Occlusion management strategies for pedestrians tracking across fisheye camera networks

Abstract

This paper presents a particle filtering algorithm for multiple pedestrian objects tracking across fisheye cameras network. For this purpose, we first suggest a particle filter tracking algorithm that integrates a best view selection strategy to ensure tracking consistency across multiple cameras. The proposed best view selection strategy, based on a model describing the known static occluders of the scene, favors the view in which the static occlusion suffered by a partially occluded pedestrian object is the least severe. We propose to estimate the severity of a partial occlusion through a new metric, called visibility index. Secondly, to perform multi-pedestrian object tracking, we propose a framework in which multiple autonomous instances of the proposed particle filter are used, each filter tracking one specific pedestrian object and taking dynamic occlusions into account. The efficiency of the proposed distributed framework lies in the fact that it performs tracking in linear complexity in terms of the number of tracked objects.