Interactive physical education method based on large model

Abstract

The invention relates to the technical field of physical education teaching, in particular to an interactive physical education teaching method based on a large model, and aims to realize real-time monitoring and personalized training guidance of the motion state of a student through an advanced sensor technology and a machine learning model. The method comprises the following steps: firstly, collecting motion data of a student in real time through a camera and a depth sensor, including motion posture, speed, response time, heart rate and other information; then, the collected data are input into a pre-trained large model, the model comprises a convolutional neural network, a support vector machine and a random forest model which are respectively used for evaluating the action standard, the exercise intensity and the fatigue degree of students, the large model automatically generates personalized training items, and the exercise postures of the students are captured in real time; according to the motion deviation correcting system and method, by comprehensively applying various technical means, the scientificity and individuation level of physical education are improved, and the motion deviation correcting system and method have wide application prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of physical education teaching, and particularly to an interactive physical education teaching method based on a large model. Background Art

[0002] With the continuous development of the physical education teaching field, various physical education teaching equipment and technologies have been widely used. However, there are still some problems in the actual use of these products. For example, the currently available physical education teaching equipment on the market usually adopts traditional mechanical or single-function control systems, which have a slow response speed and are difficult to achieve precise control in the face of complex teaching requirements. This leads to problems such as low teaching efficiency and inconvenient operation in some scenarios.

[0003] To improve performance, some manufacturers have tried to increase the automation level by adding electronic feedback devices. However, such improvements often face problems of poor coordination between electrical components and mechanical parts, resulting in poor overall system stability and high maintenance costs.

[0004] After retrieval, a sports teaching system with the publication number CN113689591B was disclosed, and the publication date was April 19, 2024. This design uses an identity verification module and a timing module, which can control the switching between training modes and assessment modes and count the daily training duration of people. Although this system can improve the training quality and monitoring effect to a certain extent, due to the lack of real-time feedback and personalized training functions, it cannot dynamically adjust the training content according to the actual situation of students, resulting in limited training effects. In addition, this system mainly relies on video recording and post-event analysis, lacking real-time interactivity and instant feedback, and cannot meet the requirements for real-time and interactivity in modern teaching.

[0005] For example, in traditional physical education teaching during basketball shooting training, it is difficult for coaches to simultaneously monitor the takeoff angle, wrist force trajectory, and overall body coordination of trainees; in football shooting teaching, it is impossible to quantitatively analyze the dynamic relationship between the standing position of the supporting foot and the swing amplitude of the leg. In gymnastics teaching, when trainees perform somersaults and twists, coaches cannot real-time capture micro-posture data such as the spinal curvature, resulting in lagging action correction.

[0006] The above problems indicate that the traditional physical education teaching equipment and technologies on the market are difficult to effectively meet the new requirements for precise control, real-time feedback, and personalized training under complex teaching requirements. Summary of the Invention

[0007] Based on the above purposes, the present invention provides an interactive physical education teaching method based on a large model, including the following steps: S1: Real-time collect the motion data of students through a camera and a depth sensor, including the action postures, speeds, reaction times, and heart rates of students; The cameras are installed at multiple locations in the training ground, and the depth sensors are installed on key teaching equipment. The cameras and depth sensors are connected to the central processor through data acquisition cables to transmit the collected data in real time. The central processor preprocesses the collected motion data, including filtering, denoising, and normalization. S2: The central processor inputs the preprocessed data into a pre-trained large model to evaluate the student's motion state, including action standardization, exercise intensity, and fatigue level. S3: According to the evaluation results of the student's motion state, the large model automatically generates personalized training programs. S4: Real-time capture the student's motion posture, analyze whether the action is standardized, and provide real-time feedback. S5: Generate interactive sports games based on the student's basic motor ability and interest preference.

[0008] Preferably, the large model described in S2 includes a convolutional neural network model, a support vector machine model, and a random forest model. The convolutional neural network model, the support vector machine model, and the random forest model are respectively used to evaluate action standardization, exercise intensity, and fatigue level.

[0009] Preferably, the input of the convolutional neural network model is the motion data collected in real time, including action posture image data, and the output is the action standardization score ; The algorithm is as follows: ; Where N is the number of data samples, is the weight, is the convolution kernel, is the input image, is the bias term, is the activation function, and i is the index; The input of the support vector machine model (SVM) is the exercise intensity data collected in real time, including speed, acceleration, and heart rate, and the output is the exercise intensity evaluation ; The algorithm is as follows: ; Where, is the Lagrange multiplier, is the sample label, is the kernel function, is the bias term, and i is the index; The input of the random forest model (RF) is the fatigue level data collected in real time, including heart rate, action posture stability, etc., and the output is the fatigue level evaluation ; The algorithm is as follows: ; Among them, is the number of decision trees, is the output of the th decision tree.

[0010] Preferably, the specific steps for generating personalized training items in S3 are as follows: S3.1: Evaluate the standardness of actions. If the action standardness score is lower than the threshold , generate corrective action exercises; S3.2: Evaluate the exercise intensity. If the exercise intensity is lower than the threshold , generate training to increase intensity; S3.3: Evaluate the fatigue level. If the fatigue level is higher than the threshold , generate restorative training.

[0011] Preferably, the corrective action exercises, training to increase intensity, and restorative training all include multiple specific actions, and each action includes an action name, an action description, a standard action demonstration video, and a scoring criterion; Among them, the corrective action exercises, training items to increase intensity, and restorative training items retrieve their corresponding action names and action descriptions from a pre-stored database, and retrieve the standard action demonstration videos and scoring criteria from a pre-stored video library to generate corrective action exercises, training to increase intensity, or restorative training.

[0012] Preferably, in S4, it specifically includes the following steps: S4.1: Real-time collect the action data of students through the camera and depth sensor in S1, including joint positions, movement trajectories, and speeds. The depth sensor captures the action details of students, and the camera records the movement postures of students from multiple angles; S4.2: Compare the real-time collected action data with the preset standard action demonstrations obtained in S3, and identify the action deviations of students through a pose estimation algorithm. The input of the pose estimation algorithm is the real-time collected action data, and the output is the quantization value of the action deviations; The pose estimation algorithm includes a three-level decomposition architecture: 1. Macro-action stage division: Decompose continuous actions into a preparation period (muscle pre-activation), an execution period (power burst), and a finishing period (posture stabilization) 2. Key frame extraction: Based on the energy mutation point recognition technology, accurately capture the action turning moments (such as the badminton smash hitting point) 3. Micro-joint group analysis: Establish a human motion chain model to distinguish active joints (such as the shoulder joint during throwing) from passive joint groups; S4.3: Generate real-time feedback information based on the quantified value of the motion deviation. The feedback information includes text prompts, voice prompts, and video demonstrations, which are played in real time through multimedia devices to guide students to correct their motions. The input is the quantified value of the motion deviation, and the output is the real-time feedback information; S4.4: The feedback information is presented in real time through monitors, speakers, and smart wearable devices installed in the training venue. The screen shows a comparison diagram of the student's current motion posture and the standard posture. The speaker plays voice prompts, and the smart wearable device vibrates or flashes lights to attract the student's attention.

[0013] Preferably, in S4.2, it specifically includes the following steps: S4.2.1: Compare the joint positions in the real-time collected motion data with the joint positions of the preset standard motion to calculate the joint position deviation; S4.2.2: Compare the motion trajectories in the real-time collected motion data with the motion trajectories of the preset standard motion to calculate the motion trajectory deviation; S4.2.3: Compare the speeds in the real-time collected motion data with the speeds of the preset standard motion to calculate the speed deviation; S4.2.4: Comprehensive deviation calculation: Combine the joint position deviation, motion trajectory deviation, and speed deviation to calculate the final quantified value of the motion deviation.

[0014] Preferably, the weights of the joint position deviation, motion trajectory deviation, and speed deviation are 0.4, 0.3, and 0.3 respectively. The comprehensive deviation value is calculated through the following formula: .

[0015] Preferably, in S5, it specifically includes the following steps: S5.1: Evaluate the student's basic motor ability based on the motion data collected in S1; S5.2: Analyze the student's interest preferences through the student's historical records. The historical records are the game types, training times, frequencies, participation degrees, training data, and feedback situations selected by the student during the historical training process. The interest preference analysis is realized through a data mining algorithm. The input of the data mining algorithm is the student's historical records, and the output is the preference scores of the student for different sports events; S5.3: Generate personalized interactive sports games based on the evaluation results of the basic motor ability and interest preferences. The game generation algorithm is realized through a multi-objective optimization model. The input is the evaluation result of the student's basic motor ability and the interest preference scores, and the output is personalized game content, including game types, difficulty settings, game scenes, and interaction methods.

