Efficient Plane Detection in Multilevel Surface Maps

An automatic system aimed at producing a compact tridimensional description of indoor environments using a mobile 3D laser scanner is described in this paper. The resulting description is made up of a Multi-Level Map (ML map) and a series of plane patches extracted from the MLSM. We propose a novel plane detection algorithm, based on the efficient RANSAC algorithm, that operates directly over the data structures of a ML map and does not need to rely on the low level laser data cloud. The mobile 3D scanner is built from a Hokuyo laser range sensor attached to a 2DOF pan-tilt, which is installed on top of a 3DX Pioneer mobile robot. The 3D spatial information acquired by the laser sensor from different poses is used to build a large single map of the environment using the SLAM 6D library. Experimental results demonstrate that the described system is capable of efficiently building compact and accurate 3D representations of complex large indoor environments at multiple semantic levels.