Video follow-up exercise evaluation and exercise intensity monitoring method and system

Abstract

The invention discloses a video follow-up exercise evaluation and exercise intensity monitoring method and system. The method comprises the following steps: acquiring basic information of a user; collecting the heart rate, expression, breathing and sweating conditions of the user during exercise; collecting a user motion video; obtaining a standard motion video; extracting a user action model of the user motion video and a standard action model in the standard action video; scoring the user action standard degree according to the relationship between the overall similarity and a threshold value; and performing real-time exercise intensity reminding on the user according to a comparison result of the real-time exercise intensity of the user and the exercise intensity threshold range. According to the method, comprehensive judgment is carried out by combining parameters such as heart rate, breath, expression and sweating conditions, so that the result is more accurate. In addition, in the setting of the exercise intensity threshold value, parameters such as age, gender, body weight and subjective willingness of the user are integrated, so that the alarm is more referential.

Description

Technical Field

[0001] The present invention belongs to the technical field of sports monitoring, and in particular relates to a video training evaluation and sports intensity monitoring method and system. Background Art

[0002] In recent years, exercise has become increasingly important for people's health. However, many people have relatively weak knowledge of how to exercise properly. They may just blindly follow the video to exercise, but fail to realize whether their movements are correct and standard, and lack awareness of their own abilities and limitations. This practice often leads to incorrect movements, or causes exercise to reach an intensity that is not suitable for oneself, and ultimately does not achieve a good exercise effect, but instead causes damage to the body.

[0003] With the rapid development of computer technology, people can easily find various digital exercise tutorials on the Internet. However, these tutorials can usually only understand and experience their own movements through the user's own imagination and proprioception, which is not accurate enough for most people. Through the teaching video on the screen, you can only see the coach's action demonstration, but you can't really feel the position, posture and accuracy of your body. When pursuing the correct way and technique of movement, it is easy to make errors based on spatial imagination and proprioception alone. And it is difficult for people who don't exercise for a long time to find the right exercise intensity through proprioception. So without correct feedback from the outside world, our movements may be inaccurate, our postures are not standardized, or the exercise intensity is unreasonable, which makes it impossible to exercise effectively. Summary of the invention

[0004] In view of this, the present invention provides a video follow-up evaluation and exercise intensity monitoring method and system, which can not only evaluate the follow-up action according to the action similarity, but also monitor the user's exercise intensity in real time to avoid excessive exercise intensity or insufficient exercise intensity.

[0005] In order to solve the above technical problems, the technical solution of the present invention is to adopt a video training evaluation and exercise intensity monitoring method, including: Obtaining basic user information, including age, gender, weight, exercise method, and expected exercise intensity; Collect the user's heart rate, facial expression, breathing and sweating during exercise; Collecting a user motion video, wherein the user motion video is a full-body video of the user during motion; acquiring a standard motion video, wherein the standard motion video is a standard motion video of the user performing the motion; Extracting a user motion model from a user motion video and a standard motion model from a standard motion video; Calculate the similarity of each joint angle in the user action model and the standard action model to obtain the overall similarity, and score the user action standard according to the relationship between the overall similarity and the threshold; The exercise intensity threshold range is set according to the user's basic information, and the user's real-time exercise intensity is obtained according to the user's heart rate, facial expression, breathing and sweating during exercise; the user is given a real-time exercise intensity reminder based on the comparison result between the user's real-time exercise intensity and the exercise intensity threshold range.

[0006] As an improvement, the method for extracting a user motion model from a user motion video includes: Detect the face part in the video and obtain the bounding box of the face; Preliminarily estimate the primary key points of the human body based on the face boundary; Segment the human body from the video based on the first-level key points; Accurately estimate the segmented human body to obtain the secondary key points of the human body; Perform two-dimensional reconstruction of human body posture based on primary key points and secondary key points; The user's three-dimensional motion model is obtained based on the key points after two-dimensional reconstruction and the depth information of the human body.