[0016] Preferably, the basic motor ability assessment includes muscle strength assessment, endurance assessment, and coordination assessment. Among them, the muscle strength of students is evaluated by analyzing their movement postures and speeds. The specific formula is: ; Among them, the number of accurate movements refers to the number of times a student has a correct posture when completing a movement, and the speed score refers to the ratio of the speed at which a student completes a movement to the standard value; The endurance of students is evaluated by analyzing their movement time, heart rate, and degree of fatigue. The specific formula is: ; Among them, the movement time refers to the time a student continuously moves, the heart rate stability refers to the fluctuation range of the heart rate of a student during movement, and the fatigue index is the degree of fatigue evaluated by a large model.

[0017] The flexibility of students is evaluated by analyzing their movement postures and reaction times. The specific formula is: ; Among them, the movement range refers to the range of activities of a student when completing a movement, and the reaction time score refers to the ratio of the reaction time of a student when completing a movement to the standard value; The coordination of students is evaluated by analyzing their movement postures, speeds, and reaction times. The specific formula is: ; Among them, the number of accurate movements refers to the number of times a student has a correct posture when completing a movement, the speed score refers to the ratio of the speed at which a student completes a movement to the standard value, and the reaction time score refers to the ratio of the reaction time of a student when completing a movement to the standard value.

[0018] Advantages of the present invention: 1. The present invention can comprehensively and accurately evaluate the movement state of students, including movement standardization, exercise intensity, and degree of fatigue, by using a camera and a depth sensor to collect multi-dimensional movement data such as students' movement postures, speeds, reaction times, and heart rates in real time. Compared with the traditional teaching method that relies on the subjective judgment of coaches, this data-based evaluation is more objective and accurate, can better grasp the actual situation of students during training, and provides a reliable basis for subsequent teaching guidance; 2. The present invention can automatically generate personalized training programs based on the student's sports status assessment results. For example, if the student's movement standard score is lower than the threshold, the system will generate corrective movement exercises; if the exercise intensity is lower than the threshold, it will generate increased intensity training; if the fatigue level is higher than the threshold, it will generate recovery training. This personalized training arrangement can specifically solve the problems that students have in sports, improve the pertinence and effectiveness of training, and enable students to improve their sports skills and physical fitness levels more quickly during training; 3. The present invention can capture students' movement postures in real time, analyze whether the movements are standardized, and provide real-time feedback, including text prompts, voice prompts, and video demonstrations. This real-time feedback mechanism enables students to understand their movement deviations in time during training, and make adjustments and improvements based on feedback information, which enhances the interactivity of teaching, enables students to participate in training more actively, and improves the enthusiasm and effectiveness of training; 4. The system of the present invention can generate personalized interactive sports games based on students' basic sports ability and interest preferences. This way of integrating game elements into physical education can greatly enhance the fun of training, stimulate students' learning interest and enthusiasm for participation, enable students to exercise in a relaxed and pleasant atmosphere, and improve students' love and persistence for sports; 5. The present invention reduces the time and energy that coaches spend on individual guidance of each student during the teaching process through the automated evaluation and training project generation of a large model, allowing coaches to devote more energy to the formulation of teaching strategies and the optimization of the overall teaching plan, thereby improving teaching efficiency; 6. The present invention can help students comprehensively improve their athletic ability through comprehensive assessment and training of basic athletic abilities such as muscle strength, endurance, flexibility and coordination, lay a solid foundation for students to participate in various sports and activities, and promote the all-round development of students' physical fitness. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 is a flow chart of the steps of the method of the present invention; Figure 2 is a flow chart of the steps of method S4 of the present invention; Figure 3 This is a flow chart of the steps of method S5 of the present invention. DETAILED DESCRIPTION

[0021] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the accompanying drawings are only for more specifically describing the embodiments and are not intended to specifically limit the present invention.

[0022] Embodiment 1: Please refer to Figures 1 - 3 , the embodiment of the present invention provides an interactive sports teaching method based on a large model, which is characterized by including the following steps: S1: Real-time collect the motion data of students through cameras and depth sensors, including the action postures, speeds, reaction times, and heart rates of students; The cameras are installed at multiple positions in the training venue, and the depth sensors are installed on key teaching equipment. The cameras and depth sensors are connected to the central processor through data acquisition cables to transmit the collected data in real time; The central processor preprocesses the collected motion data, including filtering, denoising, and normalization processing; In this embodiment, S1 further includes: collecting the muscle activation state through a flexible strain sensor array, obtaining the plantar pressure distribution through an intelligent insole system, overcoming visual occlusion through a millimeter-wave radar group to achieve three-dimensional action capture, and monitoring the core body temperature change through an infrared thermal imager; the multi-source data is input into the preprocessing module after spatio-temporal alignment; S2: The central processor inputs the preprocessed data into a pre-trained large model to evaluate the motion state of students, including action standardization, exercise intensity, and fatigue degree; S3: According to the evaluation results of the motion state of students, the large model automatically generates personalized training items; S4: Real-time capture the motion postures of students, analyze whether the actions are standard, and provide real-time feedback; S5: Generate interactive sports games according to the basic sports ability and interest preferences of students to improve the participation and interest of students.

[0023] In a possible implementation, first, multiple cameras and depth sensors are installed at different positions in the training venue to collect students' motion data in real time. These devices are connected to the central processor through data cables to ensure real-time data transmission. The cameras mainly capture the students' movement postures and trajectories, while the depth sensors capture detailed information such as joint positions, speeds, accelerations, reaction times, and heart rates. The data collection in this stage provides comprehensive basic information for subsequent analysis. Through filtering, denoising, and normalization of the data, the accuracy and consistency of the data are ensured, avoiding the influence of noise interference and outliers. Flexible sensors are embedded at key muscle group positions in the sports uniform at 5 cm intervals, with a sampling frequency of 500 Hz; the smart insole contains 128 pressure sensing units to form a heat map of foot force; the millimeter-wave radar group covers the training area in a triangular layout, with a minimum detection distance of 0.2 m.

[0024] Furthermore, after data preprocessing, the central processor inputs this data into a pre-trained large model. The large model uses deep learning algorithms (such as convolutional neural networks, support vector machines, etc.) to evaluate the students' motion states, including action standardization, exercise intensity, and fatigue level. By analyzing the deviation of the students' movement postures from the standard actions, the standardization of the actions is evaluated; by speed, acceleration, and heart rate data, the intensity of the exercise is evaluated; by heart rate fluctuations and action stability, the fatigue level is evaluated. This stage provides an accurate basis for the generation of subsequent personalized training programs.

[0025] Furthermore, according to the results evaluated by the large model, the system automatically generates personalized training items. If the student's action standardization score is low, the system will automatically recommend action correction exercises; if the exercise intensity is low, it is recommended to increase the intensity training; if the student shows fatigue, the system will recommend restorative training. The generation of personalized training items helps to train according to the specific problems of the students, avoiding a one-size-fits-all training method and improving the training effect and efficiency.

[0026] Furthermore, when the students are training, the system captures the students' actions in real time through cameras and depth sensors and compares them with the standard actions for analysis. If non-standard actions are found, the system will provide feedback in a timely manner to remind the students to adjust their postures or improve their actions. The feedback can be displayed in various forms such as text, voice, or video to ensure that the students can understand and correct in a timely manner. This real-time feedback mechanism helps the students to continuously improve the standardization of their actions during training, thereby improving the exercise effect.

[0027] Furthermore, based on the students' basic motor abilities and interest preferences, the system generates interactive sports games to further improve students' participation and interest. By evaluating the students' motor abilities, the system can adjust the difficulty of the games to make them challenging but not overly difficult, thus stimulating students' enthusiasm. In addition, the gamified training method enables students to exercise in a relaxed and pleasant atmosphere, enhancing the fun and sustainability of the exercise.

[0028] In the embodiment of the present invention, the large models described in S2 include a convolutional neural network model, a support vector machine (SVM) model, and a random forest model. The convolutional neural network model, the support vector machine (SVM) model, and the random forest model are respectively used to evaluate the action standardization, exercise intensity, and fatigue level.

