Quaternion-based three-dimensional human pose estimation motion correction method
By employing a quaternion-based 3D human pose estimation method, which utilizes Euclidean distance and angle change calculations, and combines convolutional neural networks for feature extraction and pose correction, the problem of insufficient accuracy in 3D human pose recognition is solved, achieving a more efficient motion correction effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV OF TECH
- Filing Date
- 2023-02-24
- Publication Date
- 2026-06-26
AI Technical Summary
Existing technologies lack accuracy in 3D human pose recognition, especially in situations with occlusion and complex environments where it is difficult to accurately identify user movements, which increases the likelihood of sports injuries or failure to meet standards. Furthermore, hiring a personal trainer is expensive.
A quaternion-based 3D human pose estimation method is adopted. By calculating the Euclidean distance and angle changes of the joints, a convolutional neural network is used for feature extraction and pose correction. Quaternion rotation is used to improve the accuracy of pose recognition, and a velocity correction method is used to reduce occlusion and environmental influences.
It improves the accuracy of 3D human posture recognition and the reliability of motion correction, reduces the probability of user injury, and reduces the reliance on professional guidance.
Smart Images
Figure CN116188520B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of three-dimensional human posture estimation and sports health, and particularly relates to a method for achieving the purpose of movement correction by combining human posture recognition with quaternions. Background Technology
[0002] It is well known that maintaining good health can significantly reduce the incidence of disease; for students, good physical fitness is also a basic requirement for diligent study. How students can exercise efficiently and professionally is a crucial issue. Currently, due to insufficient funds, limited space, and other unfavorable factors, most students lack access to professional fitness equipment. Therefore, they should focus on exercises that require less reliance on or use reasonably priced equipment. Without the guidance of a professional fitness coach, the risk of injury is high due to exercising too fast or failing to meet fitness standards due to exercising too slowly. However, hiring a personal trainer is often quite expensive.
[0003] Technically, 3D human pose recognition faces challenges in accuracy due to factors such as human occlusion and complex environmental backgrounds. In 3D space, quaternions can be used in pose representation. A quaternion is composed of a real number plus three imaginary units i, j, and k, and can be considered as a linear combination of 1, i, j, and k. In other words, a quaternion can generally be represented as:
[0004] q = a + bi + cj + dk
[0005] Where a, b, c, and d are real numbers, i 2 =j 2 =k 2 =ijk=-1.
[0006] In the representation of quaternions, the real and imaginary parts are often separated, and the imaginary part is represented by a three-dimensional vector, written in the following form:
[0007]
[0008] When the real part a is 0, it means that q is a pure quaternion.
[0009] In quaternions, the imaginary units i, j, and k have the following geometric meanings: i represents rotation from the positive X-axis to the positive Y-axis in the plane where the X and Y axes intersect; j represents rotation from the positive Z-axis to the positive X-axis in the plane where the Z and X axes intersect; and k represents rotation from the positive Y-axis to the positive Z-axis in the plane where the Y and Z axes intersect. Therefore, in three-dimensional space, quaternions can express the rotation of an object about any vector axis. Let the axis of rotation be a unit vector. If the rotation angle is θ, then the quaternion q can also be expressed in the following form:
[0010]
[0011] The above representation is also equivalent to:
[0012]
[0013] In quaternions, q * The conjugate of a quaternion, and the rotation opposite to that of the quaternion q, are represented as:
[0014]
[0015] The above representation is also equivalent to:
[0016]
[0017] When the quaternion q is a unit quaternion, i.e., ||q|| = 1, then the inverse of the quaternion q is... -1 It can be expressed by the following formula:
[0018] q -1 =q * / ||q||=q *
[0019] To address these issues, there is an urgent need for a method and technology that can more accurately identify user movements and correct posture, enabling users to exercise under professional guidance and reducing the probability of injury due to speed issues. Summary of the Invention
[0020] The purpose of this invention is to provide a quaternion-based three-dimensional human posture estimation and motion correction method. This method improves the accuracy of three-dimensional posture recognition by calculating angles and comparing skeletons, and improves the accuracy of speed by calculating the Euclidean distance between joints and the rate of angle change, thereby enhancing the reliability of user action recognition and posture correction. It also significantly reduces the impact of occlusion, environmental background and other issues on accuracy, thereby improving the accuracy of human posture recognition.
