Intelligent exercise training teaching device for stadium

Abstract

The invention relates to the technical field of sports teaching, in particular to an intelligent sports training teaching device for a stadium. The device comprises an action monitoring module, an error frame point identification module, a posture matching evaluation module and a feedback optimization adjustment module, the monitoring terminal can be connected with a corresponding camera, an intelligent display screen and a training equipment sensor in a training area of the stadium, and is used for locking and monitoring a training action video and training action sensing data corresponding to a trainee in real time; a high-speed wireless communication module is adopted to transmit to a corresponding central data processing unit in real time, wrong action time frame point recognition and wrong time synchronization determination are carried out, meanwhile, action posture matching evaluation is carried out, and the training action posture matching degree corresponding to a trainee is obtained; and performing teaching feedback optimization adjustment on the personalized teaching training plan corresponding to the trainer to generate personalized teaching feedback suggestions corresponding to the trainer on the stadium. The physical training effect can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of sports teaching, and in particular to an intelligent sports training teaching device for a stadium. Background Art

[0002] With the continuous development of science and technology, intelligent technology has gradually penetrated into all walks of life, especially in the field of sports. The application of intelligent sports training equipment has become an important means to improve the efficiency of sports training and optimize teaching effects. At present, there are some sports training equipment on the market that use sensors, cameras, motion capture and other technologies to collect and analyze the movement status of students. Some intelligent systems can provide corrections and suggestions for sports movements. However, the training process in traditional stadiums is usually based on static action demonstrations and repeated exercises, which makes it difficult to fully mobilize the training enthusiasm of students. In addition, there is a lack of comprehensive sports scene coverage and comprehensive guidance for multi-project training during the training process. At the same time, existing intelligent devices are often unable to accurately capture the movement status and performance of students when dealing with complex training scenes, and thus cannot obtain accurate training guidance. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide a stadium intelligent sports training teaching device to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a stadium intelligent sports training teaching device includes the following modules: The motion monitoring module is used to connect with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area, and use the smart display screen to display the personalized teaching training plan while using the cameras and training equipment sensors to lock and monitor the corresponding training motion videos and training motion sensor data of the trainees in real time; and use the high-speed wireless communication module to transmit the corresponding training motion videos and training motion sensor data of the trainees in real time to the corresponding central data processing unit; The error frame point recognition module is used to use the central data processing unit to perform error action time frame point recognition on the training action video corresponding to the trainee, so as to obtain the error action time frame point corresponding to the trainee; The posture fit evaluation module is used to perform error time synchronization determination on the training action sensor data corresponding to the trainee based on the time frame point of the trainee's corresponding motion error action, so as to obtain the training action sensor data corresponding to the trainee at the error time point; obtain the standard action of the sports training project, and perform action posture fit evaluation on the training action sensor data corresponding to the trainee at the error time point based on the standard action of the sports training project, so as to obtain the training action posture fit degree corresponding to the trainee; The feedback optimization and adjustment module is used to optimize and adjust the teaching feedback of the trainee's corresponding personalized teaching training plan based on the trainee's corresponding training movement posture fit, so as to generate personalized teaching feedback suggestions for the trainee in the stadium.

[0005] Furthermore, the action monitoring module includes the following functions: By connecting with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area, a comprehensive training visual monitoring network is built in the stadium training area; When the all-round training visual monitoring network detects that the trainee enters the sports training area and touches the smart display screen, the smart display screen responds to the trainee's corresponding training display touch operation instructions, and displays and formulates a personalized teaching training plan corresponding to the trainee on the smart display screen according to the training display touch operation instructions; The trainee performs sports training according to the corresponding personalized teaching training plan, and uses the camera to lock and monitor the trainee's corresponding position and training movements in real time during the sports training to generate the trainee's corresponding training movement video; The training equipment sensor is used to monitor the pressure distribution of the equipment corresponding to the trainee, the acceleration of the training action and the change of the training action posture in real time, so as to obtain the training action sensor data corresponding to the trainee; By adopting a high-speed wireless communication module, the trainee's corresponding training action video and training action sensor data are transmitted to the corresponding central data processing unit in real time.

[0006] Furthermore, the training equipment sensor includes a pressure sensor, an acceleration sensor and a gyroscope.

[0007] Furthermore, the error frame point identification module includes the following functions: Using the central data processing unit to adjust the video frame rate of the training action video corresponding to the trainee, so as to generate the action frame-adjusted video corresponding to the trainee; Performing video denoising on the action frame-adjusted video corresponding to the trainee to generate an action denoised video corresponding to the trainee; By setting video frame points corresponding to 5-second intervals, and dividing the action denoising video corresponding to the trainee based on the corresponding video frame points, so as to generate training action frame segments corresponding to the trainee at each video frame point; The wrong action time frame points are identified for the training action frame segments corresponding to the trainee at each video frame point, so as to obtain the corresponding wrong action time frame points of the trainee.

[0008] Furthermore, the step of identifying the time frame points of the wrong actions of the trainee at each video frame point corresponding to the training action frame segments includes: Obtain sports training action data after marking standard training actions and common incorrect actions; Based on the sports training action data after the training standard action and common incorrect action are annotated as training samples and combined with the convolutional neural network and the recurrent neural network to train the action recognition model, a sports training action recognition model is generated to accurately identify the standard actions and common incorrect actions corresponding to various sports items, and output the action features corresponding to the sports training actions, including the action rhythm and action amplitude corresponding to the exercise parts of the trainee; Input the training action frame segments corresponding to the trainee at each video frame point into the previously trained sports training action recognition model for action feature recognition analysis to obtain the training action video frame features corresponding to the trainee at each video frame point; Based on the training action video frame features corresponding to the trainee at each video frame point, the training action frame segments corresponding to the trainee at each video frame point are identified with the wrong action time frame points to obtain the corresponding motion wrong action time frame points of the trainee.

[0009] Further, the identifying of the wrong action time frame points of the training action frame segments corresponding to the trainee at each video frame point based on the training action video frame features corresponding to the trainee at each video frame point includes: The preliminary action detail differences between the video frame features of the trainee's training action at each video frame point and the standard action in the pre-built action template library are quantified to obtain the action detail deviation between the trainee and the standard action at each video frame point; Based on the deviation of the trainee's action details between the standard action at each video frame point, the video frame points corresponding to the trainee are screened for suspected frames. If the corresponding action detail deviation is greater than or equal to the preset action deviation threshold, it is considered that the trainee is suspected of having a corresponding wrong action at the video frame point, otherwise it is not considered, and the video frame points corresponding to the suspected wrong action are screened out to obtain a time frame point set of suspected wrong actions corresponding to the trainee; The action detail deviation between the adjacent time frames and the standard action is obtained through the time frame point set of the suspected wrong action corresponding to the trainee, and the action deviation trend analysis is performed according to the action detail deviation between the adjacent time frames and the standard action, so as to obtain the change trend of the action deviation between the trainee and the standard action in the suspected adjacent time frames; Based on the variation trend of the trainee's movement deviation from the standard movement in the suspected adjacent time frames, the corresponding adjacent time frames in the suspected erroneous movement time frame point set are subjected to the position movement difference correlation analysis, so as to obtain the corresponding mutual relationship of the trainee's position movement difference between the suspected adjacent time frames; Based on the mutual relationship between the differences in the trainees' part movements between suspected adjacent time frames, the incorrect movement time frame points are identified for the set of suspected incorrect movement time frame points corresponding to the trainees. If there is a correlation between the deviations of the part movements between adjacent time frames and the deviations show a trend of gradually increasing, it is judged that there is an incorrect movement between the suspected adjacent time frames, and the corresponding average time points between the suspected adjacent time frames are calculated as the corresponding incorrect movement frame points, so as to obtain the corresponding movement incorrect movement time frame points of the trainees.

[0010] Furthermore, the posture fit assessment module includes the following functions: Based on the time frame point of the trainee's corresponding motion error action, the trainee's corresponding training motion sensor data is synchronously determined to obtain the trainee's corresponding training motion sensor data at the error time point, including pressure distribution, motion acceleration and motion posture; Obtaining corresponding sports training items through the corresponding training action sensor data of the trainee at the wrong time point; According to the sports training items corresponding to the trainee, the corresponding standard movements of the sports training items are obtained from the corresponding movement template library; Based on the standard movements of the sports training items, the training movement sensor data corresponding to the trainee at the wrong time point is evaluated for movement posture fit, and the training movement posture fit of the trainee is obtained.

