Research And Implementation Of Key Techniques For Real-Time Video Semantic Analysis Of Tennis Match Based On Deep Learning

With the arrival of the artificial intelligence era,the traditional broadcast form of sports competitions has changed greatly.The tennis eagle eye system,which came out as early as the beginning of the 20th century,also serves as a referee aid and plays an important role in tennis competitions.However,there is currently no available technology for semantic analysis of live video of tennis matches.It is our main goal to bring a more digital and intelligent experience to form of traditional tennis competition broadcast through advanced video analysis technology based on deep learning.The thesis focuses on the two major sports objects in the tennis competition—athletes and tennis.The semantic information include the type of action of the athlete,the athletic distance,the speed and the landing area of the tennis.First of all,people and tennis ball in the video frame of the tennis are very small,we propose a detection algorithm for small targets.Secondly,we design a sports player tracking algorithm based on character division and continue to lock the target players to be tracked and output the player area.After that,our research goal is from coarse to fine,and design discrimination algorithms of action category and athlete' s running distance and speed based on the displacement of the human key points and detection algorithm of athlete's moving distance and speed.Through the two-dimensional attitude estimation of the players in the player area,the body skeleton and key point information are analyzed to complete the estimation of player's action category.At the same time,according to the displacement information of the athlete's body key points and the system running time,the athlete's movement distance and movement speed are obtained.Then because small tennis is difficult to capture in high-speed sports,we design a tennis motion and landing area prediction algorithm based on prior knowledge to get drop area of tennis.Finally,we implement a semantic analysis prototype system for real-time tennis video,test and analyz the performance of the system.Experiment results indicate that the system has good performance in real-time,accuracy and stability.