Understanding Tracking Methodology of Kernelized Correlation Filter

Visual tracking as a field of research has undergone tremendous progress in the past decade. Researchers around the world have presented state-of-art trackers which work in presence of occlusions, clutter, variations in illumination and many others. Despite the significant progress the challenge continues in presenting real-time trackers which are computationally efficient and accurate. Kernelized Correlation Filter (KCF) is one of the recent finding which has shown good results. Based on the idea of traditional correlational filter, it uses kernel trick and circulant matrices to significantly improve the computation speed. Given the complexity of this tracker, a clear step-by-step explanation is highly desirable in order to fully appreciate and expedite the research in real-time visual tracking. This paper aims to make the understanding of this tracker simpler for the benefit of the research community