Exercise training method, apparatus and system based on visual ai model

Abstract

The present invention relates to the technical field of data processing. Provided in the present invention are an exercise training method, apparatus and system based on a visual AI model. User information, training data information and dynamic training data are collected in real time by means of a camera, and an actual exercise curve is drawn on the basis of the dynamic training data; on the basis of the actual exercise curve, a training frequency, current planned exercise progress information and current exercise progress information, actual exercise progress information after exercise training is calculated; and on the basis of the actual exercise progress information and the exercise training frequency, the current planned exercise progress information is corrected to obtain corrected planned exercise progress information, so as to guide a user toward more suitable exercise training. Thus, the exercise training effect for users is effectively improved, and the dependence on professionals for exercise training is also effectively reduced. Moreover, by using digital virtual guidance to help users learn relevant professional theory, the reliability of non-professional guidance is significantly improved, and scenarios where exercise-oriented individuals complete self-training with digital guidance are also significantly expanded.

Description

Motion training method, device and system based on visual AI model TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to a motion training method, device and system based on a visual AI model. BACKGROUND

[0002] With the implementation of the national technological revolution and industrial revolution, the development and application process of artificial intelligence in China should be effectively tracked, and important issues such as the promotion of basic research and original innovation in the field of artificial intelligence, the deep integration of artificial intelligence and industrial development, the deep application of artificial intelligence in the fields of people's livelihood protection and social governance, and the governance and international cooperation of artificial intelligence should be further researched, and targeted and operable countermeasures and suggestions should be proposed to help artificial intelligence better empower the construction of modern industrial system and high-quality development, and better create and serve the people's better life.

[0003] The new generation of artificial intelligence is taken as a driving force to promote the development of science and technology, the optimization and upgrading of industry, and the overall leap of productivity, so as to promote new breakthroughs in the technological innovation of artificial intelligence in China, the rapid development of the artificial intelligence industry, and the deep development of the application of artificial intelligence in the integration with modern industry, and to make solid progress in cultivating new productivity, building a modern industrial system, and promoting high-quality development.

[0004] In the process of rapid development and deep application of the new generation of artificial intelligence industry, the importance of combining artificial intelligence with sports is increasingly valued. Skeletal movement is actually the movement of skeletal muscle responsible for the skeleton, and movement is an action on the function of certain skeletal muscle. Due to various factors, including diseases, accidental injuries, etc., users may face challenges in the recovery or improvement of motor function.

[0005] Traditional motion training often needs long-term tracking and guidance in order to correct the movement action and the number of movements, etc., which is a challenge to the time and energy of the instructors. At the same time, the medical resources in some areas are scarce, and users cannot obtain professional motion guidance in time, which further increases the difficulty of movement. In addition, in the current cognition, the standard of movement action is relatively broad, which is mostly the summary of professional experience, which will inevitably lead to the fact that some actions are not stable enough to help the function recovery of certain skeletal muscle, so that users cannot enjoy high-quality motion training.

[0006] However, with the continuous development of computer vision and deep learning technology, we have the ability to realize a child-centered, family-centered medical service system, based on this point, we hope to provide a convenient and effective exercise training method for exercise programs. Combined with mobile devices such as smartphones, these technologies can be used to provide personalized, intelligent, and professional exercise training programs in home or out-of-hospital settings, enabling comprehensive monitoring and management of the exercise process. This exercise method not only alleviates the problem of medical resource shortage, but also improves exercise efficiency and exercise training participation, providing strong support for the healthy growth and future development of users. SUMMARY

[0007] The application provides a kind of exercise training method, device and system based on visual AI model, to solve the above technical problems.

[0008] The application is implemented in this way, on the one hand, the application provides an exercise training method based on visual AI model, comprising: collecting user information of a user to be trained through a mobile terminal; obtaining training data information of the user according to the user information, the training data information including the number of times of exercise training of the user, current planned exercise progress information and current exercise progress information; real-time collection of dynamic training data during the user's exercise training process, the dynamic training data including joint angles during the user's exercise training process and conclusion data given by the visual AI model for the joint angles; drawing the actual exercise curve of the user according to the collected joint angles and conclusion data; calculating the actual exercise progress information of the user after exercise training according to the actual exercise curve, the number of training times, the current planned exercise progress information and the current exercise progress information; correcting the above current planned exercise progress information according to the actual exercise progress information and the number of exercise training times to obtain the corrected planned exercise progress information to guide the user to exercise.

[0009] Preferably, the actual exercise curve of the user is drawn according to the collected joint angles and conclusion data, comprising: obtaining functional recovery index data related to the user information; according to the statistical algorithm, the actual exercise curve with time as the horizontal coordinate and exercise state as the vertical coordinate is calculated and obtained according to the joint angle, the conclusion data and the functional recovery index data, containing P value (with 0.05 as the statistical significance); wherein the time is in units of days, weeks or months; the functional recovery index data at least includes the recovery angle of the disabled joint, muscle strength, activity ability and imaging evaluation result.

[0010] Preferably, the actual exercise progress information of the user after the exercise training is obtained according to the actual exercise curve, the training frequency of the user, the current planned exercise progress information and the current exercise progress information, and includes: obtaining the functional recovery index data of the user after the exercise training according to the actual exercise curve; comparing the functional recovery index data of the user after the exercise training with the dynamic training data in the current exercise progress information; and obtaining the actual exercise progress information of the user after the exercise training according to the comparison result, the training frequency of the user and the current planned exercise progress information.