[0007] As a further improvement, the method for calculating the similarity between the user action model and the joint angles in the standard action model includes: Proportionalize the user motion model and the standard motion model; The user's joint angle is calculated by the Euclidean distance between each joint and the adjacent joint point in the user's motion model and the Euclidean distance between the end points of the two limbs of the joint; The standard joint angle is calculated by the Euclidean distance between each joint and the adjacent joint point in the standard action model and the Euclidean distance between the end points of the two limbs of the joint; Calculate the similarity between the user's joint angle and the corresponding standard joint angle; The weighted average of all obtained joint similarities is performed to obtain the overall similarity.

[0008] As another further improvement, the method for obtaining the user's real-time exercise intensity based on the user's heart rate, facial expression, breathing and sweating during exercise is: calculate the heart rate exercise intensity, facial expression exercise intensity, breathing exercise intensity and sweating exercise intensity respectively according to the user's heart rate, facial expression, breathing and sweating during exercise, and perform weighted calculation of the heart rate exercise intensity, facial expression exercise intensity, breathing exercise intensity and sweating exercise intensity to obtain the user's real-time exercise intensity.

[0009] As an improvement, the method for calculating the heart rate exercise intensity according to the user's heart rate during exercise includes: Using the formula HRmax = 216.6-0.84*age (years) Heart rate exercise intensity = real-time heart rate / HRmax Calculate heart rate exercise intensity, where HRmax is the maximum heart rate.

[0010] As an improvement, the method for calculating the intensity of the expression movement according to the user's expression includes: Decompose the user's motion video frame by frame into images and grayscale them, and calculate the gradient of each pixel in the image; Convert the image into a histogram of oriented gradients to extract image features and thus obtain the face position; Locate the feature points of the face and calculate the real-time expression based on the position change of the feature points; The real-time expression is input into the trained expression-motion intensity model to obtain the expression motion intensity.

[0011] As an improvement, the method for calculating the intensity of the breathing exercise according to the user's breathing includes: While calculating the real-time expression based on the position transformation of the feature points, the number of times the lips open and close is recorded; The corresponding breathing exercise intensity is obtained by the number of times the lips open and close.

[0012] As an improvement, the method for calculating the sweating exercise intensity according to the sweating condition of the user includes: After locating the feature points of the human face, selecting the facial sweat mark retention areas; the sweat mark retention areas include the forehead, nose tip, chin, and cheeks; The facial sweat mark retention areas are input into the trained sweat mark-exercise intensity model to obtain the sweating exercise intensity.

[0013] As an improvement, a method for weighting the heart rate exercise intensity, facial expression exercise intensity, breathing exercise intensity, and sweating exercise intensity to obtain the user's real-time exercise intensity includes: Real-time exercise intensity = 0.5*heart rate exercise intensity + 0.2*expression exercise intensity + 0.2*breathing exercise intensity + 0.1*sweating exercise intensity.

[0014] The present invention is beneficial in that: The present invention provides a video follow-up evaluation system, which detects the user's sports movements and compares them with standard movements to obtain the movement familiarity, so as to promptly remind the user to adjust the movement posture, so that the user can master the correct movement, reduce the risk of injury, and improve the exercise effect.

[0015] In addition, the present invention provides a system for real-time monitoring of exercise intensity during exercise, real-time feedback of exercise intensity, reduction of invalid exercise rate, and reduction of the risk of excessive exercise. In the monitoring of exercise intensity, the present invention combines parameters such as heart rate, breathing, expression, and sweating to make a comprehensive judgment, so that the result is more accurate. And in the setting of the exercise intensity threshold, the present invention combines parameters such as age, gender, weight, and user's subjective will, so that its alarm is more referenceable. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the present invention.

[0017] Figure 2 It is a structural principle diagram of the present invention. Implementation

[0018] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods.

[0019] Prior art Chinese patent application CN202310243710.6 discloses a mobile terminal design method for lightweight recognition of Tai Chi movements, which belongs to the field of human posture recognition technology. The method includes: lightweight model training, video analysis, image preprocessing, key point confidence acquisition, joint angle distance classification action, program function design, and mobile terminal deployment. The lightweight model training uses the COCO data set and the standard Tai Chi movement data set; the video analysis extracts key frames through motion analysis, obtains the iconic movements of each Tai Chi posture or directly performs video recognition, and evaluates the overall situation. Tai Chi movements are judged by the dual standards of angle and distance, the angle and distance errors are calculated, and probabilistic actions are given. The best match is selected from the standard actions as the judgment action after comprehensive consideration, and the calculation results are fed back to give the recognition score.

