linearly separable in that space.

Citation Information

Patent Citations

  • Fitness action recognition model, model training method and fitness action recognition method

    CN115294660A

  • Fitness equipment monitoring method and system

    CN116966513A

  • Preparation method of Bi-Te based thermoelectric material using hot isostatic pressing post-treatment

    KR1020250114742A

  • System and method for determining running power

    US20230158368A1

  • Accurate ambulatory gait analysis with wearable sensors using transductive learning inference models

    WO2022232697A1

Cited By

  • Treadmill falling early warning and emergency shutdown method and system based on posture recognition

    CN121570786A

  • Pedestrian behavior prediction method for autonomous vehicle

    CN121583001A

  • Treadmill stride pattern recognition method and system

    CN122615800A