Fast Spatially-Regularized Correlation Filters For Visual Object Tracking

Spatially-regularized correlation filters have achieved great successes in visual object tracking, with excellent tracking accuracy and robustness to various interferences. The performance improvement mainly attributes to spatial regularization (SR), which is a powerful tool to alleviate the boundary effects of correlation filters (CF) based tracking, but on the other hand, also severely harms the efficiency. In this paper, we propose an effective fast spatial regularization model that can be learned within the joint frequency and spatial domain. Extensive experiments on OTB-100 validate the effectiveness and generality of our model in helping state-of-the-art CF trackers to achieve much faster (near 5 times) frame rate and even better tracking accuracy.