Ball tracking and virtual replays for innovative tennis broadcasts

Abstract

This paper presents a real-time computer vision system that tracks the motion of a tennis ball in 3 0 using multiple cameras. Ball tracking enables virtual replays, new game statistics, and other visualizations which result in very new ways of experiencing and analyzing tennis matches. The system has been used in international television broadcasts and webcasts of more than 15 matches. Six cameras around a stadium, divided into four pairs, are currently used to track the ball on serves which sometimes exceed speeds of 225 kmph. A multi-threaded approach is taken to tracking where each thread tracks the ball in a pair of cameras based on motion, intensity and shape, performs stereo matching CO obtain the 30 trajectory, detects when a ball goes out of view of its camera pair, and initializes and triggers a subsequent thread. This efSlcient approach is scalable to many more cameras tracking multiple objects. The ready acceptance of the system indicates the growing potential for multi-camera based real-time tracking in broadcast applications.

TenniVis: Visualization for Tennis Match Analysis

“…For tennis, much of the existing work utilizes ball and player tracking information to provide virtual replays of tennis matches [20]. This tracking information is also used to provide summary level information about player strategies such as player movement [21,28].…”

Intelligent Middleware for Adaptive Sensing of Tennis Coaching Sessions

“…However, the majority of approaches use broadcast sports video and can exploit the editing style of the content, such as the close-up of a player after a soccer goal, for example. Player-tracking and ball-tracking are also common methods for extracting semantic knowledge and the majority of approaches employ background subtraction or frame differencing to obtain candidate blobs which are then classified, tracked or discarded [8], [5], [14]. The main difficulty in such approaches is the problem of occlusion, but this can be overcome by using multiple cameras [14] or using an overhead camera, as we do in this work.…”

“…Player-tracking and ball-tracking are also common methods for extracting semantic knowledge and the majority of approaches employ background subtraction or frame differencing to obtain candidate blobs which are then classified, tracked or discarded [8], [5], [14]. The main difficulty in such approaches is the problem of occlusion, but this can be overcome by using multiple cameras [14] or using an overhead camera, as we do in this work. Candidate detections can be filtered if they do not conform to certain constraints, such as knowledge of the ball's colour [5].…”

SINGLE-IMAGE 3D RECONSTRUCTION OF BALL VELOCITY AND SPIN FROM MOTION BLUR - An Experiment in Motion-from-Blur

“…On the other hand, the problem of estimating the motion of a ball in the 3D space has been extensively treated in video tracking literature (Gopal Pingali and Jean, 2000;J Ren and Xu, 2004;Jonathan Rubin and Stevens, 2005). These methods assume the ball visible from multiple synchronized cameras, in order to triangulate the ball position in the corresponding frames.…”