Learning Place-Dependent Feature Detectors for Localisation Across Extreme Lighting and Weather Conditions

This work is about metric localisation across extreme lighting and weather conditions. The typical approach in robot vision is to use a point-feature-based system for localisation tasks. However, these system typically fail when appearance changes are too drastic. This research takes a contrary view and asks what is possible if instead we learn a bespoke detector for every place. Our localisation task then turns into curating a large bank of spatially indexed detectors and we show that this yields vastly superior performance in terms of robustness in exchange for a reduced but tolerable metric precision. We present an unsupervised system that produces broad-region detectors for distinctive visual elements, called scene signatures, which can be associated across almost all appearance changes.