Middle school student broadcast gymnastics teaching evaluation system and method based on Al technology

Abstract

The invention discloses a middle school student broadcast gymnastics teaching evaluation system and method based on the Al technology, and the system comprises a visual collection module, an image processing module, an operation module, and a display module, the visual collection module is electrically connected with the image processing module, the image processing module is electrically connected with the operation module, and the display module is electrically connected with the operation module. The operation module is electrically connected with the display module, and the display module is further provided with a voice broadcast device. The problems that in the prior art, broadcasting gymnastics training actions of middle school students cannot be efficiently and accurately evaluated, and the dance teaching and training effect can be affected are solved.

Description

Technical Field

[0001] The present invention relates to the field of deep learning and computer vision technology, and in particular to an AI technology-based system and method for evaluating the teaching of radio gymnastics for middle school students. Background Art

[0002] For the online evaluation of radio gymnastics for middle school students, there is currently no complete system process and method for analyzing and evaluating radio gymnastics for middle school students. Therefore, the online evaluation and teaching of radio gymnastics for middle school students still face the following challenges. First, the existing evaluation methods of radio gymnastics for middle school students lack accurate evaluation scales. The current standards are mostly descriptive words rather than quantitative values, which makes individual evaluations have the characteristics of weak objectivity, quantification and repeatability. Second, the online teaching mode of radio gymnastics for middle school students makes the manual evaluation workload of teachers large, and teachers urgently need an efficient, accurate and scientific system to reduce the workload. Third, the one-to-many teaching mode makes it difficult for teachers to accurately grasp the overall teaching quality and individual level. Fourth, the existing AI technology model is only made of templates that integrate dance and exercise projects. Fifth, in addition to in-class learning, students cannot practice and correct errors independently at any time outside of class.

[0003] Combining the above five analysis points, it can be seen that the existing technology cannot efficiently and accurately evaluate the radio gymnastics training movements of middle school students, which will affect the effect of gymnastics teaching and training. Summary of the invention

[0004] The purpose of the present invention is to overcome the existing defects and provide a system and method for evaluating the teaching of radio gymnastics for middle school students based on AI technology, which can efficiently and accurately evaluate the training movements of radio gymnastics and improve the quality and level of teaching.

[0005] The technical scheme for achieving the above object is: a middle school student radio gymnastics teaching evaluation system based on AI technology, including a visual acquisition module, an image processing module, a calculation module and a display module, the visual acquisition module is electrically connected to the image processing module, the image processing module is electrically connected to the calculation module, the calculation module is electrically connected to the display module, and the display module is also provided with a voice announcer; The radio gymnastics teaching evaluation method includes the above-mentioned radio gymnastics teaching evaluation system, and the evaluation method includes the following steps: S1 students do exercises facing the camera in the visual acquisition module, and the visual acquisition module transmits the picture data to the image processing module; The image processing module S2 extracts image frame data, converts the raw data into a form that can be understood by the machine learning algorithm, extracts 13 key points of the human body from the image frame through the MediaPipe processing framework, and calculates the body posture angle data of 8 parts to generate a feature vector; The S3 image processing module imports the processed feature vector into the dance model in the operation module to obtain the processed data. The model output data is a digital number classified by humans. The digital number is parsed into intuitive text action name information according to the classification dictionary table. The S4 operation module transmits the processed information to the display module, which displays the action judgment and overall scoring information.