[0011] Preferably, the current planned exercise progress information is corrected according to the actual exercise progress information and the exercise training frequency to obtain the corrected planned exercise progress information to guide the user to perform the exercise training, and includes: obtaining the planned exercise training frequency, the planned exercise training time and the planned exercise training index data in the current planned exercise progress information; obtaining the activity angle of the disabled joint after the actual exercise training according to the actual exercise progress information; and correcting the planned exercise training frequency and the planned exercise training time according to the activity angle and the exercise training frequency to obtain the new planned exercise training time, the planned exercise training frequency and the planned exercise training index data as the corrected planned exercise progress information.

[0012] Preferably, the planned exercise training frequency and the planned exercise training time are corrected according to the activity angle and the exercise training frequency, and includes: searching the exercise training frequency and the exercise training time corresponding to the activity angle and the exercise training frequency in the preset rule database; comparing the exercise training frequency and the exercise training time with the planned exercise training frequency and the planned exercise training time, and calculating the difference of the two respectively; and adding the integer of half of the difference to the planned exercise training frequency and the planned exercise training time to obtain the new planned exercise training time and the planned exercise training frequency.

[0013] Preferably, the dynamic training data in the exercise training process of the user is collected in real time, and includes: obtaining the joint angle of the user in the exercise training process and the time of maintaining the joint angle in real time; associating the data according to the combination of the generative model and the recognition capture; comparing the joint angle of the user in the exercise training process and the time of maintaining the joint angle with the data in the preset rule database to analyze and judge the joint angle of the user in the exercise training process and the time of maintaining the joint angle; and taking the joint angle meeting the data in the preset rule database and the conclusion data of the joint angle given by the visual AI model as the dynamic training data in the exercise training process of the user to be collected.

[0014] Preferably, the method further comprises: acquiring, by the server, training data of each time of movement training of the user, the training data comprising angle, position, times and qualified times of the disabled joint movement training; classifying the training data; training the classified training data, actual movement curve and actual movement progress information to obtain key data affecting the movement training, and establishing a corresponding movement database according to the key data.

[0015] Preferably, the training of the classified training data, actual movement curve and actual movement progress information comprises: detecting the disabled joint of the user to acquire the disability information of the joint of the user; comparing the disability information with at least a movement database comprising a movement knowledge graph or a standard movement action to obtain a planned movement value; deriving the cycle and training content of the required movement according to the planned movement value to obtain the angle of the joint movement training required by the user every day, and taking the angle as the key data affecting the movement training so as to guide the user to perform the movement training.

[0016] On the other hand, the application further provides a movement training device based on a visual AI model, comprising: an identity acquisition module for acquiring user information of a user to be trained; an information acquisition module for acquiring training data information of the user according to the user information; a data acquisition module for acquiring dynamic training data in a movement training process of the user in real time; a curve drawing module for drawing an actual movement curve of the user according to the acquired joint angle and conclusion data; an information calculation module for calculating actual movement progress information of the user after movement training according to the actual movement curve, training times of the user, current planned movement progress information and current movement progress information; and an information correction module for correcting the current planned movement progress information according to the actual movement progress information and movement training times to obtain corrected planned movement progress information for guiding the user to perform the movement training.

[0017] In yet another aspect, the application also provides a sports training system based on a visual AI model, comprising a mobile terminal and a server, wherein the mobile terminal collects user information of a user to be trained based on the visual AI model, the mobile terminal is also used to obtain training data information of the user according to the user information and to collect dynamic training data in a real-time manner during the sports training process of the user; the server is used to draw an actual movement curve of the user according to the collected joint angle and conclusion data, the server is also used to calculate actual movement progress information of the user after sports training according to the actual movement curve, the training times of the user, current planned movement progress information and current movement progress information, and the server is also used to correct the current planned movement progress information according to the actual movement progress information and the movement training times to obtain corrected planned movement progress information to guide the user to perform sports training.

[0018] This invention provides a motion training method based on a visual AI model. Compared with existing technologies, the beneficial effects of this invention are as follows: The motion training method based on a visual AI model includes: collecting user information via a mobile device; obtaining user training data information based on the user information, the training data information including the number of user's motion training sessions, current planned motion progress information, and current motion progress information; collecting dynamic training data in real time during the user's motion training process, the dynamic training data including joint angles during the user's motion training process and conclusion data given by the visual AI model for the joint angles; plotting the user's actual motion curve based on the collected joint angles and conclusion data; calculating the user's actual motion progress information after motion training based on the actual motion curve, the number of user's training sessions, current planned motion progress information, and current motion progress information; and correcting the current planned motion progress information based on the actual motion progress information and the number of motion training sessions to obtain corrected planned motion progress information to guide the user in motion training. The exercise training method based on a visual AI model provided in this application collects training data and dynamic training data from a mobile device based on a visual AI model. It then calculates the user's actual exercise progress based on this data, and subsequently modifies the planned exercise progress based on the actual progress. This revised planned progress guides the user's exercise training, providing appropriate guidance throughout the process. Furthermore, this method effectively modifies subsequent exercise plans, guiding the user on the number of repetitions and joint angles of affected joints. It provides comprehensive and effective guidance for the user's exercise training, addressing the challenges of achieving standardized and effective exercise training regardless of location or expert guidance. This visual AI model-based exercise training method not only improves the user's exercise training effectiveness but also reduces reliance on medical personnel and enhances the reliability of exercise training. Attached Figure Description