[0029] In a possible implementation manner, the CNN model is mainly used to evaluate the students' action standardization. By processing the data from the camera and depth sensor, the CNN can extract features such as the joint positions, postures, and movement trajectories during the students' exercise. The convolutional layer of the CNN extracts spatial features from this data, thereby classifying and identifying the students' actions and comparing them with the standard actions. The CNN can effectively identify the deviations of the students when performing actions, such as non-standard actions and abnormal joint angles. The output result of this model will be transmitted to the subsequent decision-making system to provide evaluation data on action standardization.

[0030] Furthermore, the SVM model is used to evaluate the students' exercise intensity. The exercise intensity not only depends on the students' exercise speed and acceleration but is also closely related to physiological data such as heart rate. The SVM model uses this multi-dimensional data to divide different categories of exercise intensity (such as low intensity, medium intensity, and high intensity) by constructing a high-dimensional feature space. This model can accurately identify whether the students' current exercise intensity meets the target training intensity and can provide timely feedback on the students' exercise status according to the training plan. For example, when the SVM model detects that the students' exercise intensity is lower than the preset target, the system can suggest increasing the training intensity.

[0031] Furthermore, the random forest model is used to evaluate the students' fatigue level. The RF model analyzes whether the students are in a fatigued state by integrating multiple decision trees and combining data such as the students' heart rate changes, exercise stability, and action performance. Each decision tree independently makes a prediction on the fatigue level, and finally, a comprehensive evaluation result is obtained through a majority voting method. The advantage of the RF is that it can handle high-dimensional and noisy data and has strong adaptability to the non-linear relationships between features, thus accurately determining whether the students need to rest or reduce the exercise load.

[0032] By organically combining the CNN, SVM, and RF models in step S2, the system can comprehensively evaluate the student's exercise status from multiple dimensions and form a comprehensive and accurate evaluation result. First, CNN focuses on the standardization of movements to help students correct their movement postures; SVM pays attention to the exercise intensity to ensure that the training intensity is suitable for the student's physical condition and training goals; RF evaluates the degree of fatigue to prevent students from getting injured or being inefficient due to overtraining.

[0033] In the embodiment of the present invention, the input of the convolutional neural network model (CNN) is the exercise data collected in real time, including action posture image data (2D or 3D images), and the output is the action standardization score. ; The algorithm is as follows: ; Among them, is the number of data samples, is the weight, is the convolutional kernel, is the input image, is the bias term, is the activation function; The input of the support vector machine model (SVM) is the exercise intensity data collected in real time, including speed, acceleration, and heart rate, and the output is the exercise intensity evaluation. ; The algorithm is as follows: ; Among them, is the Lagrange multiplier, is the sample label, is the kernel function, is the bias term; The input of the random forest model (RF) is the fatigue degree data collected in real time, including heart rate, action posture stability, etc., and the output is the fatigue degree evaluation. ; The algorithm is as follows: ; Among them, is the number of decision trees, is the output of the th decision tree.

[0034] In a possible implementation manner, the core of data input and collection is to collect the student's exercise data in real time from multiple sensors, including physiological data such as action posture images (2D or 3D images), speed, acceleration, and heart rate. Each data type corresponds to the input of a different model.

[0035] Further, the input of the CNN is the motion posture image data collected in real time. This data is processed through the convolutional layer, and the spatial features are extracted through the convolutional kernel and the activation function (ReLU) for pattern recognition. The output in the formula represents the action standard score, which measures the deviation between the student's action and the standard action, and accurately reflects the student's action standard. The score output by the CNN model is used for subsequent decision-making. If the action standard is poor, the system can give corresponding feedback to help the student correct the action. This step helps to accurately identify the action deviation during exercise and ensure that the student can train according to the standard action.

[0036] Further, the input of the SVM model is the exercise intensity data (such as speed, acceleration, heart rate, etc.). Its core is to construct a high-dimensional space through the Lagrange multiplier and the kernel function to classify different exercise intensities. The output of the SVM model represents the exercise intensity evaluation. The system determines whether the current training intensity meets the student's training needs through this evaluation. If the intensity is too low or too high, the system will give adjustment suggestions. The evaluation of exercise intensity can guide the student to adjust the training intensity in real time, ensure the training effect and avoid overtraining. Combining the evaluation of action standard, the system can dynamically adjust the student's training plan.

[0037] Further, the input of the random forest model is the fatigue degree data, including heart rate, action posture stability, etc. The RF comprehensively evaluates the fatigue degree through the integration of multiple decision trees and outputs the fatigue degree score . When the fatigue degree is too high, the system will suggest that the student rest or reduce the training intensity to avoid the harm caused by overtraining. The advantage of the RF model lies in its ability to process high-dimensional data, which can judge the student's fatigue state through diverse features (such as heart rate, action stability) and provide a basis for personalized training programs.

[0038] These three models work together in the system in an independent but interrelated manner. The specific steps are as follows: In the first step, the system first evaluates the student's action standard through the CNN model to ensure that the student meets the correct posture requirements when performing the action.

[0039] It is understandable that the CNN model, SVM model, and random forest model all belong to machine learning algorithm models. In this embodiment, the machine learning model may include decision trees, naive Bayes classification, nearest neighbors, neural networks, convolutional neural networks, generative adversarial networks, support vector machines, linear regression, logistic regression, random forests, and / or gradient boosting algorithms, all of which can be adaptively replaced. Preferably, the machine learning algorithm is organized to process an input with a high dimension into an output with a much lower dimension. This machine learning algorithm is called "intelligent" because it can be "trained". The algorithm can be trained using records of training data. The records of training data may include training input data and corresponding training output data. The training output information of the training information record may be the result expected to be produced by the machine learning algorithm when the training input data of the same training data record is given as input. The deviation between the expected result and the actual result produced by the algorithm can be observed and rated through a "loss function". The loss function can be used as feedback for adjusting the parameters of the internal processing chain of the machine learning algorithm. For example, the parameters can be adjusted with the optimization goal of minimizing the value of the loss function, which is produced when all the training input data is fed into the machine learning algorithm and the result is compared with the corresponding training output data. The result of this training may be that, given a relatively small number of training data records as "ground truth", the machine learning algorithm can perform its work well for a large number of input data records that are many orders of magnitude higher. Therefore, the simulation model may include at least one algorithm and model parameters. The parameters of the simulation model can be generated by using at least one artificial neural network.

[0040] In the second step, the SVM model evaluates whether the current training intensity is appropriate based on the student's exercise intensity data to ensure the effectiveness and safety of the training.

[0041] In the third step, the RF model evaluates the student's fatigue level in real time to prevent training injuries caused by excessive fatigue.

[0042] This integrated multi-model approach can not only comprehensively monitor the student's exercise status from different dimensions but also make personalized adjustments based on real-time feedback, maximizing the training effect and reducing the risk of injury. The outputs of each model complement each other, forming a closed-loop feedback system to achieve precise and dynamic training guidance.

[0043] For example, when the CNN identifies that the student's movements are not standard, the system can prompt the student to adjust their posture; when the SVM determines that the student's exercise intensity is insufficient, the effect can be optimized by appropriately increasing the training intensity; when the RF detects signs of fatigue in the student, it can automatically recommend reducing the load or increasing the rest time. Such multi-dimensional feedback helps to comprehensively improve the training quality of students and ensure the safety and personalization of sports training.

[0044] Through the comprehensive application of this model, the training process of students is not only more intelligent, but also can be adjusted in real time according to the actual needs of individuals, so as to achieve a more efficient and safer sports training effect.

[0045] In the embodiment of the present invention, the specific steps for generating personalized training items in S3 are as follows: S3.1: Evaluate the action standardization. If the action standardization score is lower than the threshold , generate corrective action exercises; S3.2: Evaluate the exercise intensity. If the exercise intensity is lower than the threshold , generate training to increase the intensity; S3.3: Evaluate the fatigue level. If the fatigue level is higher than the threshold , generate restorative training.

[0046] In a possible implementation manner, the action standardization evaluation (S3.1): The action standardization evaluation is based on the analysis result of the CNN model on the student's movement posture image. If the system evaluates that the student's action standardization score is lower than the set threshold , targeted corrective action exercises will be generated. The content of the corrective action exercises includes: Repeated practice: Repeat the standard action to strengthen the muscle memory of the correct posture.

[0047] Action correction tips: Provide specific correction suggestions according to the system's recognition of action deviations, such as adjusting the foot position, changing the swing angle, etc.

