Person Re-identification based on Interactive Transfer Deep Learning for Mobile Robots and Its Application to TA Task Support

Abstract

This paper proposes interactive person re-identification method for mobile robots that periodically provide services to specific groups. This method consists of a CNN-based person feature extractor that is trained based on Triplet Loss, and a CNN-based person re-identifier that is trained based on transfer learning. Person re-ID is executed through a cooperative human-in-the-loop learning approach. As an example of a service, we apply this method to a Teaching Assistant (TA) support. This application aims to support students’ study based on their identification by the proposed method and student card reading in which appearances and names are linked. Performance of the proposed method is evaluated by experiments using a large open dataset and a self-made dataset periodically collected for the same group by a mobile robot. In addition, the feasibility of the TA support is verified by experiments in which robots are operated in actual classes.