BEHAVIORAL FILTERING OF HUMAN TRAJECTORIES FOR AUTOMATIC-MULTI-TRACK INITIATION
In this paper we describe a new approach for the multitrack initiation problem. We propose an extensive use of the top-view reconstruction of the scene to solve the detection step in tracking pedestrians. We leave a large set of starting hypothetical moving objects free to evolve in the scene for a certain number of frames. The number of trajectories is pre-filtered using distance and direction constraints on a one-step movement of pedestrians. The output trajectories from pre-filtering step are then filtered using a discrete choice model for pedestrian behavior, calibrated on real data. The results show that is possible to use this technique to perform multitarget tracking in real situations. We particularly focus on an application in the context of automatic video surveillance.
Record created on 2006-06-14, modified on 2016-08-08