Exercise movement evaluation system and method based on posture analysis

Abstract

The present invention discloses an exercise action evaluation system and method based on posture analysis. The action evaluation system includes a user information database, an exercise database, a face acquisition module, a face recognition module, an exercise action acquisition module and an image analysis module. The user information database is used to store the recognized face image and historical exercise information of each user. The exercise database is used to store standard exercise videos, wherein each standard exercise video is divided into a plurality of sub-videos. The face acquisition module is used to acquire the face image of the current user as an image to be authenticated. The face recognition module compares the image to be authenticated with the authenticated face image in the user information database. If there is a certain authenticated face image whose similarity with the image to be authenticated is greater than the authentication similarity threshold, the exercise action acquisition module acquires the exercise action image of the user following the standard exercise video.

Description

Technical Field

[0001] The present invention relates to the technical field of exercise action evaluation, and in particular to an exercise action evaluation system and method based on posture analysis. Background Art

[0002] At present, students have more time after school, but the types of after-school activities are relatively scarce. Arranging and promoting physical exercise can enrich students' extracurricular life and enhance the physical fitness of young people. However, in the actual implementation of physical exercise, schools often verbally assign physical exercise activities to students and require parents to supervise and accompany their children. During the students' home exercise, because there is no corresponding guidance, the standardization of exercise movements cannot be guaranteed, and the students' exercise situation cannot be effectively evaluated. Summary of the invention

[0003] The object of the present invention is to provide a system and method for evaluating exercise movements based on posture analysis to solve the problems raised in the above-mentioned background technology.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: an exercise action evaluation system based on posture analysis, the action evaluation system includes a user information library, an exercise database, a face acquisition module, a face recognition module, an exercise action acquisition module and an image analysis module, the user information library is used to store the authentication face image and historical exercise information of each user, the exercise database is used to store standard exercise videos, wherein each standard exercise video is divided into a plurality of sub-videos, the face acquisition module is used to acquire the face image of the current user as the image to be authenticated, the face recognition module compares the image to be authenticated with the authentication face image in the user information library, if there is a certain authentication face image and the image to be authenticated whose similarity is greater than the authentication similarity threshold, the exercise action acquisition module acquires the exercise action image of the user following the standard exercise video, the image of the user exercising following a certain sub-video of the standard exercise video is assumed to be the actual sub-image, the corresponding sub-video is the reference video of the actual sub-image, the image analysis module analyzes the acquired actual sub-image, and outputs the exercise action evaluation result of the user.

[0005] Furthermore, the action evaluation system also includes a video segmentation module, which obtains an action image at a certain moment in the standard exercise video. If the similarity between the action image at this moment and the action image at the previous moment is less than a consistent similarity threshold, a dividing point is set between this moment and the previous moment, and the video segment between two adjacent dividing points in the standard exercise video is a sub-video.

[0006] Furthermore, the image analysis module includes an effective similarity acquisition module, a comprehensive evaluation value calculation module and a comprehensive evaluation value comparison module. The effective similarity acquisition module extracts the user's exercise action in a certain actual sub-image, and sets the similarity between the exercise action and the reference video of the actual sub-image as the effective similarity. The comprehensive evaluation value calculation module calculates the comprehensive evaluation value of the user's exercise following the standard exercise video. in, h i is the duration of the ith sub-video of the corresponding standard exercise video, S i is the effective similarity value of the ith actual sub-image of the user's exercise following the standard exercise video, k is the number of sub-videos of the standard exercise video that the user is following this time, and the comprehensive evaluation value comparison module compares the comprehensive evaluation value of the user's exercise following the standard exercise video with the comprehensive threshold, and when the comprehensive evaluation value is less than the comprehensive threshold, the exercise action unqualified information is transmitted.

