Using consistency of depth gradient to improve visual tracking in RGB-D sequences
Huizhang Shi, Changxin Gao, Nong Sang · 2015
Color information is not quite effective for tracking in complex scenes, such as dynamic illumination and similar color background. Depth information has been one of the good resolutions to this problem, with the widely usage of RGB-D sensors. To effectively utilize depth information in visual tracking, this paper proposes a tracking method, where the motion model is extracted by integrating depth information and color information. To extract motion model from depth images, depth gradient information is used due to its robustness to complex scenes. Promising experimental results demonstrate the good performance of proposed method, especially in complex scenes.