[0011] Furthermore, the performing of movement posture matching evaluation on the training movement sensor data corresponding to the trainee at the wrong time point based on the standard movement of the sports training item includes: By mapping the standard movements of the sports training items and the movement postures in the training movement sensor data corresponding to the trainee at the wrong time point to the same spatial coordinate system, the corresponding standard training movements and the trainee's wrong movement postures in the same spatial coordinate system are generated; Determine the posture position and direction of the standard training action and the trainer's incorrect posture in the same spatial coordinate system, so as to obtain the posture position and posture direction corresponding to the standard training action and the trainer's incorrect posture in the same spatial coordinate system; Calculating the position distance between the posture positions corresponding to the standard training action and the trainer's erroneous action in the same spatial coordinate system to obtain the posture position distance between the standard training action and the trainer's erroneous action in the same spatial coordinate system; According to the direction deviation measurement between the posture directions corresponding to the standard training action and the trainer's erroneous action in the same spatial coordinate system, the posture direction deviation between the standard training action and the trainer's erroneous action in the same spatial coordinate system is obtained; The posture fit calculation formula is used to evaluate the posture fit of the posture position distance and posture direction deviation between the standard training movement and the trainee's incorrect movement in the same spatial coordinate system, and the corresponding training posture fit of the trainee is obtained.

[0012] Furthermore, the action posture matching calculation formula is specifically: ; In the formula, To train the coordination of movements and postures, is the total number of nodes corresponding to the action posture position, To train standard movements The node coordinates corresponding to the action posture position, For trainers wrong action The node coordinates corresponding to the action posture position, It is the distance between the standard training action and the trainer's wrong action. To train the posture direction deviation between the standard action and the trainee's wrong action, For the The error time frame point corresponding to the action posture position, is the time decay factor.

[0013] Furthermore, the feedback optimization and adjustment module includes the following functions: The training action posture conformity of the trainee is compared and judged according to the preset training action conformity threshold. If the training action posture conformity of the trainee is greater than or equal to the preset training action conformity threshold, it is considered that the actual training effect of the trainee has reached the standard; if the training action posture conformity of the trainee is less than the preset training action conformity threshold, it is considered that the actual training effect of the trainee has not reached the standard, and there are problems of poor teaching training effect or irregular action. According to the trainee's corresponding poor teaching and training results or irregular movements, the personalized teaching and training plan corresponding to the trainee is optimized and adjusted through teaching feedback to automatically generate corresponding personalized feedback suggestions, including adjusting training movements and changing training intensity, and timely feedback is pushed to the trainee through the smart display screen to generate personalized teaching feedback suggestions for the trainee on the stadium.

[0014] Beneficial effects of the present invention: The stadium intelligent sports training teaching device proposed by the present invention is generally composed of a motion monitoring module, an error frame point recognition module, a posture fit evaluation module and a feedback optimization adjustment module. Compared with the prior art, the beneficial effect of the present application lies in that the equipment and technology in the stadium training area are closely integrated to achieve real-time monitoring and accurate analysis of the training process. Through the connection with the camera, the smart display screen and the training equipment sensor, the system can capture the athletes' action videos and sensor data in real time during training, and accurately record each training detail. This method not only improves the visualization of training, but also can directly display personalized teaching and training plans through the smart display screen when the trainee performs a certain action, so that the trainee's movement state and performance can be accurately captured, and more targeted guidance can be provided. Through the high-speed wireless communication module, all training videos and sensor data will be transmitted to the central data processing unit in a timely and stable manner, which enables subsequent analysis and feedback to be quickly started, avoiding the lag of traditional manual recording and analysis. This real-time data stream and connection method effectively improves the intelligence level of training and ensures that each action in the training process can be accurately monitored. Secondly, the core task of the central data processing unit in this step is to use advanced image recognition and motion analysis technology to conduct real-time analysis of the trainee's training video. By carefully comparing each frame of the video image, the system can identify the incorrect movements in the training process. The accurate identification of the time frame points of the incorrect movement is the key technical link of the system. It uses algorithm calculation to compare the training movement standard with the actual movement, accurately calibrate when the trainee made a mistake, and how the incorrect movement occurred. This process is not just a simple image processing, but a deep combination of deep learning and motion recognition technology to accurately locate the problems such as non-standard movements and irregular postures in training, and mark the specific time points of the incorrect movements. This precise identification and positioning provides a strong basis for subsequent movement correction and feedback, and can achieve real-time tracking of errors from the beginning of training, minimize the accumulation of improper movements in training, and thereby improve training effects and athlete performance. Then, the trainee's training data is synchronously calibrated according to the previously identified erroneous movement time frame points to ensure that the motion sensor data corresponds one-to-one with the erroneous time frame points. The data collected by the sensor, including pressure distribution, movement acceleration, movement posture and other indicators, can accurately describe the trainee's movement posture and movement details. By matching the erroneous time points with these sensor data, the system can provide specific and quantitative motion data analysis. Then, the system will compare the trainee's movement data with the standard movement data to evaluate the trainee's movement posture fit, thereby accurately identifying defects or deficiencies in posture, making the trainee's movement improvement more efficient.Finally, the training program was further optimized by feeding back the fit of movement posture into the personalized teaching plan. The key to this feedback mechanism is to adjust the training plan according to the specific performance of each trainee, rather than relying solely on traditional fixed teaching methods. The fit of movement posture of each trainee is different, which is affected by various factors such as body shape, strength, flexibility, etc. By analyzing the movement data and fit evaluation results in real time, the corresponding training strategy is automatically tailored for each trainee to help them correct mistakes and improve athletic performance in practice, ensuring the scientific nature and pertinence of the training plan. This real-time feedback and optimization adjustment mechanism will not only help improve the quality of training in the short term, but also help provide comprehensive guidance for comprehensive project training during the training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings: Figure 1 This is a module schematic diagram of the stadium intelligent sports training and teaching device of the present invention; Figure 2 for Figure 1 Functional flow diagram of the action monitoring module; Figure 3 for Figure 1 Schematic diagram of the functional flow of the error frame point identification module. DETAILED DESCRIPTION

[0016] The technical system of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0017] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor systems and / or microcontroller systems.

[0018] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0019] To achieve this, please refer to Figures 1 to 3 The present invention provides a stadium intelligent sports training and teaching device, the system includes the following modules: The motion monitoring module is used to connect with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area, and use the smart display screen to display the personalized teaching training plan while using the cameras and training equipment sensors to lock and monitor the corresponding training motion videos and training motion sensor data of the trainees in real time; and use the high-speed wireless communication module to transmit the corresponding training motion videos and training motion sensor data of the trainees in real time to the corresponding central data processing unit; The error frame point recognition module is used to use the central data processing unit to perform error action time frame point recognition on the training action video corresponding to the trainee, so as to obtain the error action time frame point corresponding to the trainee; The posture fit evaluation module is used to perform error time synchronization determination on the training action sensor data corresponding to the trainee based on the time frame point of the trainee's corresponding motion error action, so as to obtain the training action sensor data corresponding to the trainee at the error time point; obtain the standard action of the sports training project, and perform action posture fit evaluation on the training action sensor data corresponding to the trainee at the error time point based on the standard action of the sports training project, so as to obtain the training action posture fit degree corresponding to the trainee; The feedback optimization and adjustment module is used to optimize and adjust the teaching feedback of the trainee's corresponding personalized teaching training plan based on the trainee's corresponding training movement posture fit, so as to generate personalized teaching feedback suggestions for the trainee in the stadium.

[0020] In the embodiment of the present invention, please refer to Figure 1 FIG. 1 is a schematic diagram of modules of the stadium intelligent sports training and teaching device of the present invention. In this example, the stadium intelligent sports training and teaching device includes the following modules: S1: a motion monitoring module, which is used to connect with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area, and use the smart display screen to display the personalized teaching training plan, while using the camera and training equipment sensors to lock and monitor the corresponding training action video and training action sensor data of the trainee in real time; and use a high-speed wireless communication module to transmit the corresponding training action video and training action sensor data of the trainee in real time to the corresponding central data processing unit; In the embodiment of the present invention, by selecting corresponding high-definition smart cameras, they are installed in different positions according to the layout of the stadium training area to ensure that there are no blind spots in vision. For example, one camera is installed at each of the four corners of the football field and above the center line on both sides. The smart display screen adopts Lenovo's touch-screen all-in-one machine and is placed in a conspicuous position at the entrance of the training area. The training equipment sensor is adapted according to the type of equipment. For example, a pressure sensor of AVIC Electronics is installed on the barbell, and a speed and acceleration sensor of Bosch is installed on the treadmill. The camera, smart display screen and training equipment sensor are connected to a Huawei switch through a Category 5e network cable, and then connected to a Dell server through optical fiber to realize device connection. When the training equipment is in motion, the camera, smart display screen and training equipment sensor are connected to the Huawei switch through an optical fiber. The trainee approaches the smart display screen and touches the screen. The system retrieves and displays his personalized teaching training plan from the database. The plan includes detailed content such as training items, duration, intensity, etc. The camera uses image recognition algorithm to lock the key parts of the trainee's body, such as shoulders and knees, and shoots training action videos at a frame rate of 30 frames per second. At the same time, the training equipment sensors collect real-time data on the pressure distribution, movement acceleration, and posture changes applied by the trainee on the equipment. These data are transmitted in real time to the central data processing unit through Qualcomm's 5G communication module in accordance with specific communication protocols. The unit is composed of high-performance workstations equipped with corresponding equipment for subsequent data processing.

