Learning Dual-Fused Modality-Aware Representations for RGBD Tracking

Object tracking is to localize an arbitrary object in a video sequence, given only the object description in the first frame. It can be applied in lots of applications in video surveillance, autonomous driving [18,35,23], and robotics [19]. Recent years witness the development of RGBD (RGB+Depth) object tracking thanks to the affordable and accurate depth cameras. RGBD tracking aims to track the objects more robustly and accurately with the help of depth information, even in color-failed scenarios, e.g., target occlusion and dark scenes. Compared to conventional RGB-based tracking, the major difficulty of RGBD S. Gao et al.