Robust Visual Tracking via Binocular Consistent Sparse Learning

Abstract

In spite of the rapid development of visual tracking technologies, robust object tracking in the monocular images under complex environments still remains a challenging problem. In contrast to its monocular counterpart, stereo vision features more images from another camera looking from different viewpoints and has the capability of generating depth information for scenes. In this paper, a novel Binocular Consistent Sparse learning based Tracker (BCST) is proposed. With the popular sparse learning framework, the new method greatly improves tracking performance via efficiently exploiting the appearance and depth information from the binocular configuration. Valuable prior appearance of tracking object obtained through the second camera is integrated into an augmented dictionary via the proposed crossover templates. The depth is integrated into the sparse learning framework in three aspects. First, an extra depth view is added to the color image-based visual features as an independent view. Then a special depth consistency constraint is designed in the objective function. At last most of the stray particles can be removed according to the depth consistency property with the assumption of small range variations of tracking object between frames. An effective ADMM based optimization algorithm to solve the proposed objective function is also given. Extensive experiments on KITTI Vision Benchmark show that the proposed BCST outperforms the state-of-the-art trackers, including both the sparse and stereo-based methods.