[0021] The technical solution of the present invention is as follows:
[0022] A quaternion-based three-dimensional human pose estimation and motion correction method includes the following steps:
[0023] Step 1: Collect exercise data. On the one hand, obtain video data of standard exercises through the official website of the National Student Physical Fitness and Health Center. On the other hand, collect video data through students using data collection devices, and then preprocess the videos.
[0024] Step 2: Use a convolutional neural network to extract human joint features, extract the coordinates and number of joints, and normalize the vectors formed by each joint and the central hip joint to solve the problems caused by different body shapes.
[0025] Step 3: Using the extracted joint features, construct quaternions, rotate the vectors formed by each joint and the central hip joint, and then recalculate the coordinates of the corresponding joints.
[0026] Step 4: Average the 3D coordinates of each joint before and after rotation to obtain the average 3D coordinates, minimizing the error with the true values. Using the average 3D coordinates and methods for calculating angles in 3D space, calculate the angles between joints and the skeleton for posture correction. Furthermore, using the average 3D coordinates, construct a curve showing the rate of change of joint distances and angles over time, and calculate the slope for velocity correction.
[0027] Furthermore, in step 1, the acquired video data should be processed and converted into images. Specifically, this includes the following:
[0028] Step 1.1: Students use mobile devices to shoot various types of sports videos from different angles.
[0029] Step 1.2: Extract the captured motion video frame by frame, and use a portion of the frames as training data for processing. Some frames suffer from insufficient data quantity and monotonous backgrounds. Data augmentation methods such as brightness adjustment, image rotation, and random cropping are used to address the insufficient data quantity. Images with various backgrounds are selected, and a blending algorithm is employed to blend the backgrounds and address the issue of monotonous backgrounds.
[0030] Furthermore, in step 2, a lightweight network is used to extract and normalize features from the preprocessed image. The specific steps are as follows:
[0031] Step 2.1: Input the preprocessed image into the lightweight network, with the central hip joint as the origin of the coordinate system, and output the features of each joint, i.e., the three-dimensional coordinates of each joint.
[0032] Step 2.2: Normalize the vectors formed by each joint point and the central hip joint point in each frame. Let x be the x-coordinate of the nth joint in the i-th frame. This is represented by the y-coordinate of the nth joint in the i-th frame. Let z be the z-coordinate of the nth joint in the i-th frame, and let n be the z-coordinate of the joint. in Represents the three-dimensional coordinates of the central hip joint point. Let represent the vector difference between the nth joint point and the central hip joint point in the i-th frame along the x, y, and z coordinates. Then, the normalization formula for the vector formed by the nth joint point and the central hip joint point in the i-th frame is:
[0033]
[0034] Where N represents the total number of key points, and I represents the number of frames in the video. Represents a positive integer.
[0035] Furthermore, in step 3, a quaternion rotation is constructed using the vectors formed by the joints of the normalized standard motion and student motion and the central hip joint point in each frame. The specific process includes the following steps:
[0036] Step 3.1: Taking the central hip joint as the origin, the student's head joint point H can be obtained from the normalized vector. stu (H x H y H z ) and foot joint point F stu (F x ,F y ,F z ), and H stu and F stu The resulting vector Will With the unit normal vector perpendicular to the horizontal plane passing through the center hip joint point Rotate θ degrees around the axis of rotation, thus aligning with the head joint H of the standard motion. coach (H' x ,H' y ,H' z ) and foot joint point F coach (H' x ,H' y ,H' z The vector formed Parallel, of which The rotation angle θ can be calculated using the following formula:
[0037] V'=qVq * =qVq -1
[0038] in V and V′ represent the vectors... and Constructed pure quaternions, Let q represent the unit normal vector perpendicular to the horizontal plane passing through the center hip joint, and let q represent the unit quaternion. *q represents the conjugate of a unit quaternion. -1 Denotes the inverse of a unit quaternion, in which q * =q -1 .
[0039] Step 3.2: Based on the calculated angle θ, the vectors formed by each joint point and the central hip joint point after rotation normalization are obtained. Since the central hip joint point is the central origin, the coordinates of each joint point after rotation can be obtained.
[0040] Furthermore, in step 4, the average value of the joint coordinates before and after rotation is taken to obtain optimized 3D coordinates. Then, the correctness of the movement is judged based on the skeleton and angles to improve accuracy; at the same time, speed correction is performed by calculating the slope. The specific process is as follows:
[0041] Step 4.1: Represent the average three-dimensional coordinates of the nth joint point in the i-th frame as follows: Then, between the key points K, L, M in the i-th frame and Angles can be expressed by the following formula:
[0042]
[0043] Step 4.2: Compare the rotated skeleton with the standard skeleton; in addition, based on the coordinates of each joint after rotation and different motion types, use the formula in Step 4.1 to calculate the angle of the vector formed between special joints and the joints, and compare it with the angle of the standard motion; combine the skeleton and the angle to evaluate the motion and reduce the impact of inaccurate recognition.
