An Occlusion-Aware RGB-D Visual Object Tracking Method Based on Siamese Network

In recent years, Siamese trackers have been extensively studied because of their high accuracy and speed. However, when the target is occluded by other objects, the result will be greatly drifted, which affects the quality of the tracking results. This study based on the RGB-D data proposes an object tracking method integrating a target occlusion estimation module and a target location correction module, called Siamese-Occlusion-Correction (SiamOC). When a target is occluded, these modules can help the Siamese tracker correct the target location. In this paper, experiments demonstrate that the method which is the real-time tracker has achieved competitive results on the CDTB dataset.