[0020] It can be seen that the above-mentioned prior art can only evaluate the accuracy of the user's movements, and cannot monitor the user's real-time exercise intensity in real time. There are also some prior arts such as CN202210337706.1 that disclose a physical labor fatigue early warning system and method, which uses the warning value discrimination rule to process the collected heart rate information of the worker to obtain the first labor intensity discrimination result, uses the worker's heart rate intensity discrimination rule to process the collected heart rate information of the worker to obtain the second labor intensity discrimination result, and uses the labor degree assessment rule to jointly process the first labor intensity discrimination result and the second labor intensity discrimination result to obtain labor fatigue early warning information. The prior art uses heart rate monitoring intensity to give an early warning when the intensity is too large, and does not take into account that the simple heart rate cannot accurately monitor the real exercise intensity of each person. For example, the elderly usually reach their real limit, but the heart rate shows a safe intensity, which is greatly unfavorable for monitoring and protecting non-healthy or special populations, and after measuring some intensity data with certain reference value, the intensity is not fed back in real time, which is not convenient for users to adjust the intensity and state according to their own proprioception.

[0021] In addition, although heart rate is a key indicator of exercise intensity, it is not comprehensive and there are still some biases in judgment.

[0022] like Figure 1 As shown, the present invention provides a video follow-up evaluation and exercise intensity monitoring method, which is not only used for follow-up evaluation but also for real-time exercise intensity monitoring, and its specific steps include: S1 obtains basic user information, which includes age, gender, weight, exercise method, and expected exercise intensity.

[0023] The differences in the user's basic information, including age, gender, weight, etc., directly determine the differences in exercise intensity tolerance. Of course, due to individual differences, even two people with completely identical basic information will still have differences. Therefore, the present invention also collects the user's own subjective exercise intensity information, such as exercise method, expected exercise intensity, etc., and comprehensively specifies an exercise intensity detection index through the above information.

[0024] S2 collects the user's heart rate, facial expressions, breathing and sweating during exercise.

[0025] The user's heart rate can be collected through peripherals such as sports watches. The collection methods of facial expressions, breathing and sweating will be explained in detail later and will not be repeated here. Heart rate, facial expressions, breathing and sweating are used to determine the user's real-time exercise intensity.

[0026] S3 collects a user motion video, which is a full-body video of the user during motion; and obtains a standard motion video, which is a standard action video of the user during motion.

[0027] Standard sports videos are teaching videos, which can be built-in or downloaded from the Internet. There is no restriction on the source. The purpose of standard sports videos is to serve as a template for evaluation. For example, if a user wants to do aerobics, then a teaching video of aerobics can be found according to the user's needs.

[0028] The user's exercise video needs to be collected in real time, that is, it needs to be filmed by a camera while the user is exercising. When filming, it is important to capture the user's entire body, which is convenient for later processing.

[0029] S4 extracts the user motion model in the user motion video and the standard motion model in the standard motion video.

[0030] It is impossible to compare directly with the video, so it is necessary to extract the action model in the video. In this embodiment, the BlazePose model is used to extract the action model, and the specific steps are as follows: S41 detects the face part in the video and obtains the bounding box of the face.

[0031] In this embodiment, the BlazeFace model is used to detect and select the face part of the user action video.

[0032] S42 preliminarily estimates the primary key points of the human body based on the facial boundaries. The so-called primary key points refer to the key points of a person that are relatively obvious and easy to identify, such as the neck, shoulders, elbows, knees, etc.

[0033] S43 segments the human body from the video according to the primary key points; after identifying the primary key points, a rough human body model can be segmented out. The human body is segmented out from the original image so as to estimate the posture of the human body more accurately.

[0034] S44 accurately estimates the segmented human body to obtain the secondary key points of the human body; the secondary key points are some joints that are more difficult to identify and whose range of motion is generally not too large, such as hip joints, ankle joints, etc.

[0035] S45 reconstructs the human body posture in two dimensions according to the primary key points and the secondary key points, and can restore the human body posture on a plane through the two-dimensional reconstruction.

