3D model based map building

Abstract

Several works deal with 3D data in SLAM problem. Data come from a 3D laser sweeping unit or a stereo camera, both providing a huge amount of data. In this paper, we detail an efficient method to extract a 3D model from raw data. Each model includes planar patches that belong to planar surfaces on the real scene. Then