Abstract:

In this paper a global localization method is proposed to be used in a topological navigation scheme. Such method characterizes, in a distinctive way, places visited by a robot. The method extracts a constellation of various types of local affine covariant features from a panoramic image and then computes a local descriptor for each detected region. This constellation is then added to a map represented as a graph where nodes are places represented by local feature panoramas and edges are adjacency relations between known places. Then the robot finds its localization in this map by acquiring a new local feature constellation and comparing it to those stored in the graph.

• Tagged
• XML
• BibTex