[0019] FIG. 1 is a structural block diagram of a visual AI model-based sports training system according to an embodiment of the present application; FIG. 2 is a functional module block diagram of a visual AI model-based sports training device according to an embodiment of the present application; FIG. 3 is a flowchart of a visual AI model-based sports training method according to an embodiment of the present application; FIG. 4 is a flowchart of another visual AI model-based sports training method according to an embodiment of the present application; FIG. 5 is a flowchart of another visual AI model-based sports training method according to an embodiment of the present application; FIG. 6 is a flowchart of another visual AI model-based sports training method according to an embodiment of the present application; FIG. 7 is a flowchart of another visual AI model-based sports training method according to an embodiment of the present application; FIG. 8 is a flowchart of another visual AI model-based sports training method according to an embodiment of the present application; FIG. 9 is a flowchart of another visual AI model-based sports training method according to an embodiment of the present application; and FIG. 10 is a flowchart of another visual AI model-based sports training method according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0021] As shown in FIG. 1, a structural block diagram of a visual AI model-based sports training system according to the present application is shown. The visual AI model-based sports training methods disclosed in the present application are all processed based on the system. The visual AI model-based sports training system disclosed in the present application will be described in detail in combination with FIG. 1.

[0022] Before the embodiments of the present application are described, the visual AI model shown in the present application is a two-dimensional human pose estimation model, which can detect the positions of human key points (such as nose, eyes, shoulders or various joints, etc.) from RGB images or videos. The basic method is to first detect the human bounding box, and then estimate the pose of the human in each bounding box individually, so as to obtain the related data and correctness of the user's pose, etc.

[0023] Specifically, the visual AI model proposed in the present application can include the following steps when performing sports training: 1. Posture detection: first use human posture detection technology (see related prior art) to process the video frames in the user's movement process, detect the key point coordinates (key points such as joint positions) of the user, which can be processed in real time or offline. 2. Posture evaluation: According to the detected key point information, evaluate whether the user's posture is correct, which requires a pre-defined standard action reference template, such as the angle range of each joint and the relative position of the limbs, etc. Compare the user's actual posture with the standard template to determine whether the posture is correct. 3. Feedback prompt: According to the results of posture evaluation, give corresponding feedback and guidance, which can prompt the user to adjust the part and correct the action essentials on the interface in the form of text, image, voice, etc. 4. Action recognition: In addition to single-frame posture analysis, a continuous action sequence can also be recognized by analyzing the motion trajectory, angle and time change of the key points to determine which specific action the user is doing, such as push-ups, squats, etc. 5. Action counting: Based on the recognition of action type, the number of times each action is completed can be further counted, which is generally achieved by detecting the periodic change of key points, such as the fluctuation of elbow angle during push-ups; action counting can help users quantify the intensity of exercise and set reasonable exercise goals. 6. Historical tracking: Save the posture evaluation results and action statistics of each exercise to form a time sequence; users can review their historical performance to understand the mastery of technology and progress trends, and adjust the training plan according to these data. 7. Human-computer interaction: In addition to post-analysis, various interactive training systems can also be designed using posture tracking technology; for example, a virtual screen is displayed on the screen to guide the user to perform actions or map the user's actions in real time to game characters, thereby increasing the fun and immersion of exercise.

[0024] Firstly, the exercise training system based on the visual AI model includes a mobile terminal and a server, wherein the mobile terminal is a client, which can be a mobile device such as a mobile phone or a PAD, and the server is a server, which can be one or a combination of a WEB server, a database server, and a cloud server, and the like, which will not be described in detail. The client of the present application collects and processes data of user's exercise training based on the visual AI model. The mobile terminal of the present application collects user information based on the visual AI model, which can be the identity card number and name, gender, and the like of the user when the user logs in the system through the mobile terminal, or the face information collected by the camera of the mobile terminal or the fingerprint information collected by the fingerprint module, of course, the face information collected by the camera of the mobile terminal can also be used after the identity card number and name, gender, and the like of the user logged in through the mobile terminal are collected, and then the two are compared to determine whether they are unified, and then the next related operation is performed. That is, the mobile terminal acquires the training data information of the user and dynamically collects the dynamic training data in the user's exercise training process according to the user information. The mobile terminal of the present application collects the corresponding information based on the visual AI model.

[0025] Of course, in the specific implementation process, the mobile terminal can register, log in, modify, and retrieve related functions such as account registration and login through the user's identity card number, portrait, and the like, and each person can also set multiple accounts, multiple accounts can be associated with each other, and personal information can also be changed accordingly. At the same time, the mobile terminal can also have related functions such as calendar clock-in, plan creation / management, data upload, payment, and the like in the specific implementation process, which can be referred to the related description of the prior art, and will not be described in detail. Even in the implementation process of the present application, since the camera needs to collect the user's action (or posture) in the exercise training process in real time, in order to judge the captured posture (such as joint activity angle), the corresponding animation is formed by rendering the action of the user's exercise training to obtain the corresponding data, so that the mobile terminal or the server can process the data, and in the server processing process, the action in the preset database can also be compared with the data obtained from the captured posture, so as to judge the effectiveness or accuracy of the user's action, and the current user's action can be played through voice, of course, the above is only an example, which will not be described in detail, and the following is a related description of the exercise training system based on the visual AI model, the exercise training device based on the visual AI model, and the exercise training method based on the visual AI model.

[0026] The server end is used for drawing an actual movement curve of the user according to the collected joint angle and conclusion data, the server end is also used for calculating actual movement progress information of the user after movement training according to the actual movement curve, the training times of the user, current planned movement progress information and current movement progress information, and the server end is also used for correcting the current planned movement progress information according to the actual movement progress information and the movement training times to obtain corrected planned movement progress information to guide the user to perform movement training.

