Optimum Correlation filters for Visual Tracking via Histogram of Gradient Features and SVM classifie

Abstract

The majority of object trackers algorithms cannot recover tracking processes from problems of drifting. These problems are caused by several challenges, especially heavy occlusion, scale variation, and fast motion. In this paper, we present a new effective method with the aim of treating these challenges robustly basing on two principal tasks. First, we infer the target location using the correlation map, resulting from the combination of a learned correlation filter model with a histogram of gradient (HOG) features. Indeed, Bat algorithm (BA) is exploited for solving the update model equation of the correlation filters. Second, we use the histogram of gradient features to learn another correlation filter model in order to estimate the scale variation. Furthermore, we exploit an online training SVM classifier to re-detect target in failure cases. The extensive experiments on a commonly used tracking benchmark dataset justify that our tracker significantly outperforms the state-of-the-art trackers.