Real-Time Position Estimation and Tracking of a Basketball

Abstract

In this paper, an algorithm to detect the position of a basketball in a real time outdoor video is proposed. The problem of ball detection in sports video arises due to the occlusion of the ball with the players, the presence of many moving objects like spectators, flags, twigs and branches of the tree etc. The ball image is also getting distorted due to the high speed of the ball. Thus the conventional direct detection and tracking algorithm fails in various occasions. It is much advantageous to study the trajectory information of a ball and estimate the ball location using the trajectory information. In this trajectory based algorithm, firstly the ball like objects are detected and a set of ball candidates are generated for each frame. Then the set of candidate trajectories are plotted. From these potential trajectories, the actual ball trajectory is estimated by studying the trajectory information. The missing ball location in any frame can also be predicted by trajectory interpolation technique. In this paper, we also propose an efficient background detection method using the approximate median value of the background pixels, thus achieving a high degree of segmentations of the moving objects in the foreground. This algorithm provides excellent result under a high degree of occlusion and works satisfactorily in the presence of many moving objects along with the ball in the scene.