Robust tracking with discriminative ranking lists

In this paper, we propose to model the target object by using discriminative ranking lists of object patches under different scales. The ranking list of each object patch is its K nearest neighbors. The object patches of the same scale with ranking lists of high purities (means with high probabilities to be on the target object) constitute the object model under that scale. A pair of object models of two different scales collaborate to determine which patches may be from the target object in the next frame. The superior ability to alleviate the model drift problem over several state-of-the-art tracking approaches is demonstrated quantitatively through extensive experiments.