[0027] In the implementation process, the movement training method based on the visual AI model is also applied to a movement training device based on the visual AI model, as shown in FIG. 2, the movement training device can include an identity collection module 11, an information acquisition module 12, a data collection module 13, a curve drawing module 14, an information calculation module 15 and an information correction module 16, wherein the identity collection module 11 is used for collecting user information; the information acquisition module 12 is used for acquiring training data information of the user according to the user information; the data collection module 13 is used for collecting dynamic training data in the movement training process of the user in real time; the curve drawing module 14 is used for drawing an actual movement curve of the user according to the collected joint angle and conclusion data; the information calculation module 15 is used for calculating actual movement progress information of the user after movement training according to the actual movement curve, the training times of the user, current planned movement progress information and current movement progress information; and the information correction module 16 is used for correcting the current planned movement progress information according to the actual movement progress information and the movement training times to obtain corrected planned movement progress information to guide the user to perform movement training.

[0028] In combination with the above disclosed movement training device based on the visual AI model and the movement training system based on the visual AI model, the application further discloses an embodiment of a movement training method based on the visual AI model in combination with FIGS. 3-10, which can be specifically referred to the following description.

[0029] As shown in FIG. 3, the movement training method based on the visual AI model disclosed by the application includes: in step S100, collecting user information of a user to be trained by a mobile terminal.

[0030] In the implementation process, the user information can be the identity card number and name, gender, and the like of the user when logging in to the system through the mobile terminal, or the face information collected through the camera of the mobile terminal or the fingerprint information collected by the fingerprint module. Of course, in the implementation process, since the user needs to log in to the mobile terminal APP system first, the mobile terminal can at least obtain the fingerprint or face information of the user after logging in to the system, and then obtain the identity card number and name, gender, and the like of the user corresponding to the features according to the fingerprint and face information. Details are not described herein.

[0031] In step S200, the training data information of the user is obtained according to the user information.

[0032] In the implementation process, since the user has a corresponding case record in the hospital or has recorded the corresponding exercise training information in the system before using the system, after obtaining the user information in step S100, the user information can be used to retrieve some training information corresponding to the user stored in the database. The training data information includes the number of exercise training of the user (i.e., the number of exercise training that has been performed, and of course the number can also be 0), the current planned exercise progress information (i.e., the planned exercise progress information before this exercise training, which can include the planned exercise training number, the planned exercise training time, and the planned joint recovery degree of exercise, and the like), and the current exercise progress information (i.e., the exercise progress before this exercise training).

[0033] In step S300, dynamic training data in the user exercise training process is collected in real time.

[0034] In the implementation process, after the user logs in to the client, the corresponding exercise can be performed according to the corresponding planned exercise information given by the visual AI model, for example, the exercise of the disabled joint of the user, so that the action of the user exercise training process based on the visual AI model is collected, so that the dynamic training data can be collected. Specifically, the dynamic training data includes the joint angle in the user exercise training process and the conclusion data given by the visual AI model to the joint angle. The conclusion data is a corresponding conclusion of whether the joint angle of the active joint of the user in the exercise process meets the requirements, so as to determine whether the data is valid.

[0035] Of course, in the implementation process, the conclusion data can also be the completion degree (such as the completion percentage) of the joint angle. For example, the required joint angle of the exercise training is 100 degrees, and the actual completed joint angle is 99 degrees, so the conclusion data is 99%. The above is only an exemplary representation, and is not the technical solution that must be adopted by the present application. Only an example is shown, and details are not described herein.

[0036] In step S400, an actual movement curve of the user is drawn according to the collected joint angle and conclusion data.

[0037] In the implementation process, reference can be made to FIG. 4, which shows detailed steps of step S400.

[0038] In step S401, functional recovery index data related to the user information is obtained.

[0039] In the implementation process, the functional recovery index data refers to an index data that the user can reach after movement training, which can include a predetermined value of the joint angle that the user can reach after movement training, etc. The functional recovery index data is related data before the present movement training.

[0040] In step S402, an actual movement curve with P value (with 0.05 as statistical significance) is calculated according to the statistical algorithm, the joint angle, the conclusion data and the functional recovery index data, with time as the horizontal coordinate and movement state as the vertical coordinate.

[0041] In the implementation process, the time in the coordinate system is in units of days, weeks or months; the functional recovery index data at least includes recovery angle of the disabled joint, muscle strength, activity ability, and imaging evaluation result. The activity ability mainly refers to the activity type and intensity of the normal joint, and the imaging evaluation result mainly refers to the relevant results of X-ray or MRI, showing the situation after fracture healing. The conclusion data is the conclusion of whether the joint angle is correct after the camera captures the movement training action or posture of the user in real time.

[0042] In the implementation process, the statistical algorithm of the present application commonly uses three methods: 1) difference analysis (mainly to study the difference between different data, including single sample mean comparison, two sample mean comparison, multiple sample mean comparison, and categorical data comparison), 2) correlation study (mainly to study the correlation between data, including using Pearson correlation coefficient for analysis when meeting normal distribution, using spearman correlation coefficient for analysis when not meeting normal distribution, and using spearman correlation coefficient for analysis when multiple classification ordered data), 3) influence factor study (mainly to study the influence factors of dependent variable, which can be divided into quantitative research using linear regression method, categorical research using logistic regression method, and survival data research using Cox method). After obtaining the relevant parameters of the user, the relevant statistical research methods can be used to obtain the joint angle, conclusion data and functional recovery index data to calculate the actual motion curve containing P value. The application of the above statistical algorithm is only one way, and is not limited to this way. The present application only uses the related statistical algorithm in the prior art to calculate the corresponding horizontal coordinate and total coordinate curve, which is not described in detail here.

