Abstract: We propose a novel tracking algorithm for targets with drastically changing geometric appearance over time. To track such objects we use online update of appearance and shape. The problem here is to model foreground appearance of target with histograms in such a way that is accurate and efficient. In this paper, the constantly changing target foreground shape is modelled as a small number of patches or rectangular blocks, whose position within the tracing window are adaptively determined. We show that the robust tracking is possible by adaptively adjusting these patches. Moreover, object segmentation result is integrated into the proposed model to further enhance it. Experimental results show that the algorithm is able to track non-rigid objects undergoing large variation is appearance and shape.

Keywords: Non-rigid Object, Object Tracking, Histograms, Segmentation.