TennisMaster: an IMU-based online serve performance evaluation system

Abstract

Tennis sport has become more popular all over the world in recent years. While tennis lovers wish to improve their tennis skill set for better performance, unfortunately only few of them could be guided under professional training. Especially, serve is probably the most important skill in tennis skill set. In this paper, we present TennisMaster, an online diagnosis and feedback system, that aims at performing online assessment of tennis serve during the training process using IMU sensors. In particular, we propose a hierarchical evaluation approach based on the fusion of two IMU sensors mounted on the racket and shank of the player. In order to achieve online serve assessment, we first develop an online serve extraction algorithm to identify the serve segments and filter the non-serve events. Then we use Hidden Markov Model (HMM) to segment the serve process into eight stages. By extracting unique features on the basis of the serve segmentation, we build a regression model which outputs the score of a serve. We conduct experiments to collect 1,030 serves involving 12 subjects at various professional levels. Evaluation results show that our system achieves high accuracy of performance assessment for tennis serves.