[0044] Step 4.3: Select the speed correction method based on the type of movement. If the Euclidean distance of the joints related to the standard movement changes significantly during the movement, then based on the positional information of these joints, the entire movement process is divided into multiple parts. Within the same part of the movement, speed correction is performed by observing the rate of change of the Euclidean distance of these related joints. These joints related to the movement are denoted as N. dis ={n i ,n j ,n k If , ...}, then under the same motion process, the Euclidean distance of a node n moving between time t and time T can be expressed by the following formula:
[0045]
[0046] The Euclidean distance of all joints related to the motion is expressed as: Starting from time t=0, the calculation is repeated at regular intervals. And draw The relationship curve between time and t varies, with different curves for different joints. Based on these curves, velocity correction is applied to each joint. If either the student's or coach's movement changes during the exercise, time t is reset to 0, and the calculation restarts. Then, based on the relationship curve of the Euclidean distance of each joint point changing with time, the slope K of the curve changing with time is calculated. dis Simultaneously, using the same method, the relationship curve of the Euclidean distance between the coach's joints and time was plotted, and the slope K of the curve as a function of time was calculated. dis-coach In the same motion process, under the same joint point change, the comparison is made at the same time interval. If |K dis |<|K dis-coach | indicates that the movement speed of this joint is too slow; if |K dis |>|K dis-coach | indicates that the movement speed of the joint is too fast.
[0047] Step 4.4: If the angle changes of the vectors formed between joints related to the standard movement are significant during the motion, then based on the angle information of these joint vectors, the entire motion process is divided into multiple parts. Within the same part of the motion process, speed correction is performed by observing the rate of change of the angles of the vectors formed between these specific joints. The angle between the relevant vectors is denoted as... Representing vectors with vector The included angle. During the same motion process, starting from time t=0, the angle θ is calculated at regular intervals. angle And draw the angle θ angle The relationship curve between θ and t varies depending on the angle between the vectors at different joint points. Based on these curves, velocity correction is applied to the angle between the vectors at each joint point. If either the student's or coach's movement changes during the exercise, time t is reset to 0, and θ is recalculated. angle Then, the slope K of the curve as a function of time is calculated based on the relationship curve. angle Simultaneously, using the same method, the curve showing the relationship between the vector angle formed by the coach's joint points and time was plotted, and the slope K of the curve as a function of time was calculated. angle-coach In the same motion process, under the same angle change, and compared over the same time period, if |K angle |<|K angle-coach | indicates that the angle change rate is too slow; if |K angle |>|K angle-coach | indicates that the angle is changing too quickly.
[0048] Compared with the prior art, the technical solution of the present invention is characterized by:
[0049] This invention proposes a quaternion-based three-dimensional human posture estimation and motion correction method. By using quaternion rotation, the accuracy of posture correction in three-dimensional space is improved. By considering the rate of change of Euclidean distance and angle, the method for determining the speed of motion in three-dimensional space is redefined. This invention can achieve excellent results in motion speed determination and motion recognition in three-dimensional human posture estimation and motion health. Attached Figure Description
[0050] Figure 1 This is a flowchart of the present invention. Detailed Implementation
[0051] The following is in conjunction with the appendix Figure 1 The present invention provides a detailed description of a quaternion-based three-dimensional human posture estimation and motion correction method, including the following steps:
[0052] Step 1: Collect exercise data. On the one hand, obtain video data of standard exercises through the official website of the National Student Physical Fitness and Health Center. On the other hand, collect video data through students using data collection devices, and then preprocess the videos.
[0053] Step 1.1: Students use mobile devices to shoot various types of sports videos from different angles.
[0054] Step 1.2: Extract frames from the acquired motion video, and use a subset of frames as training data for processing. Some frames suffer from insufficient data quantity and monotonous backgrounds. Data augmentation methods such as brightness adjustment, image rotation, and random cropping are used to address the insufficient data quantity. Multiple images with different backgrounds are selected, and the Mixup algorithm (but not limited to) is employed to blend images with different backgrounds pairwise, thereby resolving the issue of monotonous backgrounds.