[0043] At the same time, in the implementation process, the statistical algorithm is only an exemplary statistical strategy. Firstly, it can include the following prediction models: logistic prediction model and Cox regression prediction model, and the above models can use Lasso algorithm to actually develop the prediction model. Secondly, real world research is carried out, including review research design points, observation research bias and its control, observation research common statistical methods, independent variable screening method principle and method, propensity score method, propensity score weighted matching, trend research analysis method, observation research paper writing method, which are not described in detail here.

[0044] Of course, in the implementation process, in addition to the joint angle, the conclusion data after capturing the posture and the functional recovery index, the actual motion curve can also be calculated by the data such as the action done by the user every day, the action angle, the effective number, the frequency (such as 9 times a day, 3 times in the morning, 3 times at noon and 3 times in the evening, 10 groups of actions every day to reach what angle, etc.), etc. The specific details are not described here.

[0045] In step S500, the actual motion progress information of the user after the motion training is calculated according to the actual motion curve, the training number of the user, the current planned motion progress information and the current motion progress information.

[0046] In the implementation, after the actual movement curve is obtained, the following steps can be performed: in step S501, the functional recovery index data of the user after the current movement training is obtained according to the actual movement curve.

[0047] In the implementation, the functional recovery index data at least includes the recovery angle of the disabled joint, the muscle strength, the activity ability, and the imaging evaluation result. The functional recovery index data in step S501 is the index data obtained after the current movement training, which is different from the functional recovery index data in step S401.

[0048] In step S502, the functional recovery index data of the user after the current movement training is compared with the dynamic training data in the current movement progress information.

[0049] In the implementation, the dynamic training data in the current movement progress information generally refers to the dynamic training data that the disabled joint should be able to reach a certain recovery degree if the current movement training is performed. Therefore, the functional recovery index data after the current movement training is compared with the dynamic training data in the current movement progress information to see whether there is a corresponding difference or how much the difference is.

[0050] In step S503, the actual movement progress information of the user after the movement training is obtained according to the comparison result, the training times of the user, and the current planned movement progress information.

[0051] In the implementation, the difference or the difference between the functional recovery index data after the current movement training and the dynamic training data in the current movement progress information is obtained in step S502. Then, the actual recovery progress information of the user after the movement training can be obtained by combining the comparison result, the training times of the user, and the current planned movement progress information (such as the dynamic training data of the planned movement, etc.). For example, the movement recovery progress of the user reaches how much compared with the planned movement progress. If it is below 100%, it represents that the recovery degree of the movement training is low. If it is higher than 100%, it represents that the recovery degree of the movement training is high or far exceeds the expectation. The above is only an exemplary description, which is not described in detail here.

[0052] In step S600, the current planned movement progress information is corrected according to the actual movement progress information and the movement training times, and the corrected planned movement progress information is obtained to guide the user to perform the movement training.

[0053] In the implementation, the current planned exercise progress information is corrected, so that the next user exercise training process can be effectively guided. For example, according to step S503, if the exercise training recovery degree is low, the exercise training time and / or exercise training intensity may be increased in the next exercise training plan, and vice versa. The specific details are not described again. Of course, the detailed steps of step S600 can be described with reference to FIG. 6.

[0054] In step S601, the planned exercise training times, the planned exercise training time, and the planned exercise training index data in the current planned exercise progress information are obtained.

[0055] In the implementation, after step S503 is completed, the planned exercise training times, the planned exercise training time, and the planned exercise training index data in the current planned exercise progress information are obtained. Here, the applicant needs to explain that the current planned exercise progress information is the exercise progress information before this exercise training (i.e., after the last exercise training is completed), and the actual exercise progress information is the exercise progress information after this exercise training. By understanding the planned exercise training times, the planned exercise training time, and the planned exercise training index data in the last planned exercise progress information, the index data after this exercise training and the like can be adjusted accordingly, so that the user's subsequent exercise training can be guided after each exercise training, so that the user's exercise training action and time can be adjusted in time, so that the exercise effect can be achieved as soon as possible.

[0056] In step S602, the activity angle of the disabled joint after the actual exercise training is obtained according to the actual exercise progress information.

[0057] In the implementation, the activity angle of the disabled joint refers to the exercise standard that can be achieved by the disabled joint after this exercise training, or the exercise purpose achieved by the disabled joint. The activity angle of the disabled joint can be stored in a database, so the activity angle can be obtained through the actual exercise progress information.

[0058] In step S603, the planned exercise training times and the planned exercise training time are corrected according to the activity angle and the exercise training times, and the new planned exercise training time, the planned exercise training times, and the planned exercise training index data are obtained as the corrected planned exercise progress information.

[0059] In the implementation process, when the activity angle of the user's disabled joint and the number of exercise training performed are obtained, the activity angle and the number of exercise training can be compared with the dynamic training data and the planned exercise training number in the last planned exercise progress information, and then the planned exercise training number and the planned exercise training time are corrected to obtain new planned exercise training time, planned exercise training number and planned exercise training index data. For example, the dynamic training data in the last planned exercise progress information is an activity angle of 10 degrees, a planned exercise training number of 28 times, and a planned exercise training time of 14 days, and according to the actual exercise training information, the activity angle of the user's disabled joint is 8 degrees and the number of exercise training performed is 16 times, then the correction method can be the activity angle / exercise training number (8 / 16) in the actual exercise training information is 0.5 degrees, that is, the activity angle of the disabled joint is 0.5 degrees for each exercise training in the actual exercise training, then the planned exercise training number in the new planned exercise training information can be the activity angle in the last planned exercise progress information minus the activity angle in the actual exercise training information (10-8) is 2 degrees, then the planned exercise training number in the new planned exercise training information is (2 / 0.5) 4 times, if the exercise training is 2 times a day, the new planned exercise training time is (4 / 2) 2 days. Of course, the above is only an example and does not represent the training method in the implementation process of the present application, and the activity angle of 10 degrees and 8 degrees and other related information described in the above embodiment do not represent a certain user, but an example. In actual practice, it also needs to be set and explained according to the actual situation, which will not be described in detail here.

