Occlusion Robust Object Tracking with Modified Particle Filter Framework

Object tracking is a classical problem of computer vision and is ubiquitous in many applications. Multiple tracking frameworks have been proposed in the past, and still attracting many researchers due to its high applicability in various fields. Major challenges in object tracking are because of constraints like illumination, occlusions, changing background, etc. This work proposes a modified Particle Filtering framework that is robust to partial and complete occlusions. In achieving so, this work suggests the use of a forward prediction filter that is fused with the proposed framework. It works irrespective of the measurement model. Also, our proposed work proposes an Uncertainty Factor for every prediction that controls the amount of uncertainty during particle update and adjusts the search area accordingly. This Uncertainty Factor also acts as a measure of tracking performance. Extensive experiments prove the better performance of the proposed work in comparison with the existing ones in presence of occlusion.