Reliable Multiple Object Tracking under Heavy Occlusions

Abstract

Tracking multiple objects in surveillance scenarios involves considerable difficulty because of occlusions. We report a novel tracker -- based on reliability tracking -- that demonstrates superior performance under high degrees of occlusion. In our method, distinguishable features between the target and non-target are represented as the object’s reliability. When the selected features are no longer reliable for sake of occlusions, the proposed method should select a new feature with more reliability by its corresponding region’s status. We present results from PETS 2006 dataset with many objects in the scene at any instant. Experimental results show that our method is robust when tracking objects during partial and serious occlusions. The object’s discriminability in appearance model is well maintained when interaction among other objects occurs.