[0007] Furthermore, the image analysis module also includes a central action judgment module, an attention index calculation module, an attention index comparison module and a playback control module. If the effective similarity corresponding to a certain actual sub-image is less than the effective similarity threshold, the central action judgment module judges that the exercise action in the reference video of the actual sub-image is a candidate attention action, and sets the exercise action as the central action. The attention index calculation module obtains the number of times mz that the user follows the standard exercise video to exercise in the images collected within the most recent preset time period, the number of times m1 that the standard exercise video contains the central action, and the number of times m2 that the central action is judged as a candidate attention action in the standard exercise video containing the central action. Then the first index w=(mz-m1) / mz, the second index x=m2 / m1, Obtain the variance e0 of the effective similarity of the actual sub-images corresponding to the most recent preset times of the central action, calculate the variance reference c=e0 / ez, wherein ez is the variance threshold, if the variance reference is greater than or equal to 1, then the third index y=1, if the variance reference is less than 1, then the third index y=c, calculate the attention index Q=0.4*w+0.4*x+0.2*y of a central action, the attention index comparison module compares the attention index of a central action with the attention threshold, when the attention index of the central action is greater than the attention threshold, the playback control module plays the sub-video at a first speed when the exercise action in a sub-video is determined to be an attention action during the most recent exercise, otherwise, plays the sub-video at a second speed, wherein the first speed is less than the second speed.

[0008] An exercise action evaluation method based on posture analysis, the action evaluation method comprising the following steps:

[0009] A user information database and an exercise database are pre-established, wherein the user information database is used to store the authenticated face image and historical exercise information of each user, and the exercise database is used to store standard exercise videos, wherein each standard exercise video is divided into a plurality of sub-videos.

[0010] Collect the current user's face image as the image to be authenticated, and compare the image to be authenticated with the authenticated face images in the user information database. If there is an authenticated face image with a similarity greater than the authentication similarity threshold,

[0011] The standard exercise video plays exercise movements, and the camera collects the exercise movement images of the user following the standard exercise video. The image of the user exercising following a certain sub-video of the standard exercise video is assumed to be the actual sub-image, and the corresponding sub-video is the reference video of the actual sub-image.

[0012] The collected actual sub-images are analyzed and the exercise action evaluation results of the user are output.

[0013] Furthermore, each of the standard exercise videos is divided into a number of sub-videos including:

[0014] Get an action image at a certain moment in the standard exercise video. If the similarity between the action image at this moment and the action image at the previous moment is less than a consistent similarity threshold, set a dividing point between this moment and the previous moment. Then the video segment between two adjacent dividing points in the standard exercise video is a sub-video.

[0015] Furthermore, the analyzing the collected actual sub-images includes:

[0016] Extract the user's exercise action in a certain actual sub-image, and set the similarity between the exercise action in the reference video of the actual sub-image and the actual sub-image as the effective similarity.

[0017] Calculate the comprehensive evaluation value of the user's exercise following the standard exercise video in, h i is the duration of the ith sub-video of the corresponding standard exercise video, S i is the effective similarity value of the ith actual sub-image of the user following the standard exercise video, k is the number of sub-videos of the standard exercise video that the user follows this time,

[0018] If the comprehensive evaluation value is less than the comprehensive threshold, the information that the exercise action is unqualified is transmitted.

[0019] Furthermore, the analyzing the collected actual sub-images also includes:

[0020] If the effective similarity corresponding to a certain actual sub-image is less than the effective similarity threshold, then the exercise action in the reference video of the actual sub-image is a candidate focus action, and the exercise action is set as the central action.

[0021] Obtain the number of times mz that the user follows the standard exercise video to exercise in the images collected within the most recent preset time period, the number of times m1 that the standard exercise video contains the central action, and the number of times m2 that the central action is judged as a candidate focus action in the standard exercise video containing the central action, then the first index w = (mz-m1) / mz, the second index x = m2 / m1,

[0022] Obtain the variance e0 of the effective similarity of the actual sub-image corresponding to the most recent preset number of central actions, and calculate the variance reference c=e0 / ez, where ez is the variance threshold. If the variance reference is greater than or equal to 1, then the third index y=1; if the variance reference is less than 1, then the third index y=c;

[0023] Calculate the attention index of a central action Q = 0.4*w+0.4*x+0.2*y,

[0024] If the attention index of a central action is greater than the attention threshold, then the exercise action is the attention action;

[0025] When the most recent exercise of an exercise action in a certain sub-video is determined to be a focus action, the sub-video is played at a first speed, otherwise, the sub-video is played at a second speed, wherein the first speed is less than the second speed.