[0006] Preferably, the steps of establishing the dance model are as follows: S11 transmits the video to the image processing module based on the provided standard action video; S12 extracts valid image frames from video files through the OpenCV framework; S13 deletes erroneous data and garbled images, and removes the data coordinate values ​​of the end nodes of the facial features, hands and feet that have little or no impact on the action score; S14 selects 13 key points of the human body, including the nose, left and right wrists, elbows, shoulders, hips, knees, and ankles. After automatically predicting the coordinates of the points using MediaPipe, it calculates the preset left elbow angle, right elbow angle, left arm angle from the body, right arm angle from the body, left shoulder-hip-knee angle, right shoulder-hip-knee angle, left leg angle, and right leg angle, and normalizes the body posture angle data of these eight parts. S15 establishes the standard for the normalized angles of the eight body parts in each section and movement of radio gymnastics; S16 combines the complexity of radio gymnastics video movements, the dimension of the data, and the interpretability of the model, and uses scikit-learn's K-nearest neighbor classifier to train the data and generate a model; S17 evaluates the model by comparing it with a large number of student dance videos in order to achieve the most standard model; S18 data segmentation: divide the data set into a training set and a test set, with a ratio of 75% for training and 25% for testing. Then perform model training and fit the model using the training set data. Finally, evaluate the generalization ability of the model through cross-validation and optimize the model through parameter and data adjustment. S19 exports the trained model as a ".pkl" model file using joblib, and then loads the stored model file into the system of the operation module through joblib. The runtime environment library version is consistent with the Python library version during model training.

[0007] Preferably, in step S14, the human body posture is effectively reconstructed by identifying and locating key points of the human body in the image, and when performing body posture angle detection, deep learning models such as convolutional neural networks and recursive neural networks are used to achieve accurate positioning and recognition of multiple categories of body postures.

[0008] Preferably, the method for effectively reconstructing human body posture also includes a progressive reconstruction algorithm and a skeleton modeling technique.

[0009] The beneficial effects of the present invention are as follows: students do exercises facing the camera in the visual acquisition module, and the processed information is transmitted to the display module through the calculation module, and the display module displays the action judgment and overall scoring information. This system can efficiently and accurately evaluate each training action of the radio gymnastics of middle school students, facilitate the improvement of the teaching quality and level of gymnastics teaching and training, and provide a new analysis method for the online education of radio gymnastics for middle school students. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is an evaluation flow chart of the present invention; Figure 2 It is a flow chart of establishing the dance model of the present invention; Figure 3 It is a flow chart of the dance model training of the present invention; Figure 4 It is a schematic diagram of the positions of key points and posture angles of the human body in the dance model of the present invention; Figure 5 is the input flow chart of the scikit-learn algorithm of the present invention; Figure 6 It is a coordinate action diagram of key points of a human body in the visual acquisition module of the present invention; Figure 7 It is an Excel table file generated by calculating the posture angle by the coordinates of the key points of the human body in the image processing module of the present invention; Figure 8 It is another Excel table file generated by data classification and annotation after the image processing module of the present invention calculates the posture angle by using the coordinates of the key points of the human body. DETAILED DESCRIPTION

[0011] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0012] The present invention will be further described below in conjunction with the accompanying drawings.

[0013] Combination Figure 1-3 Analysis, a middle school student radio gymnastics teaching evaluation system based on Al technology, including a visual acquisition module, an image processing module, a calculation module, and a display module. The visual acquisition module is electrically connected to the image processing module, the image processing module is electrically connected to the calculation module, the calculation module is electrically connected to the display module, and a voice announcer is also provided on the display module. The interface of the display module terminal is integrated, and an intuitive user interface is designed to display the model prediction results to the terminal user in an aesthetically pleasing, intuitive and easy-to-understand manner. In the main function interface, a concise voice prompt is used, and an accurate and intuitive countdown is used to enter the function. While playing a standard demonstration video, the user action video stream obtained by the camera is mirrored on the interface below the demonstration video to ensure that the user can effectively interact with the system synchronously. At the same time, the terminal computing power is used to process real-time data, and the real-time results of the model are obtained through websocket interaction with the background. Then, the scoring results are displayed in real time with intuitive charts below the demonstration video. This can help students to practice and correct radio gymnastics independently at any time outside of class.

