A human-following approach using binocular camera

Abstract

This paper presents a tracker using binocular camera capable of tracking a specific human in both indoor and outdoor environments. Based on the kernelized correlation filter (KCF), the proposed tracker detects the target in multiple scales with the integration of HoG, color naming (CN), and local depth patterns (LDP) features, which improve the tracking performance. Specifically, a precise positioning method termed depth analysis is introduced to effectively overcome the problem of template drift. In addition, a motion controller is implemented to guide the motion of the mobile robot. The effectiveness of the proposed tracker is verified through video experiments and the robotic experiment.