Research And Application Of Motion-assisted Evaluation System Based On Human Pose Recognition

In recent years,with the rapid development of computer technology,computer vision has been better applied by virtue of deep learning.Research scholars have begun to combine artificial intelligence technology to achieve scientific sports auxiliary training.On the basis of traditional sports training,human-machine interactive auxiliary training functions have been added to achieve the index of sports norms and the unmanned teaching effect of coaching.Traditional sports have the limitations of venues and equipment,professional movements require special guidance,and training data is difficult to record.This paper applies human posture recognition to gymnastics,establishes a sports-assisted evaluation system,and introduces joint angles and similarity of movements.Concept,to solve the difference in impact caused by the height of different people and the problem of screening actions in consecutive frames,to achieve a transition from a sports training method based on experience to a human vision analysis training method based on computer vision.The main research work and contributions of this article as follows:(1)The basic principles of human gesture recognition based on deep learning are studied.The open source library of human gesture recognition Open Pose is selected and the input data is improved on its original basis to achieve more rapid extraction of 2D coordinate information of 14 key points of human body.The results of posture rendering are compared with the two-dimensional coordinate data to complete its accuracy verification.(2)Researched motion recognition and evaluation based on joint angle and motion similarity,used joint angle to describe the motion to grasp more details in the motion,introduced the concept of motion similarity in motion recognition,and found two motions using Euclidean distance The degree of similarity between them can complete the recognition of actions in a single image,and can also realize the recognition of actions in videos.(3)Developed a motion assisted evaluation system based on human posture recognition.The system is mainly composed of three parts: standard action database,auxiliary teaching,and overall evaluation.The system user can customize the standard action database,and the auxiliary teaching part of the system can compare the user's actions and standards.The movements are compared and displayed intuitively to the trainer in the form of data.The overall evaluation part of the system can perform motion recognition and screening on the video file to provide the trainer with a data-based training platform.