Research On Application Of Deep Learning In Video Analysis Of Table Tennis Match

In a table tennis game,the movement data of table tennis players plays an important role in daily training.In recent years,video analysis technology has been widely used in sports competitions.With the development of deep learning,how to use deep learning to analyze sports game videos to obtain sports data of athletes has become more and more important.This article is to use deep learning to research and analyze the athletes in the table tennis competition,and then obtain the athletes' sports data,which has important guiding significance for athletes and coaches in daily training.This paper designs and implements a table tennis game video analysis system based on deep learning,which can detect,track and analyze table tennis players.First,use the deep learning target detection algorithm YOLOv4 to detect table tennis players.Second,use the Improved unscented Kalman filter tracking algorithm to continuously track the table tennis players.According to the position trajectory changes of the athlete's center point,the movement distance is obtained by calculating the Euclidean distance,and then Obtain the movement speed;according to the three performance evaluation indicators of accuracy,IOU value,and m AP accuracy,the system's detection performance of table tennis players is analyzed,and the comparison chart of the position trajectory changes of the center points of the two players in a certain round is selected and analyzed The influence of movement amplitude and movement frequency on the winning or losing of the ball and the final result of the game is analyzed;finally,the influence on the winning or losing of a certain round and the final result of the game is analyzed through the movement distance and speed of the two athletes.In order to test the detection performance of table tennis players using the deep learning target detection algorithm YOLOv4 in this system,a table tennis game video data set is selected for verification.The test results show that the algorithm has achieved good performance in terms of accuracy and precision.To a certain extent,it meets the target detection requirements of the deep learning table tennis game video analysis system.Use the Improved unscented Kalman filter algorithm to track two athletes in a certain round of the table tennis match,and analyze the tracking and analysis of the center point position of the two athletes to obtain the average difference between the upper and lower peaks and the movement frequency value,and obtain the upper and lower peaks of the athletes.When both the average difference and the movement frequency value are large,the probability of winning is greater;when the athlete's movement distance and movement speed in a certain round of the game are both greater,the probability of winning is also greater.