Improving the Correlation Filter-based Tracking Using Low-illumination Enhancement Method

Abstract

Correlation filter-based trackers are commonly used nowadays and has shown good results in both precision rate and frame per second rate, and have been developed from working on raw pixels, to features of the image, ending on applying them on layers of convolutional neural networks. However, those trackers have not shown good results dealing with complex object appearances such as blur caused by fast motion, deformation, and difficult illumination cases. In this paper, we present a method that treats tracking objects in low illumination case. It makes a low illumination enhancement to each dark image before each treatment of a sequence using two different enhancers: Light Image Enhancement via Illumination Map Estimation (LIME) and Contrast Enhancement (CE) applied on two different trackers: tracking with Kernelized Correlation Filters (KCF) and Hierarchical Convolutional Features Tracker (HCFT).