[0021] S2: an error frame point recognition module, used to use the central data processing unit to perform error action time frame point recognition on the training action video corresponding to the trainee, so as to obtain the error action time frame point corresponding to the trainee; In an embodiment of the present invention, the central data processing unit uses the TensorFlow deep learning framework, combined with pre-trained convolutional neural network (CNN) and recurrent neural network (RNN) models to identify erroneous action time frame points in the training action video. First, the OpenCV library is used to parse the video into an image frame sequence, the image size is adjusted to 224×224 pixels and normalized, and these image frames are input into the CNN part. CNN extracts spatial features in the image, such as the trainee's body posture, limb movements, etc. through multiple convolutional layers and pooling layers. Then, the feature data extracted by CNN is input into RNN, which analyzes the temporal continuity of the action and captures the changes in the rhythm of the action. For example, in the analysis of basketball shooting action videos, the model analyzes the arm extension angle, body center of gravity position and action continuity of the shooter in each frame, and compares them with the standard actions and common incorrect actions in the training samples. If the arm bending angle of the shooter in a certain frame deviates too much from the standard action and the duration exceeds a certain threshold, the model identifies the time point corresponding to the frame as the incorrect action time frame point. By analyzing the entire video frame by frame, the model finally obtains the corresponding time frame point of the trainee's sports incorrect action, providing key information for subsequent analysis.

[0022] S3: a posture fit evaluation module, which is used to perform error time synchronization determination on the training action sensor data corresponding to the trainee based on the time frame point of the corresponding movement error action of the trainee, so as to obtain the training action sensor data corresponding to the trainee at the error time point; obtain the standard action of the sports training project, and perform action posture fit evaluation on the training action sensor data corresponding to the trainee at the error time point based on the standard action of the sports training project, so as to obtain the training action posture fit degree corresponding to the trainee; In the embodiment of the present invention, based on the time frame point of the exercise error action corresponding to the trainee, a time synchronization program written in Python is used to determine the training action sensor data corresponding to the error time point in the training action sensor data. The training action sensor data is stored in the database of the central data processing unit and arranged in chronological order. A time window of ±0.1 seconds is set based on the time corresponding to the error action time frame point. The data within the time range is filtered out by querying the database. For example, if the error action time frame point corresponds to the 15th second after the start of training, within the time window, the pressure at both ends of the barbell at this time is obtained from the pressure sensor. Pressure data, vertical acceleration data is obtained from the accelerometer, body posture angle data is obtained from the gyroscope, and at the same time, the corresponding standard action data of the sports training project is obtained from the action template library. The library is stored in the local server database and classified by sports projects. The posture matching algorithm is used, such as the posture matching algorithm based on Euclidean distance, to convert the standard action and the trainee's action posture data at the wrong time point into the same format, calculate the difference between the position and posture direction of each part of the body, and finally calculate the corresponding training action posture fit of the trainee, and quantify the closeness between the trainee's wrong action and the standard action.

[0023] S4: Feedback optimization and adjustment module, used to optimize and adjust the personalized teaching training plan corresponding to the trainee based on the fitness of the trainee's corresponding training movements and postures, so as to generate personalized teaching feedback suggestions corresponding to the trainee in the stadium.

[0024] In an embodiment of the present invention, the personalized teaching training plan is optimized and adjusted based on the training action posture fit of the trainee, using an optimization algorithm written in Python, and a training action and training intensity adjustment rule library is pre-built and stored in a database. For example, if the trainee's posture fit in the tennis serve action is low, the main problem is the wrong force sequence. The corresponding adjustment strategy is extracted from the rule library, and special training actions for the force sequence are added to the teaching training plan, such as setting up decomposition exercises for multiple serving stages, each stage focusing on the force of a specific part; in terms of training intensity, the total number of serves in each training session is appropriately reduced. The number of serves per training session was adjusted from 120 to 100 to avoid deformation of movements due to overtraining. Through the Python program, adjustment strategies were extracted from the rule library according to the trainee's specific fitness and irregular movements, and personalized feedback suggestions were generated. These suggestions were displayed on the screen in the form of a combination of graphics and text by utilizing the communication interface of the smart display. For example, for trainees with irregular running training movements, it was displayed that "it is recommended to increase pace rhythm training and adjust the training intensity to 4 times a week, 40 minutes each time", thereby generating personalized teaching feedback suggestions for the trainees in the stadium to help them improve their training results.

[0025] Furthermore, the action monitoring module includes the following functions: By connecting with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area, a comprehensive training visual monitoring network is built in the stadium training area; When the all-round training visual monitoring network detects that the trainee enters the sports training area and touches the smart display screen, the smart display screen responds to the trainee's corresponding training display touch operation instructions, and displays and formulates a personalized teaching training plan corresponding to the trainee on the smart display screen according to the training display touch operation instructions; The trainee performs sports training according to the corresponding personalized teaching training plan, and uses the camera to lock and monitor the trainee's corresponding position and training movements in real time during the sports training to generate the trainee's corresponding training movement video; The training equipment sensor is used to monitor the pressure distribution of the equipment corresponding to the trainee, the acceleration of the training action and the change of the training action posture in real time, so as to obtain the training action sensor data corresponding to the trainee; By adopting a high-speed wireless communication module, the trainee's corresponding training action video and training action sensor data are transmitted to the corresponding central data processing unit in real time.

[0026] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 The functional flow diagram of the action monitoring module in the embodiment includes the following functions: S11: Connect with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area to build a comprehensive training visual monitoring network in the stadium training area; In an embodiment of the present invention, cameras with high-definition shooting and intelligent tracking functions are selected in the training area of ​​the stadium, and are installed above and around the training area according to a specific layout to ensure that the entire training space can be covered without visual blind spots. For example, in a basketball training field, cameras are installed at the four corners of the field and above the basketball stand. At the same time, sensors compatible with the training equipment are configured, such as installing pressure sensors on dumbbells and speed and acceleration sensors on treadmills. These cameras, smart displays, and training equipment sensors are connected to switches through wired networks, and the switches are used to transmit data to servers to build a full-range training visual monitoring network. The smart display is set at the entrance of the training area to facilitate the operation of the trainer. By writing a special network configuration program and setting the IP address of each device, it is ensured that the devices can communicate stably and realize comprehensive monitoring of the stadium training area.

[0027] S12: When the all-round training visual monitoring network detects that the trainee enters the sports training area and touches the smart display screen, the smart display screen responds to the trainee's corresponding training display touch operation instruction, and displays and formulates a personalized teaching training plan corresponding to the trainee on the smart display screen according to the training display touch operation instruction; In an embodiment of the present invention, a capacitive touch screen technology is adopted through a smart display screen, and a high-sensitivity touch response function is provided. When a trainee enters a sports training area and touches the smart display screen, a sensing layer on the surface of the screen detects the corresponding touch operation of the trainee, triggering an internal circuit to generate a touch control signal. After receiving the signal, a microprocessor built into the display screen calculates the coordinates of the touch position. For example, when a trainee touches the "Personal Training Plan" option on the screen, the microprocessor calculates the coordinates of the touch point and converts it into a corresponding training display touch operation instruction. The display screen is connected to a database storing personalized teaching and training plans. Through a query statement, according to the trainee's identity information (such as obtained by scanning a student card or entering an account number), the personalized teaching and training plan corresponding to the trainee is retrieved from the database and displayed on the smart display screen in a graphic and textual form, including detailed contents such as training items, training duration, and training intensity.

[0028] S13: The trainee performs sports training according to the corresponding personalized teaching training plan, and uses a camera to lock and monitor the trainee's corresponding position and training movements in real time during the sports training to generate a training movement video of the trainee; In an embodiment of the present invention, the camera has an intelligent locking and tracking function, and an image recognition algorithm is used to identify the trainee. When the trainee starts to perform sports training, the camera obtains a video image of the training area, and identifies the trainee by analyzing the human features in the image, such as human contours, movement postures, etc. For example, in track and field training, the camera identifies the trainee's running movements and locks on key points of his body, such as the head, shoulders, knees, etc., and uses the movement trajectories of these key points to monitor the trainee's position and training movements in real time. The camera shoots video at a frame rate of 30 frames per second, stores the captured video data in a local cache, and numbers them in chronological order. After completing each minute of training, the video data in the cache is packaged, and finally a training movement video corresponding to the trainee is generated, providing intuitive image data for subsequent analysis.

