Stereo-Based Outdoor Localization using a Line Drawing Building Map

Line drawing maps are frequently used by people to exchange location information. Mobile robots in the future are expected to be able to communicate with people and thus to localize themselves using such maps. We therefore develop a method of outdoor localization using a stereo camera and a line drawing building map. Omnidirectional range data obtained by a stereo camera and a pan-tilt unit are projected onto a robotcentered 2D grid map, which is to be matched with an input line drawing map. Since various objects such as trees often obstruct buildings in usual outdoor scenes, we apply a view-based object classifier in order to extract only stereo data from buildings. To cope with uncertainties in stereo data and building extraction, we adopt Monte Carlo localization. Experimental results show that our simple approach works reasonably well in an actual outdoor environment.