Real-Time Experimental Study of Kernelized Correlation Filter Tracker using RGB Kinect Camera

Over years correlation filter-based trackers have proved their worth with their increased efficiency and increased computation speed. Kernelized Correlation Filter (KCF) was one such attempt which, by using kernel trick, achieved compelling result as compared to traditional correlation filter-based trackers. In this paper, our goal is to analyze this tracker to observe its strengths and weaknesses in detail. We use Kinect RGB camera for our experimental analysis and report our findings. The analysis showed that KCF is not only computationally very fast, it is time-invariant and very robust to speed and vertical motions. However, it is not very robust to illumination variations, scale and color.