[0029] S14: using the training equipment sensor to monitor in real time the pressure distribution of the equipment, the acceleration of the training movement, and the change of the training movement posture of the trainee, so as to obtain the training movement sensor data corresponding to the trainee; In the embodiment of the present invention, the training equipment sensor is configured according to the type of training equipment. In terms of strength training equipment, such as a barbell, a pressure sensor is installed on the barbell bar. By measuring the pressure on the barbell bar, the force and pressure distribution applied by the trainee are accurately monitored. For example, when the trainee performs squat training, the pressure sensor collects the pressure data of the barbell bar at different positions (at both ends and in the middle) in real time, and converts the pressure signal into a digital signal through the built-in signal conditioning circuit. For sports training equipment, such as an exercise bike, an acceleration sensor is installed at the wheel axle to measure the acceleration of the wheel rotation, thereby reflecting the change in the trainee's riding speed. At the same time, an inertial sensor is used to monitor the trainee's posture changes on the exercise bike, such as the body's tilt angle, and the data collected by these sensors are packaged in a specific data format (such as JSON format) through a wired or wireless communication module, and transmitted to a data acquisition terminal in real time, and finally the corresponding training action sensor data of the trainee is obtained.

[0030] S15: The training action video and training action sensor data corresponding to the trainee are transmitted to the corresponding central data processing unit in real time by using a high-speed wireless communication module.

[0031] In an embodiment of the present invention, by selecting a high-speed wireless communication module, such as a 5G communication module, the training action video and training action sensor data corresponding to the trainee are transmitted to the corresponding central data processing unit in real time. The training action video is encoded locally and an efficient video encoding algorithm, such as H.265, is used to compress the video data to an appropriate size to reduce the transmission bandwidth requirement. The training action sensor data is also compressed to remove redundant information. The 5G communication module encapsulates the encoded and compressed training action video and sensor data according to a specific communication protocol, and transmits the data to the central data processing unit through the connection between the 5G base station and the core network. The central data processing unit is configured with a high-performance server with powerful data storage and computing capabilities. It can receive, store and process a large amount of training data in real time, and provide data support for subsequent training analysis and teaching adjustments.

[0032] Furthermore, the training equipment sensor includes a pressure sensor, an acceleration sensor and a gyroscope.

[0033] Furthermore, the error frame point identification module includes the following functions: Using the central data processing unit to adjust the video frame rate of the training action video corresponding to the trainee, so as to generate the action frame-adjusted video corresponding to the trainee; Performing video denoising on the action frame-adjusted video corresponding to the trainee to generate an action denoised video corresponding to the trainee; By setting video frame points corresponding to 5-second intervals, and dividing the action denoising video corresponding to the trainee based on the corresponding video frame points, so as to generate training action frame segments corresponding to the trainee at each video frame point; The wrong action time frame points are identified for the training action frame segments corresponding to the trainee at each video frame point, so as to obtain the corresponding wrong action time frame points of the trainee.

[0034] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 The functional flow diagram of the error frame point identification module in the embodiment includes the following functions: S21: using the central data processing unit to adjust the video frame rate of the training action video corresponding to the trainee, so as to generate the action frame-adjusted video corresponding to the trainee; In an embodiment of the present invention, a high-performance workstation equipped with an NVIDIA RTX series graphics card is selected through a central data processing unit, and the video frame rate of the training action video corresponding to the trainee is adjusted by using the OpenCV video processing library. Assuming that the frame rate of the original training action video is 30 frames / second, in order to more clearly analyze the action details, the frame rate is adjusted to 60 frames / second. By reading each frame of the original video, the image is processed using a bilinear interpolation algorithm, and a new image frame is inserted between two adjacent frames to improve the smoothness of the video. For example, in a running training action video, the original video shows that the runner has a large action span between the first frame and the second frame. After the frame rate is adjusted, a new image frame is inserted between the two frames to more finely show the runner's action changes. The video data after the frame rate adjustment is re-encoded according to the new frame rate specification, and finally a frame-adjusted action video corresponding to the trainee is generated, providing more accurate material for subsequent video analysis.

[0035] S22: performing video denoising processing on the action frame-adjusted video corresponding to the trainee to generate an action denoised video corresponding to the trainee; In an embodiment of the present invention, by continuing to use the OpenCV library and the CUDA acceleration technology of the NVIDIA graphics card, the action frame-adjusting video corresponding to the trainee is subjected to video denoising processing, and a non-local mean denoising algorithm is adopted. The algorithm estimates and removes the noise of each pixel by searching for similar image blocks in the video frame. Taking an action frame-adjusting video of basketball shooting training as an example, due to factors such as the shooting environment, there are some noise points in the video, which affect the accurate observation of the shooting action. By setting a suitable search window size and similarity threshold, the algorithm searches for areas similar to the image blocks around the current pixel point in the video frame, calculates the weighted average of these similar areas, and replaces the value of the current pixel point with it, thereby removing the noise points. After denoising processing, the shooting action of the basketball player is clearer in the video, and the hand movement details and body posture changes can be more accurately presented. Finally, a denoised action video corresponding to the trainee is generated, the video quality is improved, and subsequent analysis is facilitated.

[0036] S23: by setting video frame points corresponding to intervals of 5 seconds, and dividing the action denoising video corresponding to the trainee based on the corresponding video frame points, so as to generate training action frame segments corresponding to the trainee at each video frame point; In an embodiment of the present invention, by using Python's OpenCV library and NumPy library, by setting video frame points corresponding to intervals of 5 seconds, and dividing the action denoising video corresponding to the trainee into video frames based on the corresponding video frame points, it is known that the frame rate of the action denoising video is 60 frames / second, then the number of video frames corresponding to 5 seconds is 5×60=300 frames, starting from the starting frame of the video, a video frame point is determined every 300 frames, for example, in a 30-second action denoising video, video frame points are set at the 0th frame, the 300th frame, the 600th frame, the 900th frame, the 1200th frame, and the 1500th frame, respectively, and then the video is divided into multiple segments according to these frame points, each segment contains 300 frames of images, for each segment, the video write function of OpenCV is used to save it as an independent video file, and finally the training action frame segments corresponding to the trainee at each video frame point are generated, so as to facilitate the separate analysis of the training actions in different time periods.

[0037] S24: performing erroneous action time frame point identification on the training action frame segments corresponding to the trainee at each video frame point, so as to obtain the corresponding motion erroneous action time frame points of the trainee.

[0038] In an embodiment of the present invention, by using a convolutional neural network (CNN) model in deep learning, the trainee's training action frame segments corresponding to each video frame point are identified for the wrong action time frame point, a large number of video frame segments containing correct and wrong training actions are collected in advance, and they are annotated to mark the time frame points where the wrong actions appear. Using these annotated data, a CNN-based wrong action recognition model is trained under the TensorFlow deep learning framework. The input of the model is the image data of the training action frame segment. Through the processing of multiple convolutional layers, pooling layers and fully connected layers, the judgment result of whether each time frame point is an wrong action is output. Taking a weightlifting training action frame segment as an example, the model analyzes the features of the athlete's body posture, the position of the barbell, etc. in each frame image, and compares them with the correct action features in the training data. If the model determines that the action in a certain frame image does not meet the correct standard, such as the athlete's waist bending angle is too large, exceeding the range of normal weightlifting action, the time point corresponding to the frame is identified as the wrong action time frame point. By analyzing all training action frame segments, the trainee's corresponding sports wrong action time frame point is finally obtained, providing key information for subsequent training guidance.

[0039] Furthermore, the step of identifying the time frame points of the wrong actions of the trainee at each video frame point corresponding to the training action frame segments includes: Obtain sports training action data after marking standard training actions and common incorrect actions; In an embodiment of the present invention, sports training action data that has been annotated with standard training actions and common incorrect actions are obtained from a professional sports training database and teaching resources organized by a coaching team. The professional sports training database is maintained by an authoritative sports organization and contains a large number of action videos of various sports and corresponding annotation information. For example, in track and field events, for running actions, standard actions such as standard starting posture, stride and frequency of running during the run, and action characteristics of the sprint stage are annotated in detail. Common incorrect actions such as slow reaction at the start and excessive forward leaning of the body when running are also annotated. The coaching team, based on years of teaching experience, organizes common incorrect actions such as incorrect shooting hand shape and incorrect force sequence and their corresponding video clips for specific sports, such as basketball shooting, and annotates them in detail. These annotated data from different channels are integrated and stored in the database of the local server according to sports event classification for easy subsequent call.

