Unusual activity analysis in video sequences

Abstract

We present a unique representation scheme for events in an area under surveillance, which provides a mechanism to analyze videos from multiple perspectives for unusual activity analysis. We propose clustering in event component spaces and define algebraic operations on these clusters to find co-occurrences of event components. A usualness measure for clusters is proposed that not only gives a measure on how usual or unusual an activity is, but also a basis for analyzing and predicting the possibly usual or unusual activities that can occur in the surveillance region. © Springer-Verlag Berlin Heidelberg 2007.