Estimation of the ball size in broadcast soccer video using salient objects

This paper presents an algorithm for estimating the ball size for each frame in broadcast soccer video by using various types of salient objects. Four types of salient objects, which are the center line, the ellipse, the goalmouth, and the people, are used to estimate the ball size. Before estimating the ball size, we have obtained all the video segments with the field by a video segmentor. For each such video segment, the proposed algorithm takes three steps to obtain the ball size for each frame. Experimental results show that the proposed algorithm achieves above 97% recall in detecting the center line, above 82% recall in detecting the ellipse, and above 92% recall in detecting goalmouth. Moreover, we achieve 100% precision for all detections. With the accurate results of salient object detections, we can achieve the accurate estimates of the ball size.