[0048] The key at this stage is to help students correct the mistakes in their actions, ensure the accuracy of their actions, and avoid injuries caused by incorrect postures.

[0049] Furthermore, the exercise intensity evaluation analyzes the student's exercise data (such as speed, acceleration, heart rate, etc.) through the SVM model. If the evaluation result shows that the student's exercise intensity is lower than the set threshold , the system will generate training to increase the intensity. The measures to increase the training intensity include: Increase the amount of exercise: Increase the exercise time or increase the exercise speed and intensity.

[0050] Adjust the training plan: Design more challenging training content, such as interval training, high-intensity interval training (HIIT), etc., to encourage students to gradually improve their physical fitness level.

[0051] This step ensures that the training intensity of students meets their physical fitness needs, avoiding ineffective training due to too low intensity and thus affecting physical fitness improvement.

[0052] Furthermore, the fatigue degree assessment comprehensively judges the physiological data of students (such as heart rate fluctuations, movement stability, etc.) through an RF model. If the fatigue degree of a student is higher than the set threshold , the system will generate restorative training. The measures of restorative training include: Low-intensity training: such as slow stretching exercises, yoga, meditation, etc., to help students relieve muscle fatigue.

[0053] Rest and recovery: It is recommended to take appropriate rest and avoid continuing high-intensity exercise to prevent injuries caused by excessive fatigue.

[0054] The fatigue degree assessment ensures the safety and training effect of students by monitoring the physical state of students in a timely manner and preventing students from continuing training when they are overly fatigued.

[0055] In the embodiment of the present invention, the corrective action practice, intensity-increasing training, and restorative training all include multiple specific actions, and each action includes an action name, an action description, a standard action demonstration video, and a scoring criterion; Among them, the corrective action practice, intensity-increasing training items, and restorative training items retrieve their corresponding action names and action descriptions from a pre-stored database, and retrieve the standard action demonstration video and scoring criterion from a pre-stored video library to generate corrective action practice, intensity-increasing training, or restorative training.

[0056] In a possible implementation manner, the core of the corrective action practice is to provide personalized correction suggestions by identifying the movement posture deviation of students. The system retrieves corrective actions related to the standard assessment result of the student's current action from a pre-stored database. Each corrective action includes: Action name: Clearly define the name of the action, such as "Standard Squat Correction". Action description: Concisely and detailedly describe the action essentials to ensure that students understand how to make corrections. Standard action demonstration video: Demonstrate the standard action through a video to help students visually understand the correct action. Scoring criterion: List the action standards that students need to meet, such as the knees not exceeding the toes, the back being straight, etc., to help students conduct self-assessment or teacher scoring.

[0057] Furthermore, the training programs with increased exercise intensity generate appropriate training plans by evaluating the current exercise intensity of students. The system selects suitable training actions from the database, including: Action name: such as "High-intensity interval running" or "Explosive jump training". Action description: Describe the execution method and precautions of the action to ensure that students understand how to gradually increase the exercise intensity. Standard action demonstration video: Show standardized high-intensity actions to ensure that students follow the correct training methods in practice. Scoring criteria: such as the number of completed actions, duration, action accuracy, etc., to help students adjust the training intensity according to the feedback.

[0058] Furthermore, when it is evaluated that the fatigue level of the student is relatively high, the system will recommend restorative training programs. The action designs of these programs focus on low intensity, relaxation, and recovery. The specific steps include: Action name: such as "Yoga stretching" or "Deep breathing recovery training". Action description: Guide students on how to recover muscle fatigue through soothing actions, adjust the breathing rhythm, and relieve post-exercise discomfort. Standard action demonstration video: Help students understand how to perform relaxation and soothing training through standard restorative action demonstration videos. Scoring criteria: such as action accuracy, relaxation level, etc., to help students evaluate whether they have achieved the goals of restorative training.

[0059] These training programs ensure the efficiency and systematicness of each training step through the preset standardized content in the database. Specifically: After the student completes the corrective action, the system will judge whether the intensity can be increased based on the evaluation result of the student's action standardization. If the action is not yet standard, the system will first recommend corrective training actions to ensure that the student increases the training intensity on the premise of accurate actions.

[0060] When the system finds that the exercise intensity of the student exceeds their tolerance and the fatigue level is relatively high, restorative training will be recommended. At this time, the intensity increase training and restorative training achieve dynamic balance through the fatigue assessment system to avoid overtraining of students.

[0061] In the embodiment of the present invention, in S4, it specifically includes the following steps: S4.1: Real-time collect the action data of the student through the camera and depth sensor in S1, including joint positions, movement trajectories, and speeds. The depth sensor captures the action details of the student, and the camera records the movement postures of the student from multiple angles; S4.2: Compare the real-time collected action data with the preset standard action demonstration obtained in S3, and identify the action deviation of the student through the pose estimation algorithm. The input of the pose estimation algorithm is the real-time collected action data, and the output is the quantified value of the action deviation; The pose estimation algorithm includes a three-level decomposition architecture: 1. Macro-action stage division: Decompose continuous actions into a preparation period (muscle pre-activation), an execution period (power burst), and a finishing period (posture stabilization). The action decomposition adopts a hierarchical processing mechanism: First, divide the action stages through inertial sensor data, and mark the starting point of the force application period when a 300% sudden increase in acceleration is detected; Second, use the inter-frame difference method to identify key postures, such as the highest point position during a basketball shot; Finally, through inverse kinematics calculation, distinguish the linkage relationship between the shoulder joint (active) and the wrist joint (passive). 2. Key frame extraction: Based on the energy mutation point recognition technology, accurately capture the moment of action transition (such as the hitting point of a badminton smash). 3. Microscopic joint group analysis: Establish a human motion chain model to distinguish active joints (such as the shoulder joint during throwing) and passive joint groups. S4.3: Generate real-time feedback information according to the quantified value of the action deviation. The feedback information includes text prompts, voice prompts, and video demonstrations, which are played in real time through multimedia devices to guide students to correct their actions. The input is the quantified value of the action deviation, and the output is real-time feedback information.

[0062] S4.4: The feedback information is presented in real time through monitors, speakers, and smart wearable devices installed in the training venue. The screen shows a comparison diagram of the student's current action posture and the standard posture. The speaker plays voice prompts, and the smart wearable device vibrates or flashes lights to attract the student's attention.

[0063] In a possible implementation, first, the camera and depth sensor in S1 are used to collect the student's motion data in real time, mainly including the student's joint positions, motion trajectories, and motion speeds. The camera is responsible for recording the student's motion postures from multiple angles, and the depth sensor can capture more detailed action details to ensure the accuracy and comprehensiveness of the data. This step is the basis of the whole process, providing real-time data input, which will become the core of subsequent processing and analysis. Whether it is subsequent action deviation recognition or action feedback generation, it depends on these accurate real-time data.

[0064] Furthermore, compare the real-time collected action data with the preset standard action demonstration in the S3 stage. By using the pose estimation algorithm, the system can identify the student's action deviation. The input of this algorithm is the real-time collected action data, and the output is the quantified value of the action deviation, which reflects the difference between the student's current action and the standard action. This step directly connects S4.1 and S4.3. By comparing the difference between the real-time collected data and the standard demonstration, the action deviation is accurately quantified. This quantified result provides a specific data basis for the subsequent feedback mechanism.

[0065] Specifically, the pose estimation algorithm first needs to process the raw data from the camera and depth sensor. The image data collected by the camera usually contains the 2D or 3D pose information of the student, while the depth sensor provides more accurate spatial position information, such as the 3D coordinates of the joints. The algorithm needs to extract the key point data of the student from these images through feature extraction techniques (such as HOG, SIFT, etc.), including joint positions, relative positions of body parts, etc. The depth sensor provides higher-precision 3D coordinate information, which can capture the subtle changes in the student's movements.

[0066] The pose estimation algorithm uses a graph convolutional network (GCN) to estimate the body joint coordinates of the student. Based on the input image or sensor data, the algorithm first infers the student's pose at the current moment through a trained model. The estimation result of each joint position will obtain a 3D coordinate (x, y, z) and a related confidence value to ensure the accuracy of the estimation result.

[0067] After completing the pose estimation, the algorithm compares the real-time obtained student pose with the preset standard actions. The standard actions can be ideal poses analyzed and optimized by sports experts, usually including the target positions and angles of a series of joints. To calculate the deviation, the algorithm measures the difference between the actual position and the standard position of each joint. Commonly used methods include distance calculation methods such as Euclidean distance and cosine similarity.

