Full occlusion handling for pedestrian tracking via hybrid system

Abstract

Occlusion and lack of visibility even in sparse crowd scenes make it difficult to track individual pedestrianscorrectly and consistently, particularly in a single view. We present a novel pedestrian tracking approach that connectstracking with reidenti cation to locate and maintain the identity of certain people who may be occluded for a longtime. First, two models are constructed. One model tracks the pedestrian and trains a classi er, while the other modelreidenti es the pedestrian of interest from detection results with the trained classi er. Secondly, we design a set oftransition rules for model switching. Finally, the two models work alternatively based on the principle of a hybridsystem to track the pedestrian. Several typical sets of experiments show that the proposed approach outperforms thestate-of-the-art approaches and achieves robust pedestrian tracking in the presence of full occlusion.