Performance Evaluation of KCF based Trackers using VOT Dataset

Abstract

We compare the performance of sixteen variants of the Kernelised Correlation Filter (KCF) tracker using the dataset and metrics introduced through the Visual Object Tracking (VOT) Challenge. An introduction to the KCF tracking framework is given, along with brief discussion on recent variants of the KCF tracker. Their performance is compared using the metrics of speed, accuracy, robustness and expected average overlap.