[0060] Of course, in the implementation process, reference can also be made to the flowchart of the exercise training method based on the visual AI model disclosed in FIG. 7. The step includes: in step S6031, finding the exercise training number and exercise training time corresponding to the activity angle and exercise training number in the preset rule database.

[0061] In the implementation process, the preset rule database can be a database established according to the joint and the corresponding exercise action, and of course, it can also be a database related to the activity angle and the number of exercise training added on the basis of the above database, which will not be described in detail here.

[0062] In step S6032, the exercise training number and exercise training time are compared with the planned exercise training number and planned exercise training time, and the difference between the two is calculated.

[0063] In the implementation process, if the normal exercise training according to the exercise training plan, the remaining exercise training times and exercise training time are usually equal. However, due to the individual differences of each user and the fact that the user does not exercise according to the plan or the action in the exercise training process does not conform to the regulations, the two will be different. Therefore, the exercise training times and exercise training time obtained by step S6031 are compared with the planned exercise training times and planned exercise training time, and the difference between the two is obtained. If the difference is positive, it means that the user does not exercise according to the plan or the exercise action is not in compliance, and then the planned exercise training times and planned exercise training time need to be increased, and vice versa.

[0064] In step S6033, the planned exercise training times and planned exercise training time are added to the half of the difference between the two, and the new planned exercise training time and planned exercise training times are obtained.

[0065] In the implementation process, the new planned exercise training time and planned exercise training times are calculated after the integer is obtained, as described in step S6032. Since the difference between the two may be negative, the new time and times may increase or decrease, which will not be described in detail here.

[0066] The modification scheme for the planned exercise progress information disclosed in the embodiments of the present application is only an exemplary description given by the present application. The specific description can be combined with the related schemes disclosed in the above Figure 6, which will not be described in detail here.

[0067] In addition, referring to the flowchart of the exercise training method based on visual AI model shown in Figure 8. This is a further detailed description of step S300. As shown in Figure 8, step S300 disclosed in the present application further comprises the following steps: in step S301, the joint angle of the user exercise training and the time of maintaining the joint angle are obtained in real time.

[0068] In the implementation process, when the user exercises, the mobile terminal can usually send the training target of this exercise training (for example, the joint angle to be reached in this training and the time of maintaining the joint angle, etc.). The user can stretch the joint according to the training target. The joint stretching information can be obtained by real-time collection and calculation of the mobile terminal, and the time of maintaining the joint angle can also be obtained, which will not be described in detail here.

[0069] In step S302, the data association is performed according to the combination of the generative model and the recognition capture.

[0070] In the implementation process, the generative model, i.e., generative artificial intelligence (Generative AI), is an artificial intelligence technology that can autonomously generate text, images, audio, and other content. The related technology can be referred to in the prior art, and will not be described in detail here. The identification capture is to capture the posture in the user's exercise process in real time through a camera. In this application, the video and angle information of the user can be obtained by digitizing the guidance of the exercise group outside the hospital. In order to better integrate with the generative model, we need to communicate with the muscle electrical signal acquisition. Through the collection of muscle electrical signals, the visual AI model and the generative model can be integrated with each other, thereby forming a dynamic action based on the generative model (i.e., an AI training action used by the user for exercise training). The above is only an exemplary description, and specific contents can be referred to in the description before and after and the related contents of the prior art, which will not be described in detail here.

[0071] In step S303, the joint angle of the user's exercise training and the time for maintaining the joint angle are determined and analyzed according to the comparison between the data in the preset rule database and the joint angle of the user's exercise training and the time for maintaining the joint angle.

[0072] In the implementation process, since the generative model does not provide specific rules for determining the angle of joint movement, we need to optimize and provide relevant rules for the model. The visual AI model compares the theoretical value provided by us with the actual captured value to determine whether the actual value is correct or incorrect. Specifically, the theoretical value can be pre-established in the corresponding rule database. By comparing the actual dynamic training data with the theoretical value data in the preset rule database, it can be determined whether the joint movement angle corresponding to the actual dynamic training data is valid or correct. Of course, the above is only an exemplary description, and in the implementation process, each joint movement can also correspond to one or more theoretical values in the preset rule database (different age groups, such as 1-10 years old, 11-14 years old, 15-18 years old, 19-22 years old, 23-30 years old, 30-40 years old, etc. each corresponding to a corresponding theoretical value. The above is only an exemplary description and does not represent the protection scope of the present application.

[0073] In the implementation process, the pre-established corresponding rule database can also include the time for maintaining each joint angle and the time required for maintaining each joint angle. Therefore, the joint angle and the time for maintaining the joint angle collected in step S301 can be compared with the data in the preset rule database, so that the completion degree of the joint angle that can be maintained for a corresponding time or whether the joint movement angle is valid or correct can be determined. The conclusion data given by the visual AI model for the joint angle will not be described in detail here.

