QUALITATIVE SPATIO-TEMPORAL REPRESENTATION FOR EVENT EXTRACTION FROM VIDEO DATA OF FOOTBALL GAMES

摘要

We describe a qualitative representation of the spatial relations between extracted regions of video data, and discuss event occurrence based on this representation. We use video footage of football games and investigate a formal-ization to determine whether an event has occurred, specifically a pass or shot for a goal. We represent mereologi-cal and directional relations of regions in each frame based on extracted regions of objects, and determine event occurrence from the sequence of these relations. This qualitative spatio-temporal method reduces computational complexity and provides clear semantics denning an event.