Bridging the divide between Visual Place Recognition and SLAM

Abstract

This thesis explores using filtering and optimisation methods from the simultaneous localisation and mapping literature to solve the visual place recognition problem. The proposed methods address the problems of uncertainty quantification during localisation, topological map building and learning image representations for sequence alignment. The research contributions from this thesis advance visual place recognition techniques using image sequences and can be applicable for autonomous vehicles and mobile robots operating in changing environments.

Publication Details

Data Sources