[0014] The radio gymnastics teaching evaluation method includes the above-mentioned radio gymnastics teaching evaluation system, and the evaluation method includes the following steps: S1 students do exercises facing the camera in the visual acquisition module, and the visual acquisition module transmits the picture data to the image processing module; like Figure 4 As shown in the figure (the 32 human body points in the figure represent the collection points of various parts of the human body, and the 8 angle points in the figure represent the left elbow angle, the right elbow angle, the left arm off the body angle, the right arm off the body angle, the left shoulder-hip-knee angle, the right shoulder-hip-knee angle, the left leg angle and the right leg angle), the S2 image processing module extracts the image frame data, converts the original data into a form that can be understood by the machine learning algorithm, selects the nose, left and right wrists, elbows, shoulders, hips, knees, ankles, a total of 13 human key points, and uses MediaPipe to automatically predict the point coordinates. The preset left elbow angle, right elbow angle, left arm off the body angle, right arm off the body angle, left shoulder-hip-knee angle, right shoulder-hip-knee angle, left leg angle and right leg angle are calculated. Since the video may have different sizes and proportions, it is necessary to first standardize the 8 body posture angles. The normalization method is used to convert the angle values ​​to a unified scale range. Although continuous actions have time correlation, since the scoring uses rhythm point key frames, time series construction is not used; The S3 image processing module imports the processed feature vector into the dance model in the operation module to obtain the processed data. The model output data is a digital number classified by humans. The digital number is parsed into intuitive text action name information according to the classification dictionary table. The S4 operation module transmits the processed information to the display module, which displays the action judgment and overall scoring information.

[0015] The steps to establish the dance model are as follows: S11 transmits the video to the image processing module based on the provided standard action video; S12 extracts valid image frames from video files through the OpenCV framework; S13 deletes erroneous data and garbled images, and removes the data coordinate values ​​of the end nodes of the facial features, hands and feet that have little or no impact on the action score; S14 binding Figure 6-8 Analysis: S14 selects 13 key points of the human body, including the nose, left and right wrists, elbows, shoulders, hips, knees, and ankles. After automatically predicting the coordinates of the points using MediaPipe, it calculates the preset left elbow angle, right elbow angle, left arm angle from the body, right arm angle from the body, left shoulder-hip-knee angle, right shoulder-hip-knee angle, left leg angle, and right leg angle. The body posture angle data of these 8 parts are normalized and exported to an Excel file. Then, the images and data are manually labeled. The image name corresponds to the "frame" in the data list. The process actions are uniformly labeled as *3, and the standard actions are uniformly labeled as *9. S15 establishes standards for the normalized angle data of the body posture of 8 parts of each section and action of radio gymnastics. The use of supervised learning requires the labeling and classification of the feature data list, and the manual labeling of each video clip to classify its action category in detail, indicating which category of action it belongs to. The labeling work of the classification of radio gymnastics actions for middle school students is guided by relevant physical teachers to ensure the accuracy of the classification. In addition to the evaluation of posture and action, physical qualities that are easily recognized by the machine in each section of gymnastics (such as upper and lower limb muscle strength, flexibility, reaction ability, displacement speed, coordination, sensitivity, explosive power, balance ability, etc., some of which are consistent with the action detection, and some are different, all of which are labeled) are added to facilitate the multidimensionality of subsequent reports; Combination Figure 5 Analysis, S16 combines the complexity of the middle school students' radio gymnastics video movements, the dimension of the data and the interpretability of the model, and uses scikit-learn's K nearest neighbor classifier to train the data and generate a model.

[0016] S17 evaluates the model by comparing it with a large number of student dance videos in order to achieve the most standard model; S18 Data segmentation: divide the data set into a training set and a test set, with a ratio of 75% for training and 25% for testing. Then perform model training and fit the model using the training set data. Finally, evaluate the generalization ability of the model through cross-validation. Tune the model through parameter and data adjustment to adjust the model performance to a usable or even satisfactory performance state. S19 exports the trained model as a ".pkl" model file using joblib, and then loads the stored model file into the system of the operation module through joblib. The runtime environment library version is consistent with the Python library version during model training.