[0068] After calculating the joint position differences, the algorithm quantifies the deviation of each joint and comprehensively obtains the deviation value of the overall pose. This deviation quantification value not only considers the error of the joint position but also the deviation of the movement trajectory and angle change. Through a weighting algorithm, different importance weights can be set for each joint or movement stage, so as to obtain a comprehensive and accurate pose deviation measurement.

[0069] Finally, the algorithm compares the calculated pose deviation value with the set threshold. If the deviation exceeds the preset allowable range, the algorithm will trigger subsequent feedback mechanisms. These feedback mechanisms can inform the student how to correct the pose through text, voice, and video demonstrations. In addition, through multiple learning and training, the algorithm can continuously optimize the accuracy of pose estimation and deviation quantification, so as to provide more and more accurate real-time guidance.

[0070] Furthermore, based on the identified action deviation quantification value, the system generates real-time feedback information. The feedback information not only includes text prompts such as "Please keep your back straight" or "Don't let your knees go beyond your toes", but also includes voice prompts played through speakers to help students adjust at any time during training. Meanwhile, video demonstrations will also be played synchronously so that students can visually understand the correct execution method of the standard action. This step generates personalized feedback that conforms to the student's current training status based on the action deviation quantification value identified in the previous step. The real-time feedback mechanism can immediately adjust the student's actions, ensuring that they are corrected according to the standard actions and enhancing the teaching effect.

[0071] Furthermore, the feedback information is presented in real time through the monitors, speakers, and smart wearable devices in the training venue. Specifically, the monitor shows a comparison graph of the student's current action and the standard action, helping the student clearly see the gap between the two; the speaker plays voice prompts to further emphasize the parts that the student should improve; the smart wearable devices (such as watches and wristbands) remind the student of the action deviation through vibration or flashing lights, enhancing the student's sense of participation and reaction speed. This step makes the presentation of the feedback information more comprehensive and three-dimensional through the comprehensive use of multiple devices. The collaboration between the monitor, speaker, and wearable devices achieves multi-channel feedback, stimulating the student in all aspects of vision, hearing, and touch, thereby improving the effect and timeliness of the feedback. This process is closely connected to the previous steps. Through the real-time presentation of the feedback, students can adjust their training in a timely manner and avoid the accumulation of mistakes.

[0072] In the embodiment of the present invention, in S4.2, it specifically includes the following steps: S4.2.1: Compare the joint positions in the real-time collected action data with the joint positions of the preset standard action, and calculate the joint position deviation; S4.2.2: Compare the movement trajectories in the real-time collected action data with the movement trajectories of the preset standard action, and calculate the movement trajectory deviation; S4.2.3: Compare the speeds in the real-time collected action data with the speeds of the preset standard action, and calculate the speed deviation; S4.2.4: Comprehensive deviation calculation: Combine the joint position deviation, movement trajectory deviation, and speed deviation to calculate the final action deviation quantification value.

[0073] In a possible implementation, first, the joint position data collected in real time is compared with the joint positions of a preset standard action. This step focuses on the spatial positions of the joints in the student's movement (i.e., the three-dimensional coordinates of each joint). By calculating the differences in the positions of each joint, it is possible to identify whether there are joint misalignments or posture distortions when the student performs the action. For example, whether the elbow of the arm reaches the ideal angle, whether the knees are parallel to the ground, etc. The calculated deviation values can be measured by methods such as Euclidean distance and Manhattan distance. This step provides the basic data for subsequent deviation quantification.

[0074] Next, the movement trajectory collected in real time is compared with the standard action trajectory. This step ensures the coherence and smoothness of the action by capturing the movement paths of various parts of the body (especially the joints) during the entire execution of the action. For example, in a running action, the student's legs should move along a roughly arc-shaped trajectory. If the movement trajectory of the student's legs deviates from the preset trajectory, it may lead to a decrease in the efficiency of the action or cause unnecessary injuries. Therefore, the deviation value of the movement trajectory is crucial for correcting the movement posture.

[0075] Furthermore, the speed deviation is calculated by comparing the speed of the student's action with the speed of the standard action. Speed is an important factor in sports performance, and too fast or too slow a speed will affect the quality of the action. For example, when doing push-ups, if the speed is too fast, it may lead to an improper posture and affect the exercise effect; while too slow a speed may mean that the action is not fully completed. Therefore, by comparing the speed deviation, it is possible to further guide the student to adjust the exercise intensity and optimize the action execution.

[0076] Finally, by integrating the joint position deviation, movement trajectory deviation, and speed deviation, and through weighted average or other optimization algorithms, the final quantified value of the action deviation is calculated. This quantified value not only reflects the accuracy of the student's posture but also takes into account the overall coordination and efficiency of the action execution. The comprehensive deviation quantification value can provide a multi-dimensional and all-round action assessment, providing more accurate data support for the real-time feedback mechanism. In the embodiment of the present invention, the weights of the joint position deviation, movement trajectory deviation, and speed deviation are 0.4, 0.3, and 0.3 respectively, and the comprehensive deviation value is calculated by the following formula: 。

[0077] In a possible implementation, the joint position deviation reflects the difference between the joint positions of the student during the execution of the action and the standard action. Since the joint position is the basis of the action quality, any misalignment of the joint position may directly affect the correctness and efficiency of the action. For example, when doing squats, whether the knees are aligned with the toes and the angle of the hips is appropriate directly determines the safety and effectiveness of the action. Therefore, the weight of the joint position deviation is set to 0.4, emphasizing its importance in the execution of the action.

[0078] Furthermore, the motion trajectory deviation measures the difference between the motion paths of various parts of the student's body and the standard trajectory during the motion process. The trajectory deviation reflects the fluency and coherence of the movement. If the student's trajectory is not standard when performing an action, such as the knees caving in or flaring out during running, this not only affects the action effect but may also cause sports injuries. Since the motion trajectory is directly related to the overall fluency of the action, its weight is 0.3, which reasonably reflects the impact of the trajectory on the action.

[0079] Furthermore, the speed deviation calculates the difference between the speed of the student during the action and the standard speed. Excessive speed may lead to loss of control of the action, while too slow speed may affect the intensity or effect of the exercise. Controlling the speed deviation helps ensure that the student can complete the action within the specified time, thereby improving the exercise effect. Its weight is also 0.3, indicating its balance with the other two deviation dimensions in the comprehensive score.

[0080] Through the weighted calculation of the above three deviation dimensions, a comprehensive deviation value is finally obtained, and the formula is as follows:

[0081] This comprehensive deviation value quantifies the overall performance of the student's action and can comprehensively reflect the student's performance in each key dimension. Through this comprehensive value, teachers can quickly understand the main problems existing in the student's action and thus provide targeted feedback.

[0082] In the embodiment of the present invention, in S5, it specifically includes the following steps: S5.1: Evaluate the student's basic motor ability according to the motion data collected in S1; S5.2: Analyze the student's interest preferences through the student's historical records. The historical records are the game types, training time, frequency, participation, training data, and feedback situations selected by the student during the historical training process. The interest preference analysis is implemented through a data mining algorithm. The input of the data mining algorithm is the student's historical records, and the output is the preference scores of the student for different sports items; S5.3: Generate personalized interactive sports games according to the evaluation results of the basic motor ability and interest preferences. The game generation algorithm is implemented through a multi-objective optimization model. The input is the evaluation result of the student's basic motor ability and the interest preference scores, and the output is personalized game content, including game types, difficulty settings, game scenes, and interaction methods.

[0083] In a possible implementation, the system evaluates the basic motor ability of students based on the student movement data collected in S1 (such as movement duration, movement intensity, movement type, etc.). This evaluation can cover basic motor indicators such as students' strength, flexibility, endurance, and coordination. Through big data analysis and machine learning algorithms, the system can mine the student's movement ability model from the movement data without supervision, and then judge the potential of students in different movement types. For example, the running data of a student may show strong endurance and high cardiorespiratory function, while another student may perform better in strength training.

[0084] Furthermore, by analyzing the behavioral data of students in historical training, such as the game types selected, training time, training frequency, participation, and training data and feedback, to determine the students' interest preferences. This analysis usually uses data mining algorithms, such as clustering analysis, association rule learning, etc., to identify the preference patterns of students. For example, some students may prefer group competition type games, while others tend to individual competition or projects with higher challenges. Based on the analysis of historical records, the system can provide an "interest preference score" for each student, that is, evaluate the tendency of students towards different sports projects, so as to provide data support for subsequent personalized sports design.