[0055] Step 2: Use a convolutional neural network to extract human joint features, extract the location and number of joints, and normalize them to solve the problems caused by different body shapes.
[0056] Step 2.1: Input the preprocessed image into a lightweight network, but not limited to ShuffleNet, with the central hip joint as the origin of the coordinate system, and output the features of each joint, i.e., the three-dimensional coordinates of each joint.
[0057] Step 2.2: Normalize the vectors formed by each joint point and the central hip joint point in each frame. Let x be the x-coordinate of the nth joint in the i-th frame. This is represented by the y-coordinate of the nth joint in the i-th frame. Let z be the z-coordinate of the nth joint in the i-th frame, and let n be the z-coordinate of the joint. in Represents the three-dimensional coordinates of the central hip joint point. Let represent the vector difference between the nth joint point and the central hip joint point in the x, y, z coordinates of the i-th frame. Then, the normalization formula for the vector formed by the nth joint point and the central hip joint point in the i-th frame is:
[0058]
[0059] Where N represents all the key points, and I represents the number of frames in the video. Represents a positive integer.
[0060] Step 3: Construct quaternion rotations using the vectors formed by the joints and the central hip joint point of the normalized standard and student movements in each frame. The specific process includes the following steps:
[0061] Step 3.1: Taking the central hip joint as the origin, the student's head joint point H can be obtained from the normalized vector. stu (H x H y H z ) and foot joint point F stu (F x ,F y ,F z ), and H stu and F stu The resulting vector Will With the unit normal vector perpendicular to the horizontal plane passing through the center hip joint point Rotate θ degrees around the axis of rotation, thus aligning with the head joint H of the standard motion. coach (H' x ,H' y ,H' z ) and foot joint point F coach (H' x ,H' y ,H' z The vector formed Parallel, of which The rotation angle θ can be calculated using the following formula:
[0062] V'=qVq * =qVq -1
[0063] in V and V′ both represent the vector and Constructed pure quaternions, Let q represent the unit normal vector perpendicular to the horizontal plane passing through the center hip joint, and let q represent the unit quaternion. * q represents the conjugate of a unit quaternion. -1 Denotes the inverse of a unit quaternion, in which q * =q -1 The product of quaternions follows the distributive law of multiplication.
[0064] Step 3.2: Based on the calculated angle θ, the vectors formed by each joint point and the central hip joint point after rotation normalization are obtained. Since the central hip joint point is the central origin, the coordinates of each joint point after rotation can be obtained.
[0065] Step 4: Average the 3D coordinates of the joints before and after rotation to obtain the average 3D coordinates, minimizing the error with the true values. Calculate the angles between joints and the skeleton using the average 3D coordinates and methods for calculating angles in 3D space, for posture correction; simultaneously, calculate the slope for velocity correction.
[0066] Step 4.1: Represent the average three-dimensional coordinates of the nth joint point in the i-th frame as follows: Then, between the key points K, L, M in the i-th frame and Angles can be expressed by the following formula:
[0067]
[0068] Step 4.2: Compare the rotated skeleton with the standard skeleton; in addition, based on the coordinates of each joint after rotation and different motion types, use the formula in Step 4.1 to calculate the angle of the vector formed between special joints and the joints, and compare it with the angle of the standard motion; combine the skeleton and the angle to evaluate the motion and reduce the impact of inaccurate recognition.
[0069] Step 4.3: Select the speed correction method based on the type of movement. If the Euclidean distance of the joints related to the standard movement changes significantly during the movement, then based on the positional information of these joints, the entire movement process is divided into multiple parts. Within the same part of the movement, speed correction is performed by observing the rate of change of the Euclidean distance of these related joints. These joints related to the movement are denoted as N. dis ={n i ,n j ,n kIf , ...}, then under the same motion process, the Euclidean distance of a node n moving between time t and time T can be expressed by the following formula:
[0070]
[0071] The Euclidean distance of all joints related to the motion is expressed as: Starting from time t=0, the calculation is repeated at regular intervals. And draw The relationship curve between time and t varies, with different curves for different joints. Based on these curves, velocity correction is applied to each joint. If either the student's or coach's movement changes during the exercise, time t is reset to 0, and the calculation restarts. Then, based on the relationship curve of the Euclidean distance of each joint point changing with time, the slope K of the curve changing with time is calculated. dis Simultaneously, using the same method, the relationship curve of the Euclidean distance between the coach's joints and time was plotted, and the slope K of the curve as a function of time was calculated. dis-coach In the same motion process, under the same joint point change, the comparison is made at the same time interval. If |K dis |<|K dis-coach | indicates that the movement speed of this joint is too slow; if |K dis |>|K dis-coach | indicates that the movement speed of the joint is too fast.
