A Novel Approach to Tennis Game Analysis and Emulation using Computer Vision

Continuous practice, analysis and game experience are essential to tennis players hoping to improve their game. While practice for tennis players is being made available through thousands of academies around the world, game analysis and game experience are lacking. In this paper, we propose a method of obtaining a set of statistics quantifying a tennis player's game from a video using an optimized version of a state-of-the-art computer vision algorithm to detect the position of the ball in a frame in real-time and infer the state of the game through a series of such detections. We also propose a system that is capable of virtually simulating a player's game in real-time using a graphics processing unit. This virtual simulation can be extended to real life with the use of three ball shooting machines, a camera, and a processing unit. We are able to obtain the statistics with an accuracy of 70% on a test set of two videos consisting of tennis rallies.