Fragment Based Tracking for Scale and Orientation Adaptation

In this work, we propose a simple yet highly effective algorithm for tracking a target through significant scale and orientation change. We divide the target into a number of fragments and tracking of the whole target is achieved by coordinated tracking of the individual fragments. We use the mean shift algorithm to move the individual fragments to the nearest minima, though any other method like integral histograms could also be used. In contrast to the other fragment based approaches, which fix the relative positions of fragments within the target, we permit the fragments to move freely within certain bounds. Furthermore, we use a constant velocity Kalman filter for two purposes. Firstly, Kalman filter achieves robust tracking because of usage of a motion model. Secondly, to maintain coherence amongst the fragments, we use a coupled state transition model for the Kalman filter. Using the proposed tracking algorithm, we have experimented on several videos consisting of several hundred frames length each and obtained excellent results.