[0036] S46 obtains a three-dimensional motion model of the user according to the two-dimensionally reorganized key points and the human body depth information.

[0037] The standard action model in the standard action video of the present invention can also be extracted in the above-mentioned manner. Of course, if the standard action video is preset, it can be extracted in advance to reduce the real-time overhead of the system.

[0038] S5 calculates the similarity between the angles of each trunk joint in the user action model and the standard action model to obtain the overall similarity, and scores the user action standard according to the relationship between the overall similarity and the threshold.

[0039] Specifically, it includes the following steps: S51 The user motion model and the standard motion model are proportional, with the aim of making the two equal in size to avoid errors caused by the proportions.

[0040] S52 calculates the user's joint angles through the Euclidean distances between each joint and adjacent joint points in the user's motion model and the Euclidean distances between the end points of two limb segments of the joint.

[0041] The formula for calculating the Euclidean distance between two points is:

[0042] Where d is the Euclidean distance, x1, y1, z1 and x2, y2, z2 are the spatial coordinates of the two points respectively.

[0043] S53 calculates the standard joint angle through the Euclidean distance between each joint and the adjacent joint point in the standard action model and the Euclidean distance between the end points of the two limbs of the joint.

[0044] The calculation method in this step is the same as that of the user motion model. If the standard action video is preset, the calculation here can also be performed in advance.

[0045] S54 calculates the similarity between the user's joint angle and the corresponding standard joint angle.

[0046] By calculating the similarity, we can obtain the similarity between each joint angle and the joint angle in the corresponding standard action model. The higher the similarity, the higher the similarity.

[0047] S55 performs weighted averaging on all obtained joint similarities to obtain the overall similarity.

[0048] The similarity of each joint has a different impact on the overall movement. For example, the knee and elbow joints have a greater impact on the overall movement similarity than the wrist and ankle joints. Therefore, it is necessary to perform weighted calculations on different joints to obtain the final similarity.

[0049] In fact, if the video is preset in advance, the weighting method can also set different weighting formulas according to different actions, so that the final overall similarity is more accurate.

[0050] After obtaining the overall similarity, you can score and provide feedback based on the result of comparing the similarity with the threshold. You can use a more professional 100-point score, or you can use a more vague grading system such as "average", "very good", "excellent", etc.

[0051] S6 sets the exercise intensity threshold range according to the user's basic information, and obtains the user's real-time exercise intensity according to the user's heart rate, expression, breathing and sweating during exercise; and provides the user with real-time exercise intensity reminders based on the comparison results between the user's real-time exercise intensity and the exercise intensity threshold range.

[0052] According to the age, gender, weight, exercise mode, and expected exercise intensity collected in step S1, a targeted exercise intensity threshold range is generated. It can be understood that for older and heavier users, their alarm thresholds are definitely lower than those for younger and lighter users.

[0053] In this embodiment, the heart rate exercise intensity, facial expression exercise intensity, breathing exercise intensity, and sweating exercise intensity are calculated respectively according to the user's heart rate, facial expression, breathing and sweating during exercise, and the heart rate exercise intensity, facial expression exercise intensity, breathing exercise intensity, and sweating exercise intensity are weightedly calculated to obtain the user's real-time exercise intensity.

[0054] S61 The method for calculating the heart rate exercise intensity according to the heart rate of the user during exercise includes: Using the formula HRmax = 216.6-0.84*age (years) Heart rate exercise intensity = real-time heart rate / HRmax Calculate heart rate exercise intensity, where HRmax is the maximum heart rate in beats per minute.

[0055] Facial expressions can well reflect a person's exercise intensity. When the exercise intensity reaches a certain level, the facial expressions will also show corresponding characteristics. We can observe these characteristics to preliminarily evaluate the exercise intensity at this time. For example, during high-intensity exercise, athletes often have expressions such as frowning, firm eyes, closed mouth or slight clenching of teeth, indicating that the athlete is bearing a large physical load and is going all out to concentrate on high-intensity challenges. If it is moderate-intensity exercise, the athlete generally shows characteristics such as slightly wrinkled eyebrows, energetic eyes, and slightly raised corners of the mouth, indicating that the intensity is relatively comfortable at this moment and the athlete is in an active state of exercise. If the intensity is low at this moment, the athlete often has a relaxed eyebrow, a smile or a raised corner of the mouth, indicating that the low intensity of exercise means that the athlete feels relaxed and even enjoys exercise. Therefore, facial expressions can be used to evaluate the intensity of exercise.

