Visual Tracking using Spatial-Temporal Regularized Support Correlation Filters

Support Correlation Filters (SCFs) have recently shown great potentials in real-time visual tracking. However, most of existing SCF trackers learn appearance models using the information of current frame, and completely neglect inter-frame information. Besides, they still suffer from unwanted boundary effects. In this paper, we proposed a novel Spatial-Temporal Regularized Support Correlation Filter (STRSCF) model, which introduces the spatial weight and temporal regularization term into SCF model. In order to improve the tracking performances, we extend STRSCF to multi-dimensional feature space. In addition, an effective optimization algorithm is developed to solve our STRSCF model in closed form solution. The experimental results on OTB-13 demonstrate that the STRSCF tracker performs superiorly against several state-of-the-art trackers in terms of accuracy and speed.