Visual Tracking via Structure Rearrangement and Multi-scale Block Appearance Model

In this paper, we propose a tracking method via structure rearrangement and multi-scale block based appearance model, plus a dynamic template mechanism within a two-stage search framework. Firstly, a wide range sampling is performed then confidence value of each candidate is calculated via a discriminative reverse sparse coefficient vote method, a part of candidates with large confidence value are selected to join in next stage. After a small range resampling in stage two, the candidates and target templates are divided into multi-scale patches, in addition, the background information is used to model the error term. Furthermore, a labeled template pool is maintained in the tracking process to dynamically generate an appropriate template set for next frame according to the occlusion map of current tracking result. Both qualitative and quantitative evaluations on challenging image sequences demonstrate that the proposed tracking algorithm performs favorably against several state-of-the-art methods.