Scale adaptive ensemble tracking
Guanbin Li, Hefeng Wu · 2011
Visual tracking is treated as a binary classification problem in recent trend. In this paper, we propose a novel approach to preprocess pixel-based examples used for training and online updating of classifiers, resulting in a label map that plays a guidance role, which can greatly alleviate the problem of model degradation and at the same time offer a convenient way to scale adaptation of model templates during the tracking process. An integrated tracking system is built through fusing our method into the ensemble tracking framework. Experiments on challenging video sequences demonstrate the effectiveness of the proposed approach.