[0056] S62 The method for calculating the intensity of the expression movement according to the user's expression includes: S621 decomposing the user's movement video into images frame by frame and graying them, and calculating the gradient of each pixel in the image; S622 converts the image into a histogram of directional gradients to extract image features, thereby obtaining the face position. In this embodiment, facial expression recognition can be implemented by referring to the GitHub project face_recognition.

[0057] S623 locates the feature points of the face and calculates the real-time expression according to the position transformation of the feature points; specifically, the Dlib facial feature recognition model is used to locate the 68 feature points of the face.

[0058] S624 inputs the real-time expression into the trained expression-motion intensity model to obtain the expression motion intensity.

[0059] The expression-exercise intensity model is actually a classification model. It only needs to record the changes in the key features of each expression during exercise of different intensities as a training set to train the model. After the model training is completed, the recognized expression is input into the model to obtain the corresponding exercise intensity.

[0060] Breathing patterns are a good reflection of exercise intensity. In low-intensity exercise, especially when running in cold weather, it is usually recommended to inhale through the nose because the nasal hair can block germs and the nasal cavity can warm the air, making breathing smoother. However, when the intensity of exercise increases and reaches 3 / 4 of the maximum intensity, breathing only through the nose may limit the amount of oxygen absorbed. At this time, mouth breathing is needed to assist in meeting the body's demand for oxygen. In high-intensity exercise, breathing through the mouth and nose at the same time is a common phenomenon. Therefore, the breathing method can be judged by the micro-movements of the face and mouth during exercise, and the intensity of exercise can be evaluated by the breathing method and the speed of the regular amplitude changes of the lips (ventilation rate).

[0061] S63 The method for calculating the breathing exercise intensity according to the user's breathing includes: S631 calculates the real-time expression based on the position transformation of the feature points and records the number of times the lips open and close.

[0062] In this step, the facial key point recognition method in step S62 is used. At this time, attention is paid to the changes in the mouth key points. For example, when the Euclidean distance of the upper and lower lip key points is greater than the mouth closure threshold (the distance between the key points when the mouth is closed) and the regularity (breathing rhythm) amplitude changes, it is mouth breathing.

[0063] S632 obtains the corresponding breathing exercise intensity according to the number of times the lips open and close. For example, when at rest, the breathing frequency is about 12 to 18 times per minute, while during high-intensity exercise, the breathing frequency may increase to 40 to 60 times per minute.

[0064] The secretion of body fluids reflects the intensity of exercise more accurately. A large amount of heat is generated in the body during exercise. The greater the intensity, the higher the heat. The body needs to eliminate excess heat through sweating, so the intensity of exercise can be evaluated more carefully through sweat secretion. During low-intensity exercise, athletes generally show slight sweat, little sweat or no sweat. During moderate-intensity exercise, athletes begin to have obvious sweat beads appearing on the forehead, back of the neck and other parts, but they do not flow out in large quantities, and the skin is slightly moist at this time. During high-intensity exercise, athletes sweat profusely, with a large number of sweat beads rolling down the face, neck, back and other parts of the body.

[0065] The method for calculating the sweating exercise intensity according to the sweating condition of the user in S64 includes: S641 After locating the feature points of the face, select the facial sweat retention areas; the sweat retention areas include the forehead, nose tip, chin, and cheeks; in this step, the facial key point recognition method in step S62 is also used to identify the sweat retention areas. The sweat secretion in the sweat retention areas is more obvious and easier to identify. In fact, the sweat secretion in the back and other parts of the body is also relatively strong, but considering the influence of clothing, the sweat secretion in the back and other parts of the body is not considered in this implementation.

[0066] S642 inputs the facial sweat mark retention area into the trained sweat mark-exercise intensity model to obtain the sweating exercise intensity.