[0040] Preferably, the motion training motion data after the training standard motion and the common wrong motion are annotated are used as training samples and the motion recognition model is trained in combination with the convolutional neural network and the recurrent neural network to generate a motion training motion recognition model to accurately identify the standard motion and the common wrong motion corresponding to various sports items, and output the motion features corresponding to the motion training motion, including the motion rhythm and the motion amplitude corresponding to the exercise part of the trainee; In the embodiment of the present invention, by using Python's deep learning framework TensorFlow, the motion training motion data after training standard motions and common incorrect motions are annotated as training samples, and the motion recognition model is trained in combination with a convolutional neural network (CNN) and a recurrent neural network (RNN) to generate a motion training motion recognition model. First, the training sample data is preprocessed, the video frame image is adjusted to a uniform size, such as 224×224 pixels, and normalized. The image data is input into a CNN part composed of multiple convolutional layers and pooling layers. The CNN is responsible for extracting spatial features in the image, such as the athlete's body posture, the position of the moving parts, etc. Then, the C The feature data extracted by CNN is input into RNN. RNN can process time series information and capture the temporal continuity of actions, such as action rhythm. For example, when training a recognition model for table tennis serve action, CNN extracts spatial features such as the athlete's arm posture and racket position at the moment of serve, while RNN analyzes the changes in the serve action at different time points and identifies the action rhythm of the serve. By setting appropriate loss functions (such as cross entropy loss function) and optimizers (such as Adam optimizer), after multiple rounds of training, the model can accurately identify the standard actions and common incorrect actions corresponding to various sports, and output the action features corresponding to the sports training actions, such as the action rhythm and action amplitude corresponding to the trainee's movement parts.

[0041] Preferably, the training action frame segments corresponding to the trainee at each video frame point are input into a previously trained sports training action recognition model for action feature recognition analysis to obtain the training action video frame features corresponding to the trainee at each video frame point; In an embodiment of the present invention, the training action frame segments corresponding to the trainee at each video frame point are input into a previously trained sports training action recognition model for action feature recognition analysis, so as to use Python's OpenCV library to read the video data of the training action frame segments and convert them into the input format required by the model. For example, the video frame image is adjusted to a size of 224×224 pixels and normalized, and the processed image data is input into the sports training action recognition model frame by frame. The CNN part of the model extracts the spatial features of each frame image, and the RNN part combines the time series information to analyze the continuity of the action. Taking a badminton racket swing training action frame segment as an example, the model analyzes the position and posture changes of the athlete's arm and wrist in each frame, and combines the time sequence to identify the rhythm of the swing action, such as the time interval from the preparation posture to the moment of hitting the ball, and the action amplitude, such as the maximum angle of the arm swing. By analyzing the entire training action frame segment, the training action video frame features corresponding to the trainee at each video frame point are finally obtained, including the action rhythm, action amplitude and other related action feature information.

[0042] Preferably, based on the training action video frame features corresponding to the trainee at each video frame point, the training action frame segments corresponding to the trainee at each video frame point are identified for the erroneous action time frame points to obtain the corresponding motion erroneous action time frame points of the trainee.

[0043] In an embodiment of the present invention, by identifying the time frame points of incorrect actions of the training action frame segments corresponding to the trainee at each video frame point based on the video frame features of the training action corresponding to the trainee at each video frame point, the video frame features of the training action are compared with the standard action features in the training sample, and the incorrect actions are identified using pre-set judgment rules. For example, for weightlifting training actions, if the video frame features of the training action show that the waist bending angle of the athlete exceeds the range specified in the standard action during the process of lifting the barbell, and the duration exceeds a certain threshold (such as 0.5 seconds), the time frame points corresponding to the time period are identified as the time frame points of incorrect actions. By analyzing the video frame features of all the training action frame segments one by one, the time frame points of incorrect actions that occur during the training process of the trainee are fully identified, providing accurate information on incorrect actions for subsequent training guidance, helping the trainee to correct errors in a timely manner, and improving the training effect.

[0044] Further, the identifying of the wrong action time frame points of the training action frame segments corresponding to the trainee at each video frame point based on the training action video frame features corresponding to the trainee at each video frame point includes: The preliminary action detail differences between the video frame features of the trainee's training action at each video frame point and the standard action in the pre-built action template library are quantified to obtain the action detail deviation between the trainee and the standard action at each video frame point; In an embodiment of the present invention, by using Python's NumPy library and a pre-built action template library, the video frame features of the training action corresponding to each video frame point of the trainer and the standard action in the action template library are quantified for preliminary action detail differences. The action template library contains detailed feature data of standard actions of various sports at different time points, such as joint angles, body part position coordinates, etc. Taking swimming training as an example, in the action template library, the standard freestyle action at a certain video frame point has an arm stroke angle of 120 degrees and a hand position coordinate of [x1, y1, z1]. The trainer has an arm stroke angle of 110 degrees and a hand position coordinate of [x2, y2, z2] at the video frame point. The angle difference is calculated as 10 degrees by the function of the NumPy library, and the position coordinate difference is calculated using the spatial distance formula, such as sqrt ((x2- x1)^2 + (y2 - y1)^2 + (z2 - z1)^2), to obtain a specific position deviation value. The motion features of the trainee at each video frame point and the standard motion features are comprehensively calculated to obtain the detailed deviations of the trainee's motions at each video frame point from the standard motions, such as angle deviation, position deviation, etc., and organize them into data tables for subsequent analysis.

[0045] Preferably, based on the deviation of the action details between the trainee and the standard action at each video frame point, the video frame points corresponding to the trainee are screened for suspected frame points. If the corresponding action detail deviation is greater than or equal to a preset action deviation threshold, it is considered that the trainee is suspected of having a corresponding wrong action at the video frame point, otherwise, it is considered that the trainee is suspected of having a corresponding wrong action at the video frame point, and the video frame points corresponding to the suspected wrong action are screened out to obtain a time frame point set of suspected wrong actions corresponding to the trainee; In an embodiment of the present invention, based on the deviation of the trainee's action details between the trainee and the standard action at each video frame point, a screening program written in Python is used to screen the video frame points corresponding to the trainee for suspected frames, and a motion deviation threshold is preset. For example, for a basketball shooting action, the arm bending angle deviation threshold is set to 5 degrees, and the body tilt angle deviation threshold is set to 3 degrees. For each video frame point, the action detail deviation is compared with the preset threshold. For example, at a certain video frame point, the trainee's arm bending angle deviation is 6 degrees when shooting, which exceeds the threshold of 5 degrees, and the body tilt angle deviation is 4 degrees, which also exceeds the threshold of 3 degrees. At this time, it is considered that the trainee is suspected of having a corresponding erroneous action at this video frame point. All video frame points suspected of having erroneous actions are screened out and stored in a list to form a time frame point set of suspected erroneous actions corresponding to the trainee, providing target data for subsequent further analysis.

[0046] Preferably, the action detail deviation between the adjacent time frames and the standard action is obtained through the time frame point set of the suspected wrong action corresponding to the trainee, and the action deviation trend analysis is performed according to the action detail deviation between the adjacent time frames and the standard action, so as to obtain the change trend of the action deviation between the trainee and the standard action in the suspected adjacent time frames; In an embodiment of the present invention, a Python data analysis script is used to obtain the action detail deviations between adjacent time frames and the standard action from the time frame point set of suspected erroneous actions corresponding to the trainee. Taking tennis serve training as an example, it is assumed that the time frame point set of suspected erroneous actions includes frame point A, frame point B, frame point C, etc. For the two adjacent time frames of frame point A and frame point B, their deviation data from the standard action are extracted from the action detail deviation data table, such as arm swing speed deviation, racket angle deviation, etc., and data analysis tools such as pandas library are used to analyze these deviation data, and the changes in the deviations between adjacent time frames are calculated, such as calculating the difference in arm swing speed deviation between adjacent time frames. By analyzing the deviation changes of multiple adjacent time frame pairs, the change trend of the trainee's action deviation between the standard action in the suspected adjacent time frames is obtained, such as the trend that the arm swing speed deviation shows a gradual increase, or the racket angle deviation first decreases and then increases, etc., and the deviation changes are presented in the form of a chart to intuitively present the deviation changes.

[0047] Preferably, based on the variation trend of the trainee's movement deviation from the standard movement in the suspected adjacent time frames, the corresponding adjacent time frames in the suspected erroneous movement time frame point set are subjected to the position movement difference correlation analysis, so as to obtain the corresponding mutual relationship of the trainee's position movement difference between the suspected adjacent time frames; In an embodiment of the present invention, based on the trend of the deviation of the trainee's movements from the standard movements in the suspected adjacent time frames, an association analysis algorithm, such as the Apriori algorithm, is used to perform an association analysis of the differences in the movements of the corresponding adjacent time frames in the suspected erroneous movement time frame point set. Taking football shooting training as an example, in the suspected adjacent time frames, the deviation changes of the movement parts such as the leg force part, the body center of gravity position, and the foot contacting the ball part are analyzed, and through the Apriori algorithm, appropriate support and confidence thresholds are set, such as the support is 0.3 and the confidence is 0.8. The algorithm searches for frequent item sets in the data. For example, it is found that when the deviation of the leg force part increases, the deviation of the body center of gravity position also increases frequently, and the set support and confidence are met, so as to obtain the corresponding relationship between the differences in the movement of the trainee's parts between the suspected adjacent time frames, and clarify the association law between the deviations of the movements of different parts.

