Scaled Context Region for Correlation Filter Tracking

Abstract

Robust target tracking is a challenging problem in visual object tracking. Most existing methods cannot find a balance between accuracy and speediness. In this paper, we follow discriminative scale space tracking and adopt scaled context region in correlation filter tracking instead of fixed region to increase the accuracy of tracking result. The scale of context region varies according to peak-sidelobe-ratio and size of the target. Meanwhile, the computational cost does not increase too much in order to retain high computational speed. Quantitatively and qualitatively experiments are conducted to demonstrate the robustness and real-time performance of our method.