The article discusses a new framework for Robot Person Following (RPF) that leverages long-term experiences for person re-identification (ReID). In many Human-Robot Interaction (HRI) applications, it’s essential for the robot to persistently follow a designated person. However, the person may often be occluded by other objects or people, making it necessary to re-identify the person when they re-appear within the robot’s field of view. The proposed ReID framework addresses this by maintaining experiences through a loss-guided keyframe selection strategy, enabling online continual learning of the person’s appearance model. The proposed method has shown promising results in re-identifying the person accurately even in the presence of severe appearance changes and distractions.
Publication date: 22 Sep 2023
Project Page: https://sites.google.com/view/oclrpf
Paper: https://arxiv.org/pdf/2309.11727