[0072] Step 4.4: If the angle changes of the vectors formed between joints related to the standard movement are significant during the motion, then based on the angle information of these joint vectors, the entire motion process is divided into multiple parts. Within the same part of the motion process, speed correction is performed by observing the rate of change of the angles of the vectors formed between these specific joints. The angle between the relevant vectors is denoted as... Representing vectors with vector The included angle. During the same motion process, starting from time t=0, the angle θ is calculated at regular intervals. angle And draw the angle θ angle The relationship curve between θ and t varies depending on the angle between the vectors at different joint points. Based on these curves, velocity correction is applied to the angle between the vectors at each joint point. If either the student's or coach's movement changes during the exercise, time t is reset to 0, and θ is recalculated. angle Then, the slope K of the curve as a function of time is calculated based on the relationship curve. angleSimultaneously, using the same method, the curve showing the relationship between the vector angle formed by the coach's joint points and time was plotted, and the slope K of the curve as a function of time was calculated. angle-coach In the same motion process, under the same angle change, and compared over the same time period, if |K angle |<|K angle-coach | indicates that the angle change rate is too slow; if |K angle |>|K angle-coach | indicates that the angle is changing too quickly.
[0073] This invention proposes a quaternion-based three-dimensional human posture estimation and motion correction method. By using quaternion rotation, the accuracy of posture correction in three-dimensional space is improved. By considering the rate of change of Euclidean distance and angle, the method for determining the speed of motion in three-dimensional space is redefined. This invention can achieve excellent results in motion speed determination and motion recognition in three-dimensional human posture estimation and motion health.
Claims
1. A three-dimensional human posture estimation and motion correction method based on quaternions, characterized in that, The specific steps are as follows: Step 1: Collect exercise data. Obtain video data of standard exercises from the official website of the National Student Physical Fitness and Health Center, as well as video data collected by students using acquisition devices. Then, preprocess the video data collected by the acquisition devices. Step 2: Use a convolutional neural network to extract human joint features, extract the coordinates and number of joints, and normalize the vectors formed by each joint and the central hip joint. Step 3: Using the extracted joint features, construct quaternions, rotate the vectors formed by each joint and the central hip joint, and then recalculate the coordinates of the corresponding joints. Step 3, which uses joint features to construct quaternions and rotates the vectors formed by each joint and the central hip joint, specifically includes: constructing quaternions using the vectors formed by each joint and the central hip joint in the standard and student movements, and rotating the vectors. The steps of constructing quaternions from vectors formed by the joints of the standard movement and the student movement and the central hip joint, and then rotating the vectors, include: 3.1: Using the central hip joint as the origin, the student's head joint point is obtained from the normalized vector. With foot joints ,as well as and The resulting vector ,Will With the unit normal vector perpendicular to the horizontal plane passing through the center hip joint point Rotate θ degrees along the axis of rotation, thus aligning with the head joint of the standard movement. With foot joints The resulting vector Parallel, of which The rotation angle θ is calculated using the following formula: in and Indicates based on vector and Constructed pure quaternions, Let q represent the unit normal vector perpendicular to the horizontal plane passing through the central hip joint point, and let q represent the unit quaternion. Describes the conjugate of a unit quaternion. Represents the inverse of a unit quaternion, in which... ; 3.2: Based on the calculated rotation angle θ, the vector formed by each joint point and the central hip joint point after rotation normalization is obtained. Since the central hip joint point is the central origin, the coordinates of each joint point after rotation are obtained. Step 4: Take the average of the three-dimensional coordinates of each joint before and after rotation to obtain the average three-dimensional coordinates to minimize the error with the true value; use the average three-dimensional coordinates and the method of calculating angles in three-dimensional space to calculate the angles between each joint and the skeleton for posture correction; use the average three-dimensional coordinates to construct a curve of the rate of change of joint distance and angle over time, and use the slope to perform speed correction.
