Unit detection in American football TV broadcasts using average energy of audio track

In this paper, we explore the domain of American Football TV broadcasting with respect to interesting event detection. Most of the existing methods achieve the goal of event detection in sports video using some type of shot or scene based approaches. Many problems arise when it comes to work in the shot or scene level in the American Football domain. The big amount of camera movement, such as tilting, zooming, panning, and the large number of different angles used by different broadcasting networks make it very difficult to work in the shot/scene level. We show how in any of these two levels, over segmentation is almost inevitable. We also propose a novel concept that answers most of the problems of the domain we work on. This is the concept of "unit". A unit is defined as good workable data, which is not necessary to be very accurate. The main purpose of the unit detection process we present is to locate these potentially interesting segments in the video, and separate them from the rest of the redundant data.