Human-Following of Mobile Robots Based on Object Tracking and Depth Vision

Abstract

Human-following is a promising ability for robots to serve and assist humankind. Correlation Filter (CF) with excellent performance and high running speed, is widely used in single object tracking, suitable for human following. In this paper, an improved correlation filter object tracking algorithm is adopted to achieve human following. For the lack of accuracy evaluation mechanism in most object trackers, fusion of Peak Side-lobe Rate (PSR) and depth information are introduced to evaluate the credibility of the tracking result. The sample and filter model are updated with a high credibility to avoid model drift. Face matching is used to re-track human when tracking fails. Based on a depth camera Kinect and a mobile robot, the experiments of following human show that proposed method has excellent abilities to deal with different scenarios and to follow target human stably for a long time.