[0085] Specifically, in step S5.2, the data mining algorithm is used to analyze the historical records of students to evaluate their interest preferences. Specifically, the core task of the data mining algorithm is to discover potential interest patterns from the behavioral data of students in historical training.

[0086] First of all, the system needs to collect the historical record data of students, which includes the game types selected by students in historical training, training time, frequency, participation, and feedback in training (such as game scores, completion status, etc.). These data will be used as input for subsequent analysis and mining. The data preprocessing steps include denoising, data filling, standardization, etc., to ensure the accuracy and consistency of the data.

[0087] Selection of data mining algorithm: To effectively analyze and identify the interest preferences of students. In this embodiment, clustering analysis is used for analysis and identification. Specifically, through unsupervised learning, students are divided into different groups (such as groups of students with similar interests). This algorithm groups similar students into one class based on characteristics such as the movement types selected by students and participation frequency, and then infers the preferences of different groups.

[0088] Through the above data mining algorithm, the system can generate an interest preference score for each student, reflecting their preference intensity for different types of sports. For example, if a student often selects football competitions with a high participation rate, then the preference score for football games will be higher. The score can be calculated using a weighted average method, comprehensively evaluating based on the frequency of student selection, participation rate, and training feedback.

[0089] Based on the mined interest preference scores, the system can further customize personalized sports games for each student. In step S5.3, these scores will be used as inputs to help generate personalized game content that not only matches the students' interests but also enhances their sports abilities.

[0090] Furthermore, after completing the assessment of basic sports abilities and interest preferences, the system will input this data into the game generation algorithm to achieve personalized sports game design through a multi-objective optimization model. The multi-objective optimization model can comprehensively consider the students' basic abilities and interest preferences and output the most suitable game content. Specifically, the design of the game content includes the following aspects: Game type: Select a suitable game type according to the students' interest preferences (such as competition, challenge, cooperation, etc.).

[0091] Difficulty setting: Adjust the difficulty of the game according to the students' sports abilities to ensure that the game is challenging but not overly difficult so as not to dampen the students' enthusiasm.

[0092] Game scene: Design a suitable virtual scene according to the students' preferences and the characteristics of the sports event, such as outdoors, indoors, race tracks, etc.

[0093] Interaction method: Optimize the interaction mode of the game according to the students' interaction preferences (such as single-player interaction, teamwork, real-time feedback, etc.).

[0094] Specifically, in step S5.3, the multi-objective optimization model is used to generate personalized interactive sports games based on the assessment results of the students' basic sports abilities and interest preference scores. The goal of this model is to balance multiple factors to achieve personalized customization of the game and enhance the students' learning interest and sports abilities.

[0095] First of all, the model needs to select the most suitable game type for the students according to their interest preference scores (the output of S5.2). For example, for students who like competitive games, the system will tend to generate game types involving competition and confrontation; while for students who prefer cooperation or exploration, group cooperation or adventure games may be recommended.

[0096] Based on the assessment of students' basic motor abilities (the output of S5.1), the optimization model will adjust the difficulty level of the game to ensure that the game is both challenging and not overly difficult or easy. For example, for students with stronger basic motor abilities, the model may set a higher game difficulty, providing more complex tasks or higher exercise intensity; while for students with weaker foundations, the game difficulty will be reduced so that they can gradually improve their abilities. The design of the scenario needs to consider the students' interests and abilities comprehensively, and the model will select a suitable scenario according to the students' preferences. For example, for students who like outdoor sports, the model may choose natural scenarios or open spaces; while for students who like indoor activities, simulated gymnasiums or virtual environment scenarios may be recommended.

[0097] Furthermore, the model also needs to adjust the interaction methods in the game according to the students' preferences. This includes the frequency of interaction, the form (e.g., competition, cooperation, or individual challenges), and the interaction with other students or AI characters. By optimizing these elements, it is ensured that students maintain a high level of engagement and interactivity in the game.

[0098] To achieve these goals, the multi-objective optimization model uses the weighted summation method technology. By defining the weights and constraints of each objective, the best balance point is found among multiple objectives. The inputs of the model include factors such as students' basic motor ability scores, interest preference scores, game types, difficulty levels, scenarios, and interaction methods, and the output is the final personalized sports game content.

[0099] Through this multi-objective optimization method, the system can not only fully consider the differences in students' interests and abilities, but also generate personalized interactive sports games that can improve students' engagement and motor abilities, thereby promoting students to continuously improve their physical fitness and skills in the game.

[0100] In the embodiments of the present invention, the assessment of basic motor abilities includes muscle strength assessment, endurance assessment, and coordination assessment. Among them, by analyzing the students' movement postures and speeds, the muscle strength of the students is evaluated. The specific formula is: ; Among them, the number of accurate movements refers to the number of times the student has the correct posture when completing the movement, and the speed score refers to the ratio of the speed at which the student completes the movement to the standard value; By analyzing the students' exercise time, heart rate, and fatigue level, the endurance of the students is evaluated. The specific formula is: ; Among them, the exercise time refers to the time that the student continuously exercises, the heart rate stability refers to the fluctuation range of the heart rate of the student during exercise, and the fatigue index is the degree of fatigue evaluated by the large model.

[0101] Evaluate the flexibility of students by analyzing their movement postures and reaction times. The specific formula is: ; Among them, the movement range refers to the range of activities of students when completing movements, and the reaction time score refers to the ratio of the reaction time of students when completing movements to the standard value; Evaluate the coordination of students by analyzing their movement postures, speeds and reaction times. The specific formula is: ; Among them, the number of accurate movements refers to the number of times students have correct postures when completing movements, the speed score refers to the ratio of the speed of students completing movements to the standard value, and the reaction time score refers to the ratio of the reaction time of students when completing movements to the standard value.

[0102] In a possible implementation, in the interactive sports teaching method based on large models, the basic motor ability assessment plays a core role. It helps to accurately understand the physical fitness and motor ability of students through a multi-dimensional scoring system. The core steps of this process involve the assessment of muscle strength, endurance, flexibility and coordination. The assessment of each ability uses specific mathematical formulas and standardized scoring methods. The comprehensive assessment of these abilities provides accurate data support for the generation of personalized sports games, and helps to customize effective teaching content and sports tasks according to the specific situation of students.

[0103] First of all, the assessment of muscle strength is calculated by analyzing the movement postures and speeds of students. In this scoring formula, the number of accurate movements and the speed score jointly determine the scoring standard of muscle strength. The number of accurate movements reflects whether the posture of students is standard when performing movements, while the speed score reflects their movement efficiency by comparing the ratio of the speed of students to the standard value. Through this scoring system, while accurately grasping the muscle strength level of students, it helps the system to adjust the difficulty of the game to ensure that the sports tasks are neither too simple nor too difficult.

[0104] Secondly, the endurance assessment comprehensively considers the movement time, heart rate stability and fatigue index. The movement time and heart rate stability indicate the persistence and endurance level of students in long-term exercises, while the fatigue index is a quantitative assessment of the fatigue degree of students by the large model. The design of the endurance score helps to judge the ability of students to continuously participate in sports, and provides a basis for the system to adjust the duration and intensity of sports tasks. This assessment helps to avoid discomfort caused by excessive fatigue of students, and at the same time optimizes the training intensity to promote the gradual improvement of students' sports endurance.

[0105] In the assessment of flexibility and coordination, the analysis of movement postures, speed, and reaction time are key factors. The flexibility score reflects the degree of freedom of movement and reaction speed of students during exercise through the combination of the range of motion and reaction time scores. The coordination score, on the other hand, is a multi-dimensional consideration that synthesizes factors such as movement accuracy, speed, and reaction time, reflecting the body coordination of students when performing movement tasks. These assessments can not only identify students' performance in different sports fields but also help the system optimize the design of movement tasks to ensure that students can continuously improve their coordination and flexibility in interactive games.

[0106] Through the combination of these ability scores, the entire assessment system can accurately reflect students' sports abilities and provide data support for subsequent personalized sports game design. The beneficial effects of this assessment method based on multi-dimensional scores are multi-faceted: it can not only help students achieve differentiated and precise improvement in various sports abilities but also increase students' interest and sense of participation through personalized interactive sports games, stimulating their sports potential. Therefore, this assessment step is crucial for the all-round development of students and the implementation of personalized physical education.

