Toward event recognition using dynamic trajectory analysis and prediction

In this paper we propose a trajectory analysis method suited for event recognition. The method works online, in the sense that it can process the data as they are acquired by the sensors and it is dynamic, since it adapts the results to the changes in the patterns of activity. For each class of objects detected by the system, the proposed method groups trajectories with common features in clusters and, based on the identification of common prefixes in the clusters, can make probabilistic predictions on the possible future positions of a moving object. This analysis can give valuable information to an event recognition system for the identification of anomalous events.