Outdoor localization using wide-angle visual feature matching and image retrieval

Abstract

This paper describes an outdoor mobile robot navigation and localization system based on visual features. We use a novel feature detector operating on wide-angle images and create a database of image features for a large outdoor environment. Images are geo-tagged by their pose. Given an arbitrary view we use algorithms inspired by "Video Google" to extract scenes from the database that overlap the current view and from which we can estimate our current pose in the environment or detect loop closures. © 2010 Springer-Verlag Berlin Heidelberg.