[0067] They are identified through convolutional neural networks (CNN). CNN contains three main types of neural layers, namely convolutional layers, pooling layers and fully connected layers. Each type of layer plays a different role. Each layer converts the input volume of neuron activation into output volume, and finally forms the final fully connected layers, thereby mapping the input data into a one-dimensional feature vector. Then, the skin sweat characteristics are identified through training with a large number of pictures of sweat on the skin. After the user video data is input, the same method is used to extract features. By comparison, the user's sweating degree at the moment is identified, which is used to evaluate the intensity of exercise at that time.

[0068] The method for S65 to obtain the user's real-time exercise intensity by weighted calculation of the heart rate exercise intensity, the expression exercise intensity, the breathing exercise intensity, and the sweating exercise intensity includes: Real-time exercise intensity = 0.5*heart rate exercise intensity + 0.2*expression exercise intensity + 0.2*breathing exercise intensity + 0.1*sweating exercise intensity.

[0069] Heart rate is still the most critical indicator for detecting exercise intensity, so it has the highest weight, followed by facial expression and breathing. Sweating varies greatly from person to person, so its weight is lower.

[0070] In addition, if Figure 2As shown, the present invention also provides a video training evaluation and exercise intensity monitoring system, which is characterized by comprising: A basic information acquisition module is used to acquire basic user information, including age, gender, weight, exercise method, and expected exercise intensity; The exercise information acquisition module is used to collect the user's heart rate, facial expression, breathing and sweating during exercise; A sports video acquisition module is used to collect a user sports video, which is a full-body video of the user during exercise; and to acquire a standard sports video, which is a standard action video of the user during exercise; An action model extraction module, used to extract a user action model from a user motion video and a standard action model from a standard action video; An evaluation module is used to calculate the similarity between the angles of each dry joint in the user action model and the standard action model to obtain the overall similarity, and to score the user action standard according to the relationship between the overall similarity and the threshold; The exercise intensity monitoring module is used to set the exercise intensity threshold range according to the user's basic information, and obtain the user's real-time exercise intensity according to the user's heart rate, expression, breathing and sweating during exercise; and provide real-time exercise intensity reminders to the user based on the comparison results between the user's real-time exercise intensity and the exercise intensity threshold range.

[0071] The above are only preferred embodiments of the present invention. It should be noted that the above preferred embodiments should not be regarded as limiting the present invention, and the protection scope of the present invention should be based on the scope defined by the claims. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the spirit and scope of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A video training evaluation and exercise intensity monitoring method, characterized in that include: Obtaining basic user information, including age, gender, weight, exercise method, and expected exercise intensity; Collect the user's heart rate, facial expression, breathing and sweating during exercise; Collecting a user motion video, wherein the user motion video is a full-body video of the user while exercising; Obtaining a standard motion video, where the standard motion video is a standard motion video of the motion performed by the user; Extracting a user motion model from a user motion video and a standard motion model from a standard motion video; Calculate the similarity of each joint angle in the user action model and the standard action model to obtain the overall similarity, and score the user action standard according to the relationship between the overall similarity and the threshold; Set the exercise intensity threshold range based on the user's basic information, and obtain the user's real-time exercise intensity based on the user's heart rate, facial expression, breathing and sweating during exercise; The user is given a real-time exercise intensity reminder based on the comparison result between the user's real-time exercise intensity and the exercise intensity threshold range.

2. A video training evaluation and exercise intensity monitoring method according to claim 1, characterized in that The method for extracting a user motion model from a user motion video comprises: Detect the face part in the video and obtain the bounding box of the face; Preliminarily estimate the primary key points of the human body based on the face boundary; Segment the human body from the video based on the first-level key points; Accurately estimate the segmented human body to obtain the secondary key points of the human body; Perform two-dimensional reconstruction of human body posture based on primary key points and secondary key points; The user's three-dimensional motion model is obtained based on the key points after two-dimensional reconstruction and the depth information of the human body.