[0107] Example 2: The system communicates with students in real time through speech recognition and synthesis technologies. For example, during training, students can ask "Are my movements standard?" through speech. The system will, based on the real-time collected data and the analysis results of the large model, answer through speech "Your movement standardization score is 0.7. It is recommended to adjust the position of your knees." This speech interaction method is not only convenient and fast but also enables students to stay focused during training without interrupting the training to check the screen.

[0108] The system can have multiple rounds of conversations with students, providing personalized guidance and suggestions. For example, students can ask "How can I improve my endurance?" The system will, based on the students' sports data and historical records, combined with the analysis of the large model, provide specific training suggestions such as "You can increase the duration of each run, gradually increasing from 30 minutes to 45 minutes." This conversation function can not only answer students' questions but also provide targeted suggestions according to the actual situation of students.

[0109] The system generates personalized training plans based on the students' sports status and goals. For example, the system can suggest "Today your training focus is to improve explosive power. It is recommended to do squat jumps and resistance band training." This training guidance is based on the comprehensive assessment of students by the large model, including movement standardization, exercise intensity, and fatigue level, etc., to ensure the scientific nature and effectiveness of the training plan.

[0110] The system encourages students to actively participate in training through a positive incentive mechanism. For example, when a student completes an action, the system will give encouragement through voice and visual feedback, such as "Well done! Your movements are becoming more and more standard." This incentive mechanism can improve students' enthusiasm and participation, and enhance the training effect.

[0111] Traditional physical education teaching often follows fixed processes and rules, while the AI personal trainer based on large models is more flexible and can dynamically adjust teaching content and methods according to the real-time status and needs of students. For example, if the system detects that a student is fatigued, it will automatically adjust the training intensity instead of strictly following the preset process. This non-process nature makes the teaching more personalized and better able to meet the needs of different students.

[0112] Specifically, the AI personal trainer does not rely on fixed rules and scripts, but provides personalized guidance through the intelligent analysis and decision-making of the large model. For example, the system can generate corrective suggestions in real time based on the student's movement deviations instead of relying on a preset rule library. The absence of such rules makes the system more intelligent and flexible, capable of handling various complex situations.

[0113] Furthermore, the embodiments of the present invention can also conduct sports interactions through gamification design. Specifically, the system uses gamification elements, such as virtual rewards, leaderboards, challenge tasks, etc., to improve students' participation and interest. For example, students can obtain virtual medals by completing specific training tasks or conduct virtual competitions with other students. This gamification design not only increases the fun of training but also stimulates students' competitive awareness and improves the training effect.

[0114] The system can also provide an immersive sports experience through virtual reality (VR) and augmented reality (AR) technologies. For example, students can conduct basketball shooting training in a virtual environment, and the system will provide real-time feedback on the accuracy of the movements and provide virtual opponents for confrontation. This way of sports interaction can not only improve students' participation but also enhance the realism and fun of training.

[0115] The system generates specific modification tips through the intelligent analysis of the large model. For example, if a student's shooting movement is not standard, the system will prompt "Your elbow is abducted too much. It is recommended to keep the upper arm at a 45° angle with the torso." Such modification tips are based on the large model's real-time analysis of the student's movements and can provide more accurate and specific guidance.

[0116] The system provides video references of standard movements to help students better understand and correct their movements. For example, the system can display the standard shooting movements of NBA players and conduct comparative analysis of key joints. This reference function can not only help students intuitively understand the standard movements but also enhance students' learning effect.

[0117] Through multimodal data fusion, large models can accurately evaluate students' exercise states. For example, by combining the motion data collected by cameras and depth sensors with heart rate and electromyogram signals, large models can more accurately assess students' fatigue levels.

[0118] Large models can conduct detailed analysis on students' movements and identify minor movement deviations. For example, by monitoring the wrist flexion and extension angles during shooting with millimeter-wave radar, large models can accurately identify the subtle movements of students at the moment of shooting.

[0119] Large models can provide professional analysis and suggestions to help students improve their sports skills. For example, by analyzing students' exercise data, large models can generate personalized training plans, including specific movement correction exercises and intensity adjustments.

[0120] Large models can analyze students' historical exercise data to identify long-term exercise trends and potential problems. For example, by analyzing students' training data in the past few weeks, large models can find that a student's endurance has declined and suggest increasing aerobic training.

[0121] At the same time, large models have unique advantages in human-computer interaction and can achieve natural conversations with students through natural language processing and speech recognition technologies. For example, students can ask "How is my endurance?" via voice, and the system will answer based on historical data: "Your endurance has declined in the past four weeks. It is recommended to increase aerobic training." Through the intelligent analysis and interaction functions of large models in the embodiments of the present invention, an upgrade of the teaching paradigm from single movement correction to sports ability development is realized. Especially in the process of acquiring complex sports skills, large models demonstrate significant technical advantages. For example, in basketball shooting training, large models can, through multimodal data fusion, provide accurate real-time movement feedback and personalized training suggestions, significantly improving students' training effects and experiences.

[0122] Based on the technical framework of Embodiment 1, this embodiment upgrades the traditional machine learning model to a deep integration system of a multimodal large model (MLLM) and a language large model (LLM), which is specifically applied to the basketball special training scenario. The technical solutions are improved as follows: Upgrade of data collection; Deploy an 8K panoramic camera array on the gymnasium ceiling to capture the movement trajectories of 20 key joint points of the whole body of trainees at 120 frames per second. Install a millimeter-wave radar array on the basketball hoop to monitor the wrist flexion and extension angles during shooting in real time (accuracy ±0.5°). Smart wristbands synchronously collect electromyogram signals and skin conductivity at the moment of shooting. Six-axis inertial sensors are built into the training shoes to capture changes in foot pressure distribution during takeoff.

[0123] Multimodal feature fusion; Build a three - layer fusion architecture: 1. Visual understanding layer: Adopt a spatio - temporal attention model to analyze continuous actions and automatically identify three key stages of the shooting action: Loading period (monitoring is triggered when the knee - bending angle < 110°) Take - off period (timestamp is marked at the moment when the heel leaves the ground) Release period (detection of the peak wrist joint speed) 2. Motion analysis layer: Detect abnormal patterns through time - series modeling: When a trainee makes a "lumbar - arching shooting" mistake, the system automatically marks the abnormal lumbar - forward - tilt angle (an alarm is triggered when the difference from the standard posture > 8°), and analyzes the problem of insufficient activation of the core muscles in combination with electromyography signals.

[0124] 3. Physiological monitoring layer: Evaluate the training load in real - time: When the heart - rate recovery rate of a trainee drops by 15% after 10 consecutive shots, the intelligent wristband vibrates for warning, and the terminal automatically pulls up the recovery training menu.

[0125] Real - time feedback; Deploy holographic projection devices beside the basketball court to achieve three - dimensional action guidance: 1. Error correction: When it is detected that the trainee's elbow abducts too much during shooting, a red laser auxiliary line is immediately projected on the ground, with a voice prompt: "Please keep the upper arm at a 45° angle with the torso." At the same time, the slow - motion of the standard actions of NBA players is shown in holographic projection, and semi - transparent bone models are superimposed at key joints for comparison.