[0074] In step S304, the joint angle meeting the data in the preset rule database and the conclusion data given by the visual AI model to the joint angle are taken as the dynamic training data in the user motion training process to be collected.

[0075] In the implementation process, when the corresponding conclusion is obtained in step S303, it can be determined that the actual dynamic training data with the judgment result of valid or correct is valid data, that is, the data that can affect or help the motion training, so as to help the user to perform the motion training in time and effectively in the subsequent user motion training process.

[0076] Meanwhile, in the implementation process, referring to FIG. 9, the application also discloses a motion training method based on a visual AI model, including the following steps: in step S700, the training data of each motion training of a user is obtained through a server, and the training data includes the angle, position, times and qualified times of the motion training of the disabled joint.

[0077] In the implementation process, during the motion training of the user, the camera collects the relevant actions of the user during the motion training, such as the position of the joint (such as the position of the active joint - upper limbs, lower limbs or elbow joint, etc.), the angle of the joint movement, the times of the joint movement and the relatively qualified times, which are not described in detail here.

[0078] In step S800, the training data is classified.

[0079] In the implementation process, a database can be established according to the joint and the corresponding motion action in the application, that is, the collection and storage of the angle, position, times and qualified times of each training of a certain joint during the motion training are established in the database. Of course, it can also be classified and processed through the EXCEL table, but the EXCEL table is only a form of record, and the core is to establish the database of the video picture angle when the user performs the corresponding action according to the matching motion action of different joints.

[0080] In step S900, the classified training data, the actual motion curve and the actual motion progress information are trained to obtain the key data affecting the motion training, and a corresponding motion database is established according to the key data.

[0081] In the implementation process, for the classified training data, the latter needs to use this part of data for research similar to randomized controlled trial (RCT), such as a random group and a control group, and changes a parameter in the database to see if it has an impact on the effect. Thus, through relevant experiments and data processing, the key data affecting sports training can be obtained, thereby facilitating the use of the corresponding data to guide the user's sports training.

[0082] In the implementation process, referring to the detailed flow chart of step S900 disclosed by the present application shown in FIG. 10, step S900 further includes the following steps: in step S901, the disabled joint of the user is detected to obtain the disability information of the joint of the user.

[0083] In the implementation process, the detection of the disabled joint of the user can be that an expert judges the relevant information of the disabled joint of the user, such as the movement angle of the disabled joint and the position of the disabled joint, through the relevant results of X-ray or MRI of the user, which will not be described in detail here and can be referred to the description of the above embodiments.

[0084] In step S902, a planned movement value is obtained by comparing the disability information with a movement database including at least a movement knowledge graph or a standard movement action.

[0085] In the implementation process, the movement database at least includes a movement knowledge graph and a standard movement action, wherein the movement knowledge graph and the standard movement action can generally collect the relevant information of the movement knowledge corresponding to different disabled joints in the existing books and medical knowledge through the database, such as standard movement actions, movement cycles, movement angles required for movement training, and corresponding atlas actions, etc. However, in the actual operation process, the database can also continuously optimize and adjust the movement knowledge graph and the standard movement action through the continuously collected big data of the actual movement of the user, which will not be described in detail here and can be referred to the prior art, etc.

[0086] Therefore, after obtaining the disability information, the disabled articulation angle and the position of the disabled joint can be known. Since the joints of users in different age groups have a corresponding range of motion in the normal state, the range of motion of the joint corresponding to the position of the disabled joint can be obtained according to the user information in combination with the range of motion of the joint in the normal state and the position of the disabled joint. For example, a user's joint can be moved 180 degrees before the elbow joint fracture in the normal state, and the joint of the current user can only be moved 150 degrees after the disability. Therefore, the difference in the required motion is 180-150, which is equal to 30 degrees. That is, the planned motion value is 30 degrees of the user's elbow joint. Of course, the above is only an exemplary description, and the specific needs are determined according to the user's age, gender, joint position, etc., which will not be described in detail here.

[0087] In step S903, the cycle and training content required for the motion are obtained according to the planned motion value, and the angle of the joint motion training required by the user per day is obtained as the key data affecting the motion training, so as to guide the user to perform the motion training.

[0088] In the specific implementation process, if the planned motion value obtained in step S902 is 30 degrees, the required motion cycle and training content can be calculated in the actual motion process. For example, 1 degree of motion can be achieved per day, so the motion cycle is 30 / 1=30 days, and the training content is to achieve 1 degree of motion, such as 10 times of 2-degree motion per day, etc., which will not be described in detail here. That is, the training content is the motion action (such as the range of motion, etc.) required by the user to achieve the motion target per day.

[0089] In the specific implementation process, the existing planned motion progress is made according to the content required by the motion course, which is only a theoretical value without user actual indicators for reference. That is, there is no logical correlation between the theoretical value and the actual value. A large amount of real user data is required to further optimize and improve the theoretical value. For example, the theoretical value is to make 10 times of 2-degree motion per day, but after a large amount of actual user data is collected, statistical analysis (such as RCT research or statistical conclusion of R language) may show that only 1-2 times of 4-degree motion per day can improve the joint function exercise, thereby optimizing the theoretical value. At present, there is only a consensus in the domestic medical field, but no guidelines. Therefore, what is missing between the consensus and the clinical guidelines is similar medical evidence. Through this mobile-based visual technology, more follow-up data can be obtained, so that hospitals can realize the inquiry and mining of related data outside the hospital.

