Event detection using multiple event probability sequences

We model human activities in videos using event probability sequences that detect events based on stable, state-level changes in learned hidden Markov models (HMM). The probability of an event occurring at every point along a motion trajectory is computed to form an event probability sequence. In this paper we propose extensions of the event probability sequences approach to handle multiple trajectories, in which events are associated with activities, rather than individual trajectories. Preliminary activity recognition experiments using indoor video sequences provide encouraging results.