Global error reduction in vision-based self-localization using a topological graph representation
Abstract
A single-sensor self-localization system which uses a monocular camera and a set of artificial landmarks is presented herein. The system represents the surrounding environment as a topological map (or graph) where each node corresponds to a marker (i.e., artificial landmark) and each edge corresponds to the existence of a relative pose between two markers. The edges are weighted based on an error metric (related to pose uncertainty) and a shortest path algorithm is applied to the map to compute the path corresponding to the least aggregate error. This path is used to localize the camera with respect to a global coordinate system whose origin lies on an arbitrary reference marker (i.e., the destination node of the path). Experimental results demonstrate the performance of the system in reducing the global error associated with large-scale localization.