[0126] 2. Dynamic difficulty adjustment: The intelligent basketball hoop automatically adjusts according to the trainee's level: Novice mode: The diameter of the basketball hoop is expanded to 60 cm, and an LED light strip is built into the net to prompt the best shooting arc Advanced mode: The basketball hoop randomly swings left and right by ± 30 cm to cultivate the ability to adapt to a dynamic environment Challenge mode: AR glasses generate virtual defensive players and implement intelligent interception through a posture prediction algorithm Generate personalized training; Build an intelligent trainer with a language large - model: 1. Training plan generation: Input trainee data: "17 years old, maximum vertical jump height 55 cm, shooting percentage 42%, shoulder fatigue warning last week", and the system automatically generates a 6 - week improvement plan including the following elements: Daily special training: Explosive power training on Monday (squat jumps + resistance - band shooting) Intensity control: Adopt the heart - rate interval control method (the proportion of the Z2 interval ≥ 60%) Action improvement: Add "medicine ball overhead toss" to correct the force chain 2. Gamification design: Develop the "Interstellar Basketball Challenge" AR game: Adjust the training intensity by setting the gravity of the virtual planet (the 1.2G gravity mode enhances lower limb strength) Dynamic task generation: "Complete 5 precise shots before the comet fragments hit" Instant reward: Hitting a key shot triggers a stadium light show and dynamic floor vibration feedback Teaching management platform; The coach's tablet presents a 3D training dashboard: Biomechanics view: Superimpose and display the heat map of the joint angle differences between the trainee and Curry's shooting actions Load monitoring wall: Real-time scroll of the fatigue index and injury risk warnings of all trainees (high-risk trainees are marked in red) Intelligent lesson preparation system: Input "Focus on improving the ability of change-of-direction breakthrough next week", and automatically push 20 sets of solution packages including video lesson plans and training prop lists Technical effects: In the actual measurement of the basketball experimental class: Improvement in action standardization: The elbow positioning error of trainees during shooting decreased from 12.7° to 3.2° (verified by millimeter-wave radar) Decrease in training injury rate: The system predicts the fatigue risk 12.3 minutes in advance, and the incidence of acute injuries is reduced by 68% Improvement in teaching efficiency: The number of trainees coached by the coach per day increased from 15 to 40, and at the same time, the individualized guidance duration increased by 3 times Through the scene-based deep integration of the large model, this embodiment realizes the upgrade of the teaching paradigm from single-action correction to the development of sports abilities, and particularly shows significant technical advantages in the process of acquiring complex motor skills.

[0127] This invention covers any substitutions, modifications, equivalent methods, and solutions made within the essence and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention without these detailed descriptions. Additionally, to avoid unnecessary confusion to the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0128] The above are only the preferred embodiments of this invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of this invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. An interactive sports teaching method based on a large model, characterized in that It includes the following steps: S1: Real-time collect the motion data of students through a camera and a depth sensor, including the students' movement postures, speeds, reaction times, and heart rates; The cameras are installed at multiple positions in the training venue, and the depth sensors are installed on key teaching equipment. The cameras and depth sensors are connected to the central processor through data acquisition cables to transmit the collected data in real time; The central processor preprocesses the collected motion data, including filtering, denoising, and normalization processing; S2: The central processor inputs the preprocessed data into a pre-trained large model to evaluate the students' motion states, including action standardization, exercise intensity, and fatigue level; S3: According to the evaluation results of the students' motion states, the large model automatically generates personalized training programs; S4: Real-time capture the students' motion postures, analyze whether the actions are standardized, and provide real-time feedback; S5: Generate interactive sports games based on the students' basic motor abilities and interest preferences.

2. The interactive sports teaching method based on a large model according to claim 1, wherein The large model described in S2 includes a convolutional neural network model, a support vector machine model, and a random forest model. The convolutional neural network model, the support vector machine model, and the random forest model are respectively used to evaluate action standardization, exercise intensity, and fatigue level.

3. The interactive sports teaching method based on a large model according to claim 2, wherein, The input of the convolutional neural network model is the motion data collected in real time, including the action posture image data, and the output is the action standard score ; The input of the support vector machine model is the motion intensity data collected in real time, including speed, acceleration, and heart rate, and the output is the motion intensity evaluation ; The input of the Random Forest Model (RF) is the fatigue degree data collected in real time, including heart rate and action posture stability, and the output is the fatigue degree evaluation .

4. An interactive sports teaching method based on a large model according to claim 3, characterized in that, The specific steps for generating personalized training programs in S3 are as follows: S3.1: Action standardization assessment. If the action standardization score is lower than the threshold , generate corrective action practice; S3.2: Exercise intensity assessment. If the exercise intensity is lower than the threshold , generate training with increased intensity; S3.3: Fatigue level assessment, if the fatigue level is higher than the threshold , generate restorative training.

5. An interactive sports teaching method based on a large model according to claim 4, characterized in that, The corrective action exercises, intensity-increasing training, and restorative training all include multiple specific actions. Each action includes an action name, an action description, a standard action demonstration video, and a scoring criterion; Among them, the corrective action exercises, intensity-increasing training, and restorative training retrieve their corresponding action names and action descriptions from a pre-stored database, and retrieve the standard action demonstration videos and scoring criteria from a pre-stored video library to generate corrective action exercises, intensity-increasing training, or restorative training.

6. The interactive sports teaching method based on a large model according to claim 5, wherein, In S4, it specifically includes the following steps: S4.1: Real-time collect the action data of students through the cameras and depth sensors in S1, including joint positions, motion trajectories, and speeds. The depth sensor captures the details of the students' actions, and the cameras record the students' motion postures from multiple angles; S4.2: Compare the real-time collected action data with the preset standard action demonstrations obtained in S3, and identify the action deviations of the students through a pose estimation algorithm. The input of the pose estimation algorithm is the real-time collected action data, and the output is the quantified value of the action deviation; S4.3: Generate real-time feedback information according to the quantified value of the action deviation. The feedback information includes text prompts, voice prompts, and video demonstrations, which are played in real time through multimedia devices to guide the students to correct their actions. The input is the quantified value of the action deviation, and the output is the real-time feedback information; S4.4: The feedback information is presented in real time through the monitors, speakers, and smart wearable devices installed in the training venue. The screen displays the comparison diagram of the students' current action postures and the standard postures. The speakers play voice prompts, and the smart wearable devices vibrate or flash lights to attract the students' attention.

7. An interactive sports teaching method based on a large model according to claim 6, characterized in that, In S4.2, it specifically includes the following steps: S4.2.1: Compare the joint positions in the real-time collected action data with the joint positions of the preset standard actions, and calculate the joint position deviations; S4.2.2: Compare the motion trajectory in the real-time collected motion data with the motion trajectory of the preset standard motion, and calculate the motion trajectory deviation; S4.2.3: Compare the speed in the real-time collected motion data with the speed of the preset standard motion, and calculate the speed deviation; S4.2.4: Comprehensive deviation calculation: Synthesize the joint position deviation, motion trajectory deviation and speed deviation to calculate the final quantified value of the motion deviation.

8. An interactive sports teaching method based on a large model according to claim 7, characterized in that, The weights of the joint position deviation, motion trajectory deviation and speed deviation are 0.4, 0.3 and 0.3 respectively, and the comprehensive deviation value is calculated by the following formula: 。 9. An interactive sports teaching method based on a large model according to claim 1, characterized in that, In S5, it specifically includes the following steps: S5.1: Evaluate the student's basic motor ability according to the motion data collected in S1; S5.2: Analyze the student's interest preferences through the student's historical records. The historical records are the game types, training time, frequency, participation, training data and feedback selected by the student during the historical training process. The interest preference analysis is realized through a data mining algorithm. The input of the data mining algorithm is the student's historical records, and the output is the preference score of the student for different sports items; S5.3: Generate personalized interactive sports games according to the evaluation results of the basic motor ability and interest preferences. The game generation algorithm is realized through a multi-objective optimization model. The input is the evaluation result of the student's basic motor ability and the preference score of the interest preference, and the output is personalized game content, including game types, difficulty settings, game scenes and interaction methods.

10. A large model-based interactive sports teaching method according to claim 9, characterized in that The evaluation of the basic motor ability includes muscle strength evaluation, endurance evaluation, coordination evaluation and flexibility evaluation. Among them, the muscle strength of the student is evaluated by analyzing the student's movement posture and speed. The specific formula is: ； Among them, the number of accurate movements refers to the number of times the student has a correct posture when completing the movement, and the speed score refers to the ratio of the speed at which the student completes the movement to the standard value; The endurance of the student is evaluated by analyzing the student's exercise time, heart rate and fatigue level. The specific formula is: ； Among them, the exercise time refers to the time that the student continuously exercises, the heart rate stability refers to the fluctuation range of the student's heart rate during the exercise process, and the fatigue index is the fatigue level evaluated by a large model; The flexibility of the student is evaluated by analyzing the student's movement posture and reaction time. The specific formula is: ； Among them, the movement range refers to the range of activities of the student when completing the movement, and the reaction time score refers to the ratio of the reaction time of the student when completing the movement to the standard value; The coordination of the student is evaluated by analyzing the student's movement posture, speed and reaction time. The specific formula is: ； Among them, the number of accurate movements refers to the number of times the student has a correct posture when completing the movement, the speed score refers to the ratio of the speed at which the student completes the movement to the standard value, and the reaction time score refers to the ratio of the reaction time of the student when completing the movement to the standard value.

Images