[0090] The above merely describes preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for sports training based on a visual AI model, characterized in that, The application relates to a visual AI model-based exercise training method. User information of a user to be trained is collected through a mobile terminal; Training data information of the user is obtained according to the user information, the training data information including the number of exercise training times of the user, current planned exercise progress information and current exercise progress information; dynamic training data in the exercise training process of the user is collected in real time, the dynamic training data including joint angles in the exercise training process of the user and conclusion data given by a visual AI model to the joint angles; an actual exercise curve of the user is drawn according to the collected joint angles and conclusion data; Actual exercise progress information after the exercise training of the user is calculated according to the actual exercise curve, the number of training times of the user, the current planned exercise progress information and the current exercise progress information; The current planned exercise progress information is corrected according to the actual exercise progress information and the number of exercise training times, so that corrected planned exercise progress information is obtained to guide the exercise training of the user. According to the visual AI model-based exercise training method, the actual exercise curve of the user is drawn according to the collected joint angles and conclusion data, including: functional recovery index data related to the user information is obtained; an actual exercise curve with P value is calculated according to a statistical algorithm, the horizontal coordinate being time and the vertical coordinate being an exercise state, the functional recovery index data including at least the recovery angle of a disabled joint, muscle strength, activity ability and imaging evaluation results. 2.The visual AI model-based athletic training method of claim 2, wherein, The actual exercise progress information after the exercise training of the user is obtained according to the actual exercise curve, the number of training times of the user, the current planned exercise progress information and the current exercise progress information, including: functional recovery index data after the exercise training of the user is obtained according to the actual exercise curve; the functional recovery index data after the exercise training of the user is compared with dynamic training data in the current exercise progress information; the actual exercise progress information after the exercise training of the user is obtained according to the comparison result, the number of training times of the user and the current planned exercise progress information. 3.The visual AI model-based exercise training method of claim 1, wherein, The current planned exercise progress information is corrected according to the actual exercise progress information and the number of exercise training times, so that corrected planned exercise progress information is obtained to guide the exercise training of the user, including: planned exercise training times, planned exercise training time and planned exercise training index data in the current planned exercise progress information are obtained; an activity angle of a disabled joint after actual exercise training is obtained according to the actual exercise progress information; the planned exercise training times and the planned exercise training time are corrected according to the activity angle and the number of exercise training times, so that new planned exercise training time, planned exercise training times and planned exercise training index data are obtained as the corrected planned exercise progress information. 4.The visual AI model-based exercise training method of claim 4, wherein, According to the activity angle and the number of exercise training times, the planned exercise training times and the planned exercise training time are corrected, including: searching the preset rule database for the exercise training times and the exercise training time corresponding to the activity angle and the number of exercise training times respectively; comparing the exercise training times and the exercise training time with the planned exercise training times and the planned exercise training time, and calculating the difference of the two respectively; adding half of the difference of the two to the planned exercise training times and the planned exercise training time to obtain the new planned exercise training time and the planned exercise training times. 5.The visual AI model-based exercise training method of claim 1, wherein, Real-time acquisition of dynamic training data in the user's exercise training process includes: real-time acquisition of joint angles and time of maintaining the joint angles during user exercise training; data correlation according to the combination of generative model and recognition capture; comparison of joint angles and time of maintaining the joint angles during user exercise training according to the data in the preset rule database, and judgment analysis of joint angles and time of maintaining the joint angles during user exercise training; joint angles conforming to the data in the preset rule database and conclusion data given by the visual AI model for the joint angles are used as the dynamic training data in the user's exercise training process. 6.The visual AI model-based exercise training method of claim 1, wherein, The method further includes: acquiring training data of each exercise training of the user through a server, the training data including angles, positions, times and qualified times of disabled joint exercise; classifying the training data; training the classified training data, actual exercise curve and actual exercise progress information to obtain key data affecting exercise training, and establishing a corresponding exercise database according to the key data. 7.The visual AI model-based athletic training method of claim 7, wherein, Training the classified training data, actual exercise curve and actual exercise progress information includes: detecting the disabled joint of the user to obtain the disability information of the joint of the user; comparing the disability information with at least an exercise knowledge graph or a standard exercise motion database to obtain a planned exercise value; deriving the cycle and training content of the required exercise according to the planned exercise value to obtain the angle of the joint required for the user to exercise every day, and using it as the key data affecting exercise training to guide the user to exercise.

8. A visual AI model-based sports training device, characterized in that, It includes: An identity acquisition module for acquiring user information of a user to be trained; An information acquisition module for acquiring training data information of the user according to the user information; A data acquisition module for real-time acquisition of dynamic training data in the user's exercise training process; A curve drawing module for drawing the actual exercise curve of the user according to the acquired joint angle and conclusion data; An information calculation module for calculating the actual exercise progress information of the user after exercise training according to the actual exercise curve, the training times of the user, the current planned exercise progress information and the current exercise progress information; An information correction module for correcting the above-mentioned current planned exercise progress information according to the actual exercise progress information and the number of exercise training times to obtain the corrected planned exercise progress information to guide the user to exercise. 9.A sports training system based on visual AI model, characterized in that, The mobile terminal and the server are included, wherein the mobile terminal collects user information of a user to be trained based on a visual AI model, the mobile terminal is also used for obtaining training data information of the user according to the user information and is used for collecting dynamic training data in a real-time manner during a training process of the user; the server is used for drawing an actual motion curve of the user according to collected joint angles and conclusion data, the server is also used for calculating actual motion progress information of the user after training according to the actual motion curve, training times of the user, current planned motion progress information and current motion progress information, and the server is also used for correcting the current planned motion progress information according to the actual motion progress information and the training times to obtain corrected planned motion progress information to guide the user to perform training.

Images