[0017] In step S14, the human body posture is effectively reconstructed by identifying and locating the key points of the human body in the image. When performing key point detection, deep learning models such as convolutional neural networks and recursive neural networks are used to achieve accurate positioning and recognition of multi-category key points.

[0018] Methods for effectively reconstructing human posture also include progressive reconstruction algorithms and skeleton modeling technology.

[0019] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some or all of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A middle school student radio gymnastics teaching evaluation system based on AI technology, characterized in that: It includes a visual acquisition module, an image processing module, a computing module and a display module, wherein the visual acquisition module is electrically connected to the image processing module, the image processing module is electrically connected to the computing module, the computing module is electrically connected to the display module, and a voice announcer is also provided on the display module; The radio gymnastics teaching evaluation method includes the above-mentioned radio gymnastics teaching evaluation system, and the evaluation method includes the following steps: S1 students do exercises facing the camera in the visual acquisition module, and the visual acquisition module transmits the picture data to the image processing module; The image processing module S2 extracts image frame data, converts the raw data into a form that can be understood by the machine learning algorithm, extracts the coordinates of 33 key points of the human body from the image frame through the MediaPipe processing framework, generates angles from the three-point coordinates, and selects 8 body posture angle data to generate feature vectors; The S3 image processing module imports the processed feature vector into the dance model in the operation module to obtain the processed data. The model output data is a digital number classified by humans. The digital number is parsed into intuitive text action name information according to the classification dictionary table. The S4 operation module transmits the processed information to the display module, which displays the action judgment and overall scoring information.

2. The method for evaluating the teaching of radio gymnastics for middle school students based on AI technology according to claim 1 is characterized in that: The steps of establishing the dance model are as follows: S11 transmits the video to the image processing module based on the provided standard action video; S12 extracts valid image frames from video files through the OpenCV framework; S13 deletes erroneous data and garbled images, and removes the data coordinate values ​​of the end nodes of the facial features, hands and feet that have little or no impact on the action score; S14 selects 13 key points of the human body, including the nose, left and right wrists, elbows, shoulders, hips, knees, and ankles. After automatically predicting the coordinates of the points using MediaPipe, it calculates the preset left elbow angle, right elbow angle, left arm angle from the body, right arm angle from the body, left shoulder-hip-knee angle, right shoulder-hip-knee angle, left leg angle, and right leg angle, and normalizes the body posture angle data of these eight parts. S15 establishes the standard for the normalized coordinates of the eight body posture angles in each section of the radio gymnastics; S16 combines the complexity of the middle school students' radio gymnastics video movements, the dimension of the data, and the interpretability of the model, and uses scikit-learn's K nearest neighbor classifier to train the data and generate a model; S17 evaluates the model by comparing it with a large number of student dance videos in order to achieve the most standard model; S18 data segmentation: divide the data set into a training set and a test set, with a ratio of 75% for training and 25% for testing. Then perform model training and fit the model using the training set data. Finally, evaluate the generalization ability of the model through cross-validation and optimize the model through parameter and data adjustment. S19 uses joblib to export the trained model as a ".pkl" model file, and then loads the stored model file into the system of the operation module through joblib. The runtime environment library version is consistent with the Python library version during model training.

3. The method for evaluating the teaching of radio gymnastics for middle school students based on AI technology according to claim 2 is characterized in that: In step S14, the human body posture is effectively reconstructed by identifying and locating the key points of the human body in the image. When detecting the body posture angle, deep learning models such as convolutional neural networks and recursive neural networks are used to achieve accurate positioning and recognition of multiple categories of body postures.

4. The method for evaluating the teaching of radio gymnastics for middle school students based on AI technology according to claim 3 is characterized in that: Methods for effectively reconstructing human posture also include progressive reconstruction algorithms and skeleton modeling technology.

Images