[0026] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention sets a standard exercise video, allowing users to follow the standard exercise video to exercise, which helps to improve the standardization of the user's exercise. At the same time, the image of the user exercising following the standard exercise video is collected, and the movements in the collected user exercise image are compared with the movements in the standard exercise video for similarity, and the user's exercise results are determined and output based on this, without the need for manual analysis. At the same time, the present application also controls the playback speed of the standard exercise video according to the user's exercise situation, thereby making the playback situation more targeted and improving the user's experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0028] Figure 1 It is a module schematic diagram of the exercise movement evaluation system based on posture analysis of the present invention. DETAILED DESCRIPTION

[0029] 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.

[0030] See also Figure 1 The present invention provides a technical solution: an exercise action evaluation system based on posture analysis, the action evaluation system includes a user information library, an exercise database, a face acquisition module, a face recognition module, an exercise action acquisition module and an image analysis module, the user information library is used to store the authentication face image and historical exercise information of each user, the exercise database is used to store standard exercise videos, wherein each standard exercise video is divided into a plurality of sub-videos, the face acquisition module is used to acquire the face image of the current user as the image to be authenticated, the face recognition module compares the image to be authenticated with the authentication face image in the user information library, if there is a certain authentication face image and the image to be authenticated whose similarity is greater than the authentication similarity threshold, the exercise action acquisition module acquires the exercise action image of the user following the standard exercise video, the image of the user exercising following a certain sub-video of the standard exercise video is assumed to be the actual sub-image, the corresponding sub-video is the reference video of the actual sub-image, the image analysis module analyzes the acquired actual sub-image, and outputs the exercise action evaluation result of the user.

[0031] The action evaluation system also includes a video segmentation module, which obtains an action image at a certain moment in the standard exercise video. If the similarity between the action image at this moment and the action image at the previous moment is less than a consistent similarity threshold, a dividing point is set between this moment and the previous moment, and the video segment between two adjacent dividing points in the standard exercise video is a sub-video.

[0032] The image analysis module includes an effective similarity acquisition module, a comprehensive evaluation value calculation module and a comprehensive evaluation value comparison module. The effective similarity acquisition module extracts the user's exercise action in a certain actual sub-image, and sets the similarity between the exercise action and the reference video of the actual sub-image as the effective similarity. The comprehensive evaluation value calculation module calculates the comprehensive evaluation value of the user's exercise following the standard exercise video. in, h i is the duration of the ith sub-video of the corresponding standard exercise video, S iis the effective similarity value of the ith actual sub-image of the user's exercise following the standard exercise video, k is the number of sub-videos of the standard exercise video that the user is following this time, and the comprehensive evaluation value comparison module compares the comprehensive evaluation value of the user's exercise following the standard exercise video with the comprehensive threshold, and when the comprehensive evaluation value is less than the comprehensive threshold, the exercise action unqualified information is transmitted.

[0033] The image analysis module also includes a central action judgment module, an attention index calculation module, an attention index comparison module and a playback control module. If the effective similarity corresponding to a certain actual sub-image is less than the effective similarity threshold, the central action judgment module judges that the exercise action in the reference video of the actual sub-image is a candidate attention action, and sets the exercise action as the central action. The attention index calculation module obtains the number mz of times the user follows the standard exercise video to exercise in the images collected within the most recent preset time period, the number m1 of times the standard exercise video contains the central action, and the number m2 of times the central action is judged as a candidate attention action in the standard exercise video containing the central action. Then the first index w=(mz-m1) / mz, the second index x=m2 / m1, and the result is obtained. The variance e0 of the effective similarity of the actual sub-image corresponding to the most recent preset number of central actions is used to calculate the variance reference c=e0 / ez, where ez is the variance threshold. If the variance reference is greater than or equal to 1, then the third index y=1; if the variance reference is less than 1, then the third index y=c. The attention index Q=0.4*w+0.4*x+0.2*y of a central action is calculated. The attention index comparison module compares the attention index of a central action with the attention threshold. When the attention index of the central action is greater than the attention threshold, the playback control module plays the sub-video at a first speed when the exercise action in the sub-video is determined to be an attention action during the most recent exercise. Otherwise, the sub-video is played at a second speed, where the first speed is less than the second speed.

