Tracking Multiple Non-Rigid Objects Using Backward Region Classification

Abstract

This paper proposes a method for tracking multiple non-rigid objects, based on Backward Region Classification. Non-rigid objects are composed of sub-regions with individual motion. Taking this into consideration, we first find for each sub-region the object which it comes from (Backward Region Classification). Grouping the sub-regions belonging to the same objects gives the current object regions. Previous methods based on Backward Region Classification, where the major goal is to track a single non-rigid object, classify each sub-region either to the object or to the background. The proposed method, in contrast, is capable of tracking multiple non-rigid objects.