[0048] Preferably, based on the mutual relationship between the differences in the trainee's part movements between suspected adjacent time frames, the time frame points of suspected erroneous movements corresponding to the trainee are identified; if there is a correlation between the deviations of the part movements between adjacent time frames and the deviations show a trend of gradually increasing, it is judged that there is an erroneous movement between the suspected adjacent time frames, and the corresponding average time points between the suspected adjacent time frames are calculated as the corresponding erroneous movement frame points, so as to obtain the corresponding movement erroneous movement time frame points of the trainee.

[0049] In an embodiment of the present invention, based on the corresponding relationship between the differences in the body movements of the trainee between suspected adjacent time frames, a judgment program written in Python is used to identify the time frame points of the suspected incorrect movements of the trainee. Taking weightlifting training as an example, between the suspected adjacent time frames, if it is found that the waist bending angle deviation is correlated with the arm force angle deviation, and both deviations show a trend of gradually increasing, and the pre-set incorrect movement judgment conditions are met, it is judged that there is an incorrect movement between the suspected adjacent time frames, and the corresponding average time points between the suspected adjacent time frames are calculated. For example, if the adjacent time frames are the 10th frame and the 12th frame, the time interval is 0.2 seconds, and the average time point is the time corresponding to the 11th frame, it is used as the corresponding incorrect movement frame point. By analyzing the entire suspected incorrect movement time frame point set, the corresponding sports incorrect movement time frame points of the trainee are comprehensively identified, and accurate incorrect movement information is provided for subsequent training guidance, helping the trainee to improve training movements.

[0050] Furthermore, the posture fit assessment module includes the following functions: Based on the time frame point of the trainee's corresponding motion error action, the trainee's corresponding training motion sensor data is synchronously determined to obtain the trainee's corresponding training motion sensor data at the error time point, including pressure distribution, motion acceleration and motion posture; In an embodiment of the present invention, by utilizing a time synchronization program written in Python, based on the time frame point of the trainee's corresponding motion error action, the training motion sensor data corresponding to the trainee is synchronized and determined at the error time. The training motion sensor data is stored in a database of a local server and arranged in chronological order. For example, it is known that the time corresponding to a time frame point of a trainee's motion error action is the 15th second after the start of the training. By querying the database, a very small time window, such as ±0.1 seconds, is set based on the time point to filter out the training motion sensor data within this time range. It is assumed that within the time window, the pressure distribution data corresponding to the trainee at the error time point is obtained from the pressure sensor, such as the pressures at both ends of the barbell are 300N and 320N respectively; the motion acceleration data is obtained from the acceleration sensor, such as the vertical acceleration of 2m / s²; the motion posture data is obtained from the gyroscope, such as the body inclination angle of 10 degrees. These data are organized into a data record, and finally the training motion sensor data corresponding to the trainee at the error time point is obtained, providing an accurate data basis for subsequent analysis.

[0051] Preferably, the corresponding sports training items are obtained through the corresponding training action sensor data of the trainee at the wrong time point; In an embodiment of the present invention, by using a pre-constructed association table of sports items and sensor data features, the corresponding sports training items are obtained through the corresponding training action sensor data of the trainee at the wrong time point. The association table records in detail the sensor data features generated by different sports items during the training process. For example, for weightlifting items, the pressure distribution data is usually large, and the acceleration data has obvious changes in the process of lifting and lowering the barbell; while for yoga items, the pressure distribution is relatively uniform, the action acceleration is small, and the action posture changes are diverse but relatively slow. Taking the pressure distribution data of the trainee at the wrong time point (such as the pressure at both ends of the barbell is 300N and 320N respectively), the action acceleration data (vertical acceleration is 2m / s²) and the action posture data (body inclination angle is 10 degrees) as an example, through a matching program written in Python, these data are compared with the features in the association table, and it is found that these data features are highly matched with the features of the weightlifting item, thereby determining that the sports training item corresponding to the trainee is weightlifting, providing a basis for subsequent acquisition of standard actions.

[0052] Preferably, according to the sports training items corresponding to the trainee, the corresponding standard movements of the sports training items are obtained in the corresponding movement template library; In an embodiment of the present invention, by using Python's database query statement according to the sports training project (such as weightlifting) corresponding to the trainee, the corresponding standard movement of the sports training project is obtained in the corresponding action template library. The action template library is stored in the database of the local server, and the standard movement data is stored according to the sports project classification. For example, for weightlifting, the action template library contains detailed data of standard snatch, clean and jerk and other movements at different stages, such as the movement trajectory of the barbell, the posture angles of various parts of the body, the force sequence, etc. By executing the query statement "SELECT standard movement data FROM action template library WHERE sports project = 'weightlifting'", the standard movement data of the weightlifting project is retrieved from the database, and these data are read into the memory and stored in the form of a data structure, which is convenient for subsequent comparison and analysis with the trainee's incorrect movement data, and provides a standard reference for the movement posture fit evaluation.

[0053] Preferably, based on the standard movements of the sports training items, the training movement sensor data corresponding to the trainee at the wrong time point is evaluated for movement posture fit, so as to obtain the training movement posture fit of the trainee.

[0054] In an embodiment of the present invention, by using a posture matching algorithm, such as a posture matching algorithm based on Euclidean distance, the training movement sensor data corresponding to the trainee at the wrong time point is evaluated for movement posture matching based on the standard movement of the sports training project. First, the posture data in the standard movement of the sports training project (such as the angles of each joint of the body, the coordinates of the position of the body part, etc.) and the movement posture data of the trainee at the wrong time point (such as the body tilt angle of 10 degrees) are converted into the same data format. Taking the body tilt angle as an example, the tilt angle of the standard movement at this time point is 8 degrees. The Euclidean distance formula is used to calculate the difference between the two, that is, sqrt ((10-8)^2)=2 degrees. For multiple posture data dimensions, such as considering the arm extension angle, leg bending angle, etc., the difference of each dimension is calculated respectively, and the weighted sum is performed according to the pre-set weight. Assuming that the weight of the body tilt angle is 0.4, the weight of the arm extension angle is 0.3, and the weight of the leg bending angle is 0.3, the comprehensive difference value is calculated and converted into the movement posture fit through a pre-set mapping relationship. For example, if the comprehensive difference value is 5, the corresponding movement posture fit is 70%. This quantifies the degree of fit between the trainee's incorrect movements and standard movements, providing a quantitative basis for training guidance.

[0055] Furthermore, the performing of movement posture matching evaluation on the training movement sensor data corresponding to the trainee at the wrong time point based on the standard movement of the sports training item includes: By mapping the standard movements of the sports training items and the movement postures in the training movement sensor data corresponding to the trainee at the wrong time point to the same spatial coordinate system, the corresponding standard training movements and the trainee's wrong movement postures in the same spatial coordinate system are generated; In an embodiment of the present invention, a spatial geometry algorithm is used to determine the position and direction of the corresponding standard training action and the trainee's incorrect action posture in the same spatial coordinate system. Taking the trainee's basketball shooting action as an example, in the unified coordinate system, at the moment of basketball shooting in the standard action, the position coordinates of the hand are [x_std, y_std, z_std]. At this time, the direction of the arm can be represented by a unit vector, such as obtained by calculating the position vector of the hand and shoulder joint and normalizing it. For the trainee's incorrect action posture, the position coordinates of the hand at the moment of basketball shooting are also determined to be [x_err, y_err, z_err], as well as the arm direction vector. A program written in Python is used to extract these position coordinates and direction vector information. By analyzing the changes in the position coordinates and direction vectors of the standard action and the incorrect action at different time points, the posture position and posture direction corresponding to the standard training action and the trainee's incorrect action in the same spatial coordinate system are obtained. For example, in the entire shooting action process, the position and direction information of each key time point are recorded respectively to provide a data basis for the subsequent calculation of distance and deviation.

[0056] Preferably, the position distance between the posture positions corresponding to the standard training action and the trainer's incorrect action in the same spatial coordinate system is calculated to obtain the posture position distance between the standard training action and the trainer's incorrect action in the same spatial coordinate system; In the embodiment of the present invention, by using the Euclidean distance formula and with the help of Python's NumPy library, the position distance between the posture positions corresponding to the standard training action and the trainer's incorrect action in the same spatial coordinate system is calculated. Assuming that at a certain moment, the position coordinates of a key part of the body (such as the shoulder) in the standard training action are [x1, y1, z1], and the position coordinates of the corresponding part in the trainer's incorrect action are [x2, y2, z2]. The Euclidean distance between the two is calculated by the function of the NumPy library, that is, Taking a soccer shot as an example, at the moment of shooting, the position coordinates of the point where the foot touches the ball in the standard action may be different from the position coordinates of the point where the foot touches the ball in the trainee's incorrect action. By calculating the distance between these two position coordinates, the position difference between the trainee's incorrect action and the standard action can be quantified. This calculation is performed on multiple key time points in the entire training action process, and finally a posture position distance sequence between the standard training action and the trainee's incorrect action in the same spatial coordinate system is obtained to reflect the changes in the position difference at different times.

