Improvement of a SIFT-based people identification for a people following robot

Use of a distance-dependent appearance model

Abstract

This paper describes person identification technique for a mobile robot which performs specific person following under dynamic complicated environment like a school canteen where many persons exist. Although the SIFT feature is used for identification of a person, the distance dependence appearance model which expects the number of collation of the SIFT feature based on the distance to a person is created and used for the improvement in reliability of identification. Moreover, improvement in the speed of identification processing was also considered. The person following experiment was actually conducted using the mobile robot, and the quality assessment of person identification was performed.