Research And Implementation Of Video-based Golf Swing Evaluation System

Golf swing is the core part of golf.The standard degree of swing posture determines whether the force is smooth when hitting the ball,the distance that the ball flies and the accuracy of the ball’s landing position.However,the golf swing is a complex whole-body movement that requires great coordination of the body.It often takes a lot of practice to get a standard pose.Amateur players cannot get guidance from professional coaches all the time during daily training,which affects training efficiency.To address this issue,this paper combines a deep learning model to design a video-based golf swing evaluation system to help amateur players obtain a portable artificial intelligence golf “coach”.The system is dedicated to analyzing several key events included in the whole golf swing process,and the degree of posture standardization in each key event is directly related to the hitting effect.It is meaningful for players to analyze the swing action at key event frames to improve the swing performance.The research content of this paper can be summarized in the following three points:(1)This paper proposes a framework to recognize key events in golf swing based on pure monocular video data.The framework combines attention mechanism in the backbone network to extract more concise features and utilizes the transformer structure to fuse multi-scale temporal information to enhance feature representation.In addition,Gaussian kernels are also introduced into the label generation process,which can effectively solve the ambiguity problem when detecting key events in adjacent similar frames.Notably,the method proposed in this paper achieves an average recognition accuracy of 83.4% for the eight golf swing events on the Golf DB dataset(7.3% improvement compared to Swing Net).At the same time,it also achieved good results on the Fine Diving dataset,which proves the effectiveness of the proposed framework and its potential to be extended to other sports key event detection.(2)According to this key event detection algorithm,combined with human body pose estimation to predict the coordinates of human skeleton points,this paper proposes a comparative analysis method for swing movements.Different from viewing the skeleton points of all frames of the entire swing,using the coordinates in the key frame can capture the problem of the golfer’s swing with a higher probability.The method compares the difference between the coordinates of the body skeleton points of professional players and amateur players,and calculates the Euclidean distance of human skeleton points in key events,supplemented by the calculation of the angle of the body joints to perform quantitative analysis of the swing.(3)The above method is implemented in the form of We Chat applet,and a golf full swing evaluation system is developed.Except for comparative analysis of swing movements,the system can also view 3D swing movements from multiple angles,which is not limited by the shooting angle of the uploaded video.The hardware device required by the system is only a smart phone,without wearing any sensors,users can use the system conveniently through We Chat.In addition to allowing student users to conduct supervised swing training anytime and anywhere,the system also provides teacher users with references for action scoring and future classroom teaching plans.At the same time,it is also convenient for teachers to manage class student information,which greatly improves the efficiency of golf teaching.