Analysis and evaluation of tennis swings using computer vision

Abstract

The need for the automated analysis of sporting activities has increased dramatically. In this study, we analyzed and quantitatively evaluated tennis swings using computer vision. Two cameras are used to detect the positions of balls and rackets in 3-D. By subtracting successive images from each other, the relative positions and 3-D trajectories of balls and racket centers are calculated. An ellipse fitting method is used to estimate the racket contours and the normal vectors of the racket face are determined. Racket face angles are calculated using the components of the normal racket vectors, with three different shots (flat shots, slice shots and drive shots) being quantitatively discriminated using these angles. Ball and racket speeds for five stroke types ranging from beginner to expert are calculated and compared to evaluate shot techniques.