Fast range image segmentation for servicing robots

Abstract

Presents a fast and scalable range image segmentation algorithm applied to a servicing robot navigation (obstacle detection) and grasping (object recognition). Our algorithm rapidly partitions range images into planar patches based on 3D scan line portioning. The key idea is that in a 3D scan line, all the points belonging to a planar surface form a straight-line segment. Secondly, any straight lines having similar parametric descriptions surely belong to the same planar surface, and thus can be merged. We also emphasize the paradigm of speed (navigation and obstacle avoidance) versus accuracy (object recognition and object grasping). Experimental results on real range images issued from a custom real time range sensor are presented.