Mobile robot self-localization by iconic matching of range maps

This paper examines techniques for a mobile robot to perform self-localization an natural terrain using stereo vision. The range map computed from a stereo image pair is first processed to generate a three-dimensional occupancy map of the terrain. This occupancy map is then compared to a similar map in a known frame of reference, The optimal relative position between the maps with respect to a Hausdorff measure is determined using efficient search techniques. These techniques allow the localization of a mobile robot to be performed robustly in natural terrain, even in the presence of noise, scene clutter, and missing data.