A wearable sensing system for timing analysis in tennis

Abstract

Wearables find in sports one of their main applications. In recent years, many wearable devices have been commercially released such as the Babolat Play or Sony Smart Tennis Sensor that detect and classify different types of tennis shots and provide a performance analysis to the player. However, available devices focus on a single technical element of tennis only - the shot. As tennis performance is the result of a full body coordination and timing of the movement, the present work wants to take a broader view at the tennis player performance and include the simultaneous work of legs and arms with the goal to time elements of movement. We design a sensor system with three inertial measurement units, one attached to each foot as well as one at the racket. We develop a pipeline to detect and classify leg and arm movement and implement a gesture recognition for the shooting arm based on LCSS (longest common subsequence). The algorithm distinguishes between forehand and backhand (with topspin and slice, respectively) as well as a smash. Footwork is first segmented into potential steps and then classified by a support vector machine between shot and side steps. In the person-dependent case the algorithm achieved 87% recall and 89% precision. The step recognition algorithm has been able to detect 76% of the steps with a classification accuracy of 95%. Based on these results timing information within the shooting state can be robustly obtained which is crucial for a thorough analysis of the whole shot.