[0034] An exercise action evaluation method based on posture analysis, the action evaluation method comprising the following steps:

[0035] A user information database and an exercise database are pre-established, wherein the user information database is used to store the authenticated face image and historical exercise information of each user, and the exercise database is used to store standard exercise videos, wherein each standard exercise video is divided into a plurality of sub-videos, and the user in this embodiment may be a student;

[0036] Each of the standard exercise videos is divided into a number of sub-videos including:

[0037] Get an action image at a certain moment in a standard exercise video. If the similarity between the action image at this moment and the action image at the previous moment is less than a consistent similarity threshold, it means that the exercise action at this moment is different from that at the previous moment. Set a dividing point between this moment and the previous moment, and the video segment between two adjacent dividing points in the standard exercise video is a sub-video. When doing physical exercise, one action is often performed several times before switching to the next action. Based on this, the standard exercise video is divided into several sub-videos according to different exercise actions. The exercise actions in different sub-videos are not the same, and there is only one exercise action in a sub-video.

[0038] Collect the current user's face image as the image to be authenticated, and use artificial intelligence technology to compare the image to be authenticated with the authenticated face images in the user information database. If there is an authenticated face image with a similarity greater than the authentication similarity threshold,

[0039] The standard exercise video plays exercise movements, and the camera collects the exercise movement images of the user following the standard exercise video. The image of the user exercising following a certain sub-video of the standard exercise video is assumed to be the actual sub-image, and the corresponding sub-video is the reference video of the actual sub-image.

[0040] The collected actual sub-images are analyzed and the exercise action evaluation results of the user are output.

[0041] The analysis of the collected actual sub-images comprises:

[0042] Extract the user's exercise action in a certain actual sub-image, and set the similarity between the exercise action and the reference video of the actual sub-image as the effective similarity. The user in an actual sub-image does the same exercise action. The effective similarity here refers to the average value of the similarity between all the user's exercise actions in the actual sub-image and the exercise actions in the reference video. For example, in a certain actual sub-image, exercise action a is performed, and exercise action a is performed 5 times. The similarity between exercise action a and the corresponding action a in the reference video is obtained. The similarities between the 5 exercise actions a and the action a in the reference video are 75%, 76%, 78%, 75%, and 75% respectively. Then the effective similarity is (75%+76%+78%+75%+75%) / 5=75.8%. In this application, the existing artificial intelligence technology is used to compare the similarity between the user's posture action in the actual sub-image and the posture action in the reference video.

[0043] Calculate the comprehensive evaluation value of the user's exercise following the standard exercise video in, h i is the duration of the ith sub-video of the corresponding standard exercise video, Si is the effective similarity value of the ith actual sub-image of the user's exercise following the standard exercise video, k is the number of sub-videos of the standard exercise video that the user follows this time, H is the total video length of the corresponding standard exercise video, and the ratio of the sub-video length corresponding to a certain action to the total video length is used as the weight of the effective similarity corresponding to the action, so that the result of the comprehensive evaluation value is more accurate, and then the evaluation result of the user's exercise action is more accurate;

[0044] If the comprehensive evaluation value is less than the comprehensive threshold, the information that the exercise action is unqualified is transmitted; the comprehensive evaluation value calculated here refers to the evaluation result of exercising following a standard exercise video. When the comprehensive evaluation value is less than the comprehensive threshold, it means that the exercise action of the user is evaluated to be non-standard;

[0045] The analyzing of the collected actual sub-images also includes:

[0046] If the effective similarity corresponding to a certain actual sub-image is less than the effective similarity threshold, then the exercise action in the reference video of the actual sub-image is a candidate focus action, and the exercise action is set as the central action.

[0047] Obtain the number of times mz that the user exercises following the standard exercise video in the images collected in the most recent preset time period, the number of times m1 that the standard exercise video contains the central action, and the number of times m2 that the central action is judged as a candidate focus action in the standard exercise video containing the central action, then the first index w=(mz-m1) / mz, the second index x=m2 / m1, when the first index is larger, it means that the number of times the user exercises the central action in the most recent preset time period is small, and it is very likely that the user is not familiar with the central action, when the second index is larger, it means that the user is relatively not very standard in performing the central action during exercise, therefore, when the first index and the second index are larger, the speed of playing the video containing the central action should be slowed down as much as possible to facilitate the user to follow and learn the exercise action;

