Depth-Based Filtration for Tracking Boost

This paper presents a novel depth information utilization method for performance boosting of tracking in traditional RGB trackers for arbitrary objects (objects not known in advance) by object segmentation/separation supported by depth information. The main focus is on real-time applications, such as robotics or surveillance, where exploitation of depth sensors, that are nowadays affordable, is not only possible but also feasible. The aim is to show that the depth information used for target segmentation significantly helps reducing incorrect model updates caused by occlusion or drifts and improves success rate and precision of traditional RGB tracker while keeping comparably efficient and thus possibly real-time. The paper also presents and discusses the achieved performance results.