3. A video training evaluation and exercise intensity monitoring method according to claim 1, characterized in that The method for calculating the similarity between the user action model and each joint angle in the standard action model includes: Proportionalize the user motion model and the standard motion model; The user's joint angle is calculated by the Euclidean distance between each joint and the adjacent joint point in the user's motion model and the Euclidean distance between the end points of the two limbs of the joint; The standard joint angle is calculated by the Euclidean distance between each joint and the adjacent joint point in the standard action model and the Euclidean distance between the end points of the two limbs of the joint; Calculate the similarity between the user's joint angle and the corresponding standard joint angle; The weighted average of all obtained joint similarities is performed to obtain the overall similarity.

4. A video training evaluation and exercise intensity monitoring method according to claim 1, characterized in that The method for obtaining the user's real-time exercise intensity based on the user's heart rate, facial expression, breathing and sweating during exercise is: calculate the heart rate exercise intensity, facial expression exercise intensity, breathing exercise intensity and sweating exercise intensity respectively according to the user's heart rate, facial expression, breathing and sweating during exercise, and perform weighted calculation on the heart rate exercise intensity, facial expression exercise intensity, breathing exercise intensity and sweating exercise intensity to obtain the user's real-time exercise intensity.

5. A video training evaluation and exercise intensity monitoring method according to claim 4, characterized in that Methods for calculating heart rate exercise intensity based on the user's heart rate during exercise include: Using the formula HRmax = 216.6-0.84*age (years) Heart rate exercise intensity = real-time heart rate / HRmax Calculate heart rate exercise intensity, where HRmax is the maximum heart rate in beats per minute.

6. A video training evaluation and exercise intensity monitoring method according to claim 4, characterized in that Methods for calculating the intensity of expression movement based on user expressions include: Decompose the user's motion video frame by frame into images and grayscale them, and calculate the gradient of each pixel in the image; Convert the image into a histogram of oriented gradients to extract image features and thus obtain the face position; Locate the feature points of the face and calculate the real-time expression based on the position change of the feature points; The real-time expression is input into the trained expression-motion intensity model to obtain the expression motion intensity.

7. A video training evaluation and exercise intensity monitoring method according to claim 5, characterized in that Methods for calculating the intensity of respiratory exercise based on the user's breathing include: While calculating the real-time expression based on the position transformation of the feature points, the number of times the lips open and close is recorded; The corresponding breathing exercise intensity is obtained by the number of times the lips open and close.

8. The video training evaluation and exercise intensity monitoring method according to claim 5, characterized in that Methods for calculating sweating exercise intensity based on the user's sweating include: After locating the feature points of the human face, selecting the facial sweat mark retention areas; the sweat mark retention areas include the forehead, nose tip, chin, and cheeks; The facial sweat mark retention areas are input into the trained sweat mark-exercise intensity model to obtain the sweating exercise intensity.

9. A video training evaluation and exercise intensity monitoring method according to claim 4, characterized in that The method of weighting the heart rate exercise intensity, the expression exercise intensity, the breathing exercise intensity, and the sweating exercise intensity to obtain the user's real-time exercise intensity includes: Real-time exercise intensity = 0.5*heart rate exercise intensity + 0.2*expression exercise intensity + 0.2*breathing exercise intensity + 0.1*sweating exercise intensity.

10. A video training evaluation and exercise intensity monitoring system, characterized in that include: A basic information acquisition module is used to acquire basic user information, including age, gender, weight, exercise method, and expected exercise intensity; The exercise information acquisition module is used to collect the user's heart rate, facial expression, breathing and sweating during exercise; A motion video acquisition module, used to collect a user motion video, wherein the user motion video is a full-body video of the user while exercising; Obtaining a standard motion video, where the standard motion video is a standard motion video of the motion performed by the user; An action model extraction module, used to extract a user action model from a user motion video and a standard action model from a standard action video; An evaluation module is used to calculate the similarity between the angles of each dry joint in the user action model and the standard action model to obtain the overall similarity, and to score the user action standard according to the relationship between the overall similarity and the threshold; The exercise intensity monitoring module is used to set the exercise intensity threshold range according to the user's basic information, and obtain the user's real-time exercise intensity according to the user's heart rate, expression, breathing and sweating during exercise; and provide real-time exercise intensity reminders to the user based on the comparison results of the user's real-time exercise intensity and the exercise intensity threshold range.

Images