[0048] Obtain the variance e0 of the effective similarity of the actual sub-image corresponding to the most recent preset times of the central action, calculate the variance reference c=e0 / ez, where ez is the variance threshold, if the variance reference is greater than or equal to 1, then the third index y=1, if the variance reference is less than 1, then the third index y=c; in this embodiment, assume that the most recent preset times are the most recent 6 times, then obtain the effective similarity values ​​v1, v2, v3, v4, v5, v6 when exercising the central action for the most recent 6 times, and calculate the variance of v1, v2, v3, v4, v5, v6, if the variance is relatively large, it means that the effective similarity value when the user performs the central action is unstable, which means that the user sometimes performs the exercise posture better and sometimes worse, and at this time, the speed should be slowed down so that the user can further learn and improve the standard accuracy of the action;

[0049] Calculate the attention index of a central action Q = 0.4*w+0.4*x+0.2*y,

[0050] If the attention index of a central action is greater than the attention threshold, then the exercise action is a focus action.

[0051] When the last exercise of an exercise action in a certain sub-video is determined to be a focus action, it means that the user is relatively non-standard when performing the exercise action. Therefore, the sub-video is played at the first speed when exercising the focus action, so that the playback speed is slow, and the user can effectively follow the action of the sub-video to learn and exercise. When the last exercise of an exercise action in a certain sub-video is not determined to be a focus action, the sub-video is played at the second speed, wherein the first speed is less than the second speed. In this embodiment, the second speed is the preset normal playback speed, and the first speed value is 0.75*the second speed value. In this embodiment, "when the last exercise of an exercise action in a certain sub-video is determined to be a focus action" means that the action is determined to be a focus action when the user exercises the action most recently.

[0052] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0053] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An exercise action evaluation system based on posture analysis, characterized in that: The action evaluation system includes a user information database, an exercise database, a face acquisition module, a face authentication module, an exercise action acquisition module and an image analysis module. The user information database is used to store the authentication face image and historical exercise information of each user. The exercise database is used to store standard exercise videos, wherein each standard exercise video is divided into a plurality of sub-videos. The face acquisition module is used to acquire the face image of the current user as the image to be authenticated. The face authentication module compares the image to be authenticated with the authentication face image in the user information database. If there is a certain authentication face image and the image to be authenticated whose similarity is greater than the authentication similarity threshold, the exercise action acquisition module acquires the exercise action image of the user following the standard exercise video. The image of the user exercising following a certain sub-video of the standard exercise video is assumed to be the actual sub-image, and the corresponding sub-video is the reference video of the actual sub-image. The image analysis module analyzes the acquired actual sub-image and outputs the exercise action evaluation result of the user. The action evaluation system further includes a video segmentation module, which obtains an action image at a certain moment in the standard exercise video, and if the similarity between the action image at this moment and the action image at the previous moment is less than a consistent similarity threshold, a dividing point is set between this moment and the previous moment, and the video segment between two adjacent dividing points in the standard exercise video is a sub-video; The image analysis module also includes a central action judgment module, an attention index calculation module, an attention index comparison module and a playback control module. If the effective similarity corresponding to a certain actual sub-image is less than the effective similarity threshold, the central action judgment module judges that the exercise action in the reference video of the actual sub-image is a candidate attention action, and sets the exercise action as the central action. The attention index calculation module obtains the number mz of times the user follows the standard exercise video to exercise in the images collected within the most recent preset time period, the number m1 of times the standard exercise video contains the central action, and the number m2 of times the central action is judged as a candidate attention action in the standard exercise video containing the central action. Then the first index w=(mz-m1) / mz, the second index x=m2 / m1, and the result is obtained. The variance e0 of the effective similarity of the actual sub-image corresponding to the most recent preset number of central actions is used to calculate the variance reference c=e0 / ez, where ez is the variance threshold. If the variance reference is greater than or equal to 1, then the third index y=1; if the variance reference is less than 1, then the third index y=c. The attention index Q=0.4*w+0.4*x+0.2*y of a central action is calculated. The attention index comparison module compares the attention index of a central action with the attention threshold. When the attention index of the central action is greater than the attention threshold, the playback control module plays the sub-video at a first speed when the exercise action in the sub-video is determined to be the attention action during the most recent exercise. Otherwise, the sub-video is played at a second speed, where the first speed is less than the second speed.

