Shot Change Detection and Shot Classification in Broadcast Soccer Video

Abstract

This paper deals with shot change detection and classification for broadcast soccer video. To extract interesting scenes in soccer games such as goal scenes, it is essential to know the movements of playersand a ball. Since broadcast soccer video is composed of many shots from several cameras, we need to extract shots which capture a wide area of the ground (called ground camera shots) for estimating their movements. We therefore develop a method of detecting shot changes, which can work even for those with a special effect such as fade in, and that of classifying shots into several ground camera shots and others. Experimental results with about 25-minute video show the effectiveness of the method.