2. The quaternion-based three-dimensional human posture estimation and motion correction method as described in claim 1, characterized in that: Step 1 involves acquiring video data using a capture device, followed by preprocessing of the acquired video data. The steps include: 1.1: Students used mobile devices to shoot various types of sports videos from different angles; 1.2: The acquired motion video is extracted frame by frame, and the frame images are extracted as training data for processing; data augmentation methods such as brightness adjustment, image rotation, and random cropping are used to solve the problem of insufficient data volume; images with multiple different backgrounds are selected, and a blending algorithm is used to blend the backgrounds to solve the problem of single background.
3. The quaternion-based three-dimensional human posture estimation and motion correction method as described in claim 1, characterized in that: Step 2 uses a lightweight convolutional neural network to extract features from the image obtained after preprocessing in Step 1 and performs vector normalization. The steps include: 2.1: Input the preprocessed image into a lightweight convolutional neural network, with the central hip joint as the origin of the coordinate system, and output the features of each joint, i.e., the three-dimensional coordinates of each joint. 2.2: Normalize the vectors formed by each joint point and the central hip joint point in each frame; Let x be the x-coordinate of the nth joint in the i-th frame. This is represented by the y-coordinate of the nth joint in the i-th frame. Let z be the z-coordinate of the nth joint in the i-th frame, and let n be the z-coordinate of the joint. ,in , , , Represents the three-dimensional coordinates of the central hip joint point. Let represent the difference vector between the nth joint point and the central hip joint point in the x, y, z coordinates of the i-th frame; then the normalization formula for the vector formed by the nth joint point and the central hip joint point in the i-th frame is: Where N represents the total number of key points, and I represents the number of frames in the video. Represents a positive integer.
4. The quaternion-based three-dimensional human posture estimation and motion correction method as described in claim 1, characterized in that: step 4 corrects posture through the skeleton and angles; simultaneously, it corrects velocity by calculating the slope, the steps of which include: 4.1: The average three-dimensional coordinates of the nth joint point in the i-th frame are expressed as... Then, between the key points K, L, M in the i-th frame and Angles are expressed by the following formula: 4.2: Compare the rotated skeleton with the standard skeleton; based on the coordinates of each joint after rotation and different motion types, calculate the angle of the vector formed between the joints using the formula in step 4.1, and compare it with the angle of the standard motion; combine the skeleton and the angle to evaluate the motion and reduce the impact of inaccurate recognition. 4.3: Select the speed correction method according to the type of movement; if the Euclidean distance of the joints related to the standard movement changes significantly during the movement, then based on the positional information of these joints, divide the entire movement into multiple parts. Under the same partial movement, use the rate of change of the Euclidean distance of these related joints for speed correction. These joints related to the movement are denoted as... Then, under the same motion process, a node n from time t to time t The Euclidean distance between the two points is expressed by the following formula: The Euclidean distance of all joints related to the motion is expressed as: Starting from time t=0, the calculation is recalculated at regular intervals. and draw The relationship curve between t and t varies, with different curves for different joints. Based on these curves, velocity correction is applied to each joint. If either the student's or coach's movement changes during the exercise, the time t is reset to 0, and the calculation restarts. Then, based on the relationship curve of the Euclidean distance of each joint point changing with time, the slope of the curve changing with time is calculated. Simultaneously, using the same method, the relationship curve of the Euclidean distance between the coach's joints and time was plotted, and the slope of the curve as a function of time was calculated. ; Under the same motion process and the same joint point change, comparisons are made over the same amount of time. This indicates that the movement speed of the joint is too slow; if This indicates that the joint is moving too fast; 4.4: If the angle of the vector formed between the joints related to the standard movement changes significantly during the movement, the entire movement process is divided into multiple parts based on the angle information of the vector formed by these joints during the movement process. Under the same part of the movement process, the speed is corrected by using the rate of change of the angle of the vector formed between these special joints. Let the angle between the relevant vectors be denoted as , Representing vectors with vector The included angle; during the same motion process, starting from time t=0, the angle is calculated at regular intervals. And draw the angle. The relationship curve between t and the vector angle at different joint points has different relationship curves. Based on their respective relationship curves, the velocity correction is performed on the vector angle at each joint point. If either the student's or coach's movement changes during the exercise, the time t is reset to 0, and the calculation restarts. Then, the slope of the curve as a function of time is calculated based on the relationship curve. Simultaneously, using the same method, the curve showing the relationship between the vector angle formed by the coach's joint points and time was plotted, and the slope of the curve as a function of time was calculated. ; In the same motion process, under the same angle change, and compared over the same time period, if This indicates that the angle is changing too slowly; if This indicates that the angle is changing too rapidly.