[0057] Preferably, the direction deviation between the posture direction corresponding to the standard training action and the trainer's erroneous action in the same spatial coordinate system is measured, so as to obtain the posture direction deviation between the standard training action and the trainer's erroneous action in the same spatial coordinate system; In an embodiment of the present invention, according to the vector operation rules, the NumPy library of Python is used to measure the direction deviation between the posture directions corresponding to the standard training action and the trainer's incorrect action in the same spatial coordinate system. For example, in a tennis serving action, the direction vector of the racket at the moment of serving in the standard action is [v1x, v1y, v1z], and the direction vector of the racket at the moment of serving in the trainer's incorrect action is [v2x, v2y, v2z]. First, the two vectors are normalized to obtain unit vectors [u1x, u1y, u1z] and [u2x, u2y, u2z]. Then, the vector dot product formula is used to calculate the cosine of the angle between the two unit vectors, that is, dot ([u1x, u1y, u1z], [u2x, u2y, u2z]), and then the included angle is obtained through the arc cosine function. The included angle is the posture direction deviation. For the direction vectors of multiple key parts, such as the direction of the central axis of the body, the direction of the arm swing, etc., the direction deviations are calculated respectively, and the weighted average is performed according to the pre-set weights. Finally, the posture direction deviation between the standard training action and the trainee's wrong action in the same spatial coordinate system is obtained. In this way, the degree of difference in direction between the trainee's wrong action and the standard action can be accurately measured.

[0058] Preferably, the posture fit calculation formula is used to evaluate the posture fit of the posture position distance and posture direction deviation between the training standard movement and the trainee's erroneous movement in the same spatial coordinate system to obtain the training posture fit of the trainee.

[0059] In an embodiment of the present invention, a suitable action posture fit calculation formula is formed by combining the total number of nodes corresponding to the action posture position, the node coordinates corresponding to the training standard action and the trainee's erroneous action, the posture position distance between the training standard action and the trainee's erroneous action, the posture direction deviation between the training standard action and the trainee's erroneous action, the erroneous time frame point corresponding to the action posture position and the corresponding time attenuation factor to perform a fit evaluation calculation, so as to comprehensively consider the position distance and the direction deviation, quantify the fit of the trainee's training action posture, and help the trainee understand the fit between his own action and the standard action, and finally obtain the corresponding training action posture fit of the trainee. In addition, the action posture fit calculation formula can also use any action posture matching method in the field to replace the action posture fit evaluation process, and is not limited to the action posture fit calculation formula.

[0060] Furthermore, the action posture matching calculation formula is specifically: ; In the formula, To train the coordination of movements and postures, is the total number of nodes corresponding to the action posture position, To train standard movements The node coordinates corresponding to the action posture position, For trainers wrong action The node coordinates corresponding to the action posture position, It is the distance between the standard training action and the trainer's wrong action. To train the posture direction deviation between the standard action and the trainee's wrong action, For the The error time frame point corresponding to the action posture position, is the time decay factor.

[0061] The present invention obtains a motion posture fit calculation formula by using a specific mathematical model and after verification, which is used to evaluate the motion posture fit of the posture position distance and posture direction deviation between the training standard action and the trainer's wrong action in the same spatial coordinate system, wherein the formula fully considers the training action posture fit , the total number of nodes corresponding to the action posture position , training standard action The node coordinates corresponding to the action posture position , the trainer's wrong action The node coordinates corresponding to the action posture position , the distance between the standard training action and the trainer's wrong action , the posture direction deviation between the training standard action and the trainer's wrong action , No. The wrong time frame points corresponding to the action posture positions , time decay factor , according to the training action posture fit The correlation between the above parameters constitutes a functional relationship , the formula can realize the posture fit evaluation process of the posture position distance and posture direction deviation between the standard training action and the trainer's wrong action in the same spatial coordinate system. At the same time, the posture fit calculation formula can accurately evaluate the trainer's action accuracy when performing sports training items by comprehensively considering the action position distance and posture direction deviation. In this way, the difference between the trainer's action and the standard action can be quantified as a fit value, making the action evaluation in the training process more objective and accurate. By introducing the influence of the time decay factor and the time point, the action posture performed at the wrong time point is given more attention. Through the time decay processing, the influence of the time error when the trainer performs the action on the final fit is properly controlled, making the training evaluation closer to the actual training progress and the timeliness of the action execution. In addition, by including a weighted decay term in the formula when calculating the posture position distance and direction deviation of each node, the errors that occur at more advanced or critical time points (such as incorrect actions in the earlier stage) have a greater impact on the final evaluation, which can help the system pay more attention to the accuracy of the critical action stage. This weighting can better reflect the trainer's action accuracy at critical moments. By combining the posture position distance with the direction deviation for evaluation, it is not only possible to determine whether the trainee's action is in the correct position (position in the spatial coordinate system), but also to evaluate whether the direction of the trainee's action is accurate (angle error). This comprehensive evaluation method provides more detailed training feedback, which helps trainees discover and improve multi-dimensional problems in their actions. Therefore, by calculating each action posture node one by one and taking a weighted average of the error of each node, the formula not only ensures a comprehensive evaluation of the entire action chain, but also provides a global analysis of the action through the total number of nodes (N). The error of each posture position is taken into consideration, so as to provide a more complete and systematic action evaluation feedback.

[0062] Furthermore, the feedback optimization and adjustment module includes the following functions: The training action posture conformity of the trainee is compared and judged according to the preset training action conformity threshold. If the training action posture conformity of the trainee is greater than or equal to the preset training action conformity threshold, it is considered that the actual training effect of the trainee has reached the standard; if the training action posture conformity of the trainee is less than the preset training action conformity threshold, it is considered that the actual training effect of the trainee has not reached the standard, and there are problems of poor teaching training effect or irregular action. In an embodiment of the present invention, a training action matching threshold is pre-set in a Python programming environment. For example, for a basketball shooting action, the matching threshold is set to 0.8, and the training action posture matching degree corresponding to the trainee is compared with the threshold. The conditional judgment statement of Python is used, such as "if training action posture matching degree>=training action matching threshold", to execute the corresponding logic. When the training action posture matching degree is less than the preset threshold, it is determined that the actual training effect corresponding to the trainee does not meet the standard, and there are problems of poor teaching and training effect or irregular action. Taking tennis serve training as an example, if the calculated posture matching degree of the trainee's serve action is 0.7, which is less than the preset threshold of 0.8, the identification information of the trainee and the non-compliance situation are recorded through the Python program to provide a basis for subsequent optimization.

[0063] Preferably, the personalized teaching and training plan corresponding to the trainee is optimized and adjusted through teaching feedback according to the trainee's corresponding problems of poor teaching and training effects or irregular movements, so as to automatically generate corresponding personalized feedback suggestions, including adjusting training movements and changing training intensity, and timely feedback is pushed to the trainee through the smart display screen, so as to generate personalized teaching feedback suggestions corresponding to the trainee in the stadium.

[0064] In an embodiment of the present invention, when it is determined that the trainee has poor teaching and training effects or irregular movements, the optimization algorithm written in Python is used to perform teaching feedback optimization and adjustment on the personalized teaching and training plan corresponding to the trainee. The algorithm is based on a pre-constructed rule library for adjusting training movements and training intensity. For example, for irregular basketball shooting movements, if the main problem is a large deviation in the shooting angle, the rule library is correspondingly adjusted to add special training movements for correcting the shooting angle, such as setting a shooting angle positioning auxiliary device for practice; in terms of training intensity, the number of shots per training is appropriately reduced from the original 100 shots per training. , adjusted to 80 times to prevent the trainees from becoming more irregular in their movements due to excessive fatigue. Through the Python program, according to the trainees' specific irregular movements, the corresponding adjustment strategies are extracted from the rule library, and corresponding personalized feedback suggestions are generated. The communication interface of the smart display screen is used to display these suggestions on the screen in the form of pictures and texts. For example, for trainees with irregular football shooting movements, the smart display screen will display "It is recommended to increase special training for the shooting force part, and the training intensity is adjusted to 3 times a week, 30 minutes each time" to generate corresponding personalized teaching feedback suggestions for trainees in the stadium, so as to help trainees improve their training effects.

[0065] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0066] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A stadium intelligent sports training teaching device, characterized in that: Includes the following modules: The motion monitoring module is used to connect with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area, and use the smart display screen to display the personalized teaching training plan while using the cameras and training equipment sensors to lock and monitor the corresponding training action videos and training action sensor data of the trainees in real time; The training action video and training action sensor data of the trainee are transmitted to the corresponding central data processing unit in real time by using a high-speed wireless communication module; The error frame point recognition module is used to use the central data processing unit to perform error action time frame point recognition on the training action video corresponding to the trainee, so as to obtain the error action time frame point corresponding to the trainee; The posture fit evaluation module is used to perform error time synchronization determination on the training action sensor data corresponding to the trainee based on the time frame point of the trainee's corresponding motion error action, so as to obtain the training action sensor data corresponding to the trainee at the error time point; obtain the standard action of the sports training project, and perform action posture fit evaluation on the training action sensor data corresponding to the trainee at the error time point based on the standard action of the sports training project, so as to obtain the training action posture fit degree corresponding to the trainee; The feedback optimization and adjustment module is used to optimize and adjust the teaching feedback of the trainee's corresponding personalized teaching training plan based on the trainee's corresponding training movement posture fit, so as to generate personalized teaching feedback suggestions for the trainee in the stadium.