2. The exercise movement evaluation system based on posture analysis according to claim 1, characterized in that: The image analysis module includes an effective similarity acquisition module, a comprehensive evaluation value calculation module and a comprehensive evaluation value comparison module. The effective similarity acquisition module extracts the user's exercise action in a certain actual sub-image, and sets the similarity between the exercise action and the reference video of the actual sub-image as the effective similarity. The comprehensive evaluation value calculation module calculates the comprehensive evaluation value of the user's exercise following the standard exercise video. ,in, , is the duration of the ith sub-video of the corresponding standard exercise video, is the effective similarity value of the ith actual sub-image of the user's exercise following the standard exercise video, k is the number of sub-videos of the standard exercise video that the user is following this time, and the comprehensive evaluation value comparison module compares the comprehensive evaluation value of the user's exercise following the standard exercise video with the comprehensive threshold, and when the comprehensive evaluation value is less than the comprehensive threshold, the exercise action unqualified information is transmitted.

3. A method for evaluating exercise movements based on posture analysis, characterized in that: The action evaluation method comprises the following steps: A user information database and an exercise database are pre-established, wherein the user information database is used to store the authenticated face image and historical exercise information of each user, and the exercise database is used to store standard exercise videos, wherein each standard exercise video is divided into a plurality of sub-videos. Collect the current user's face image as the image to be authenticated, and compare the image to be authenticated with the authenticated face images in the user information database. If there is an authenticated face image with a similarity greater than the authentication similarity threshold, The standard exercise video plays exercise movements, and the camera collects the exercise movement images of the user following the standard exercise video. The image of the user exercising following a certain sub-video of the standard exercise video is assumed to be the actual sub-image, and the corresponding sub-video is the reference video of the actual sub-image. Analyze the collected actual sub-images and output the exercise action evaluation results of the user; Each of the standard exercise videos is divided into a number of sub-videos including: Obtain an action image at a certain moment in the standard exercise video. If the similarity between the action image at this moment and the action image at the previous moment is less than a consistent similarity threshold, a dividing point is set between this moment and the previous moment, and the video segment between two adjacent dividing points in the standard exercise video is a sub-video; The analyzing of the collected actual sub-images also includes: If the effective similarity corresponding to a certain actual sub-image is less than the effective similarity threshold, then the exercise action in the reference video of the actual sub-image is a candidate focus action, and the exercise action is set as the central action. Get the number of times mz that the user follows the standard exercise video to exercise in the images collected in the most recent preset time period, the number of times m1 that the standard exercise video contains the central action, and the number of times m2 that the central action is judged as a candidate focus action in the standard exercise video containing the central action, then the first index w=(mz-m1) / mz, the second index x=m2 / m1, Obtain the variance e0 of the effective similarity of the actual sub-image corresponding to the most recent preset number of central actions, and calculate the variance reference c=e0 / ez, where ez is the variance threshold. If the variance reference is greater than or equal to 1, then the third index y=1; if the variance reference is less than 1, then the third index y=c; Calculate the attention index of a central action Q = 0.4*w + 0.4*x + 0.2*y, If the attention index of a central action is greater than the attention threshold, then the exercise action is the attention action; When the most recent exercise of an exercise action in a certain sub-video is determined to be a focus action, the sub-video is played at a first speed, otherwise, the sub-video is played at a second speed, wherein the first speed is less than the second speed.

4. The method for evaluating exercise movements based on posture analysis according to claim 3, characterized in that: The analysis of the collected actual sub-images comprises: Extract the user's exercise action in a certain actual sub-image, and set the similarity between the exercise action in the reference video of the actual sub-image and the actual sub-image as the effective similarity. Calculate the comprehensive evaluation value of the user's exercise following the standard exercise video ,in, , is the duration of the ith sub-video of the corresponding standard exercise video, is the effective similarity value of the ith actual sub-image of the user following the standard exercise video, k is the number of sub-videos of the standard exercise video that the user follows this time, If the comprehensive evaluation value is less than the comprehensive threshold, the information that the exercise action is unqualified is transmitted.

Images