Vision-based Topological Map Building and Localisation using Persistent Features

This paper proposes a topological mapping technique that utilises persistent SIFT features to reduce the amount of data storage required. It delivers, as an output, a topological map that lends itself well to conventional path planning techniques. The approach assimilates features into a statistical model which promises improved data association. Experiments were performed using omnidirectional camera images from the Cogniron dataset. A map was constructed from one of the supplied routes and the performance of the localisation algorithm evaluated using another route within the same environment. Thereafter the map was updated using the comparison route and the results discussed. The statistical feature association approach is shown to be more robust than conventional methods.