2. The stadium intelligent sports training teaching device according to claim 1 is characterized in that: The motion monitoring module includes the following functions: By connecting with the corresponding cameras, smart display screens and training equipment sensors in the stadium training area, a comprehensive training visual monitoring network is built in the stadium training area; When the all-round training visual monitoring network detects that the trainee enters the sports training area and touches the smart display screen, the smart display screen responds to the trainee's corresponding training display touch operation instructions, and displays and formulates a personalized teaching training plan corresponding to the trainee on the smart display screen according to the training display touch operation instructions; The trainee performs sports training according to the corresponding personalized teaching training plan, and uses the camera to lock and monitor the trainee's corresponding position and training movements in real time during the sports training to generate the trainee's corresponding training movement video; The training equipment sensor is used to monitor the pressure distribution of the equipment corresponding to the trainee, the acceleration of the training action and the change of the training action posture in real time, so as to obtain the training action sensor data corresponding to the trainee; By adopting a high-speed wireless communication module, the trainee's corresponding training action video and training action sensor data are transmitted to the corresponding central data processing unit in real time.

3. The stadium intelligent sports training teaching device according to claim 2 is characterized in that: The training equipment sensor includes a pressure sensor, an acceleration sensor and a gyroscope.

4. The stadium intelligent sports training teaching device according to claim 1, characterized in that: The error frame point identification module includes the following functions: Using the central data processing unit to adjust the video frame rate of the training action video corresponding to the trainee, so as to generate the action frame-adjusted video corresponding to the trainee; Performing video denoising on the action frame-adjusted video corresponding to the trainee to generate an action denoised video corresponding to the trainee; By setting video frame points corresponding to 5-second intervals, and dividing the action denoising video corresponding to the trainee based on the corresponding video frame points, so as to generate training action frame segments corresponding to the trainee at each video frame point; The wrong action time frame points are identified for the training action frame segments corresponding to the trainee at each video frame point, so as to obtain the corresponding wrong action time frame points of the trainee.

5. The stadium intelligent sports training teaching device according to claim 4 is characterized in that: The identifying of the wrong action time frame points of the training action frame segments corresponding to each video frame point of the trainee comprises: Obtain sports training action data after marking standard training actions and common incorrect actions; Based on the sports training action data after the training standard action and common incorrect action are annotated as training samples and combined with the convolutional neural network and the recurrent neural network to train the action recognition model, a sports training action recognition model is generated to accurately identify the standard actions and common incorrect actions corresponding to various sports items, and output the action features corresponding to the sports training actions, including the action rhythm and action amplitude corresponding to the exercise parts of the trainee; Input the training action frame segments corresponding to the trainee at each video frame point into the previously trained sports training action recognition model for action feature recognition analysis to obtain the training action video frame features corresponding to the trainee at each video frame point; Based on the training action video frame features corresponding to the trainee at each video frame point, the training action frame segments corresponding to the trainee at each video frame point are identified with the wrong action time frame points to obtain the corresponding motion wrong action time frame points of the trainee.

6. The stadium intelligent sports training teaching device according to claim 5, characterized in that: The step of identifying the wrong action time frame points of the training action frame segments corresponding to the trainee at each video frame point based on the training action video frame features corresponding to the trainee at each video frame point includes: The preliminary action detail differences between the video frame features of the trainee's training action at each video frame point and the standard action in the pre-built action template library are quantified to obtain the action detail deviation between the trainee and the standard action at each video frame point; Based on the deviation of the trainee's action details between the standard action at each video frame point, the video frame points corresponding to the trainee are screened for suspected frames. If the corresponding action detail deviation is greater than or equal to the preset action deviation threshold, it is considered that the trainee is suspected of having a corresponding wrong action at the video frame point, otherwise it is not considered, and the video frame points corresponding to the suspected wrong action are screened out to obtain a time frame point set of suspected wrong actions corresponding to the trainee; The action detail deviation between the adjacent time frames and the standard action is obtained through the time frame point set of the suspected wrong action corresponding to the trainee, and the action deviation trend analysis is performed according to the action detail deviation between the adjacent time frames and the standard action, so as to obtain the change trend of the action deviation between the trainee and the standard action in the suspected adjacent time frames; Based on the variation trend of the trainee's movement deviation from the standard movement in the suspected adjacent time frames, the corresponding adjacent time frames in the suspected erroneous movement time frame point set are subjected to the position movement difference correlation analysis, so as to obtain the corresponding mutual relationship of the trainee's position movement difference between the suspected adjacent time frames; Based on the mutual relationship between the differences in the trainees' part movements between suspected adjacent time frames, the incorrect movement time frame points are identified for the set of suspected incorrect movement time frame points corresponding to the trainees. If there is a correlation between the deviations of the part movements between adjacent time frames and the deviations show a trend of gradually increasing, it is judged that there is an incorrect movement between the suspected adjacent time frames, and the corresponding average time points between the suspected adjacent time frames are calculated as the corresponding incorrect movement frame points, so as to obtain the corresponding movement incorrect movement time frame points of the trainees.

7. The stadium intelligent sports training teaching device according to claim 2, characterized in that: The posture fit assessment module includes the following functions: Based on the time frame point of the trainee's corresponding motion error action, the trainee's corresponding training motion sensor data is synchronously determined to obtain the trainee's corresponding training motion sensor data at the error time point, including pressure distribution, motion acceleration and motion posture; Obtaining corresponding sports training items through the corresponding training action sensor data of the trainee at the wrong time point; According to the sports training items corresponding to the trainee, the corresponding standard movements of the sports training items are obtained from the corresponding movement template library; Based on the standard movements of the sports training items, the training movement sensor data corresponding to the trainee at the wrong time point is evaluated for movement posture fit, and the training movement posture fit of the trainee is obtained.

8. The stadium intelligent sports training teaching device according to claim 7, characterized in that: The step of evaluating the motion posture conformity of the training motion sensor data corresponding to the trainee at the wrong time point based on the standard motion of the sports training item includes: By mapping the standard movements of the sports training items and the movement postures in the training movement sensor data corresponding to the trainee at the wrong time point to the same spatial coordinate system, the corresponding standard training movements and the trainee's wrong movement postures in the same spatial coordinate system are generated; Determine the posture position and direction of the standard training action and the trainer's incorrect posture in the same spatial coordinate system, so as to obtain the posture position and posture direction corresponding to the standard training action and the trainer's incorrect posture in the same spatial coordinate system; Calculating the position distance between the posture positions corresponding to the standard training action and the trainer's erroneous action in the same spatial coordinate system to obtain the posture position distance between the standard training action and the trainer's erroneous action in the same spatial coordinate system; According to the direction deviation measurement between the posture directions corresponding to the standard training action and the trainer's erroneous action in the same spatial coordinate system, the posture direction deviation between the standard training action and the trainer's erroneous action in the same spatial coordinate system is obtained; The posture fit calculation formula is used to evaluate the posture fit of the posture position distance and posture direction deviation between the standard training movement and the trainee's incorrect movement in the same spatial coordinate system, and the corresponding training posture fit of the trainee is obtained.

9. The stadium intelligent sports training teaching device according to claim 8, characterized in that: The action posture matching calculation formula is specifically: ； In the formula, To train the coordination of movements and postures, is the total number of nodes corresponding to the action posture position, To train standard movements The node coordinates corresponding to the action posture position, For trainers wrong action The node coordinates corresponding to the action posture position, It is the distance between the standard training action and the trainer's wrong action. To train the posture direction deviation between the standard action and the trainee's wrong action, For the The error time frame point corresponding to the action posture position, is the time decay factor.

10. The stadium intelligent sports training teaching device according to claim 1, characterized in that: The feedback optimization and adjustment module includes the following functions: The training action posture conformity of the trainee is compared and judged according to the preset training action conformity threshold. If the training action posture conformity of the trainee is greater than or equal to the preset training action conformity threshold, it is considered that the actual training effect of the trainee has reached the standard. If the training action posture conformity of the trainee is less than the preset training action conformity threshold, it is considered that the actual training effect of the trainee does not meet the standard, and there are problems such as poor teaching training effect or non-standard action; According to the trainee's corresponding poor teaching and training results or irregular movements, the personalized teaching and training plan corresponding to the trainee is optimized and adjusted through teaching feedback to automatically generate corresponding personalized feedback suggestions, including adjusting training movements and changing training intensity, and timely feedback is pushed to the trainee through the smart display screen to generate personalized teaching feedback suggestions for the trainee on the stadium.

Images