Long Term Tracking of Pedestrians with Groups and Occlusions

Abstract

This chapter describes an algorithm for tracking groups of pedestrians in video sequences. The main difficulties addressed in this work concern total occlusions of the objects to be tracked, as well as group merging and splitting. Because there is ambiguity, the algorithm should be able to provide the most probable interpretation of the data. A two layer solution is proposed. The first layer produces a set of spatiotemporal trajectories based on low level operations which manage to track the pedestrians most of the time. The second layer performs a consistent labeling of the detected segments using a statistical model based on Bayesian networks. The Bayesian network is recursively computed during the tracking operation and allows the update of the tracker results every time new information is available. Interpretation/recognition errors can thus be detected after receiving enough information about the group of interacting objects. Experimental tests are included to show the performance of the algorithm in complex situations. This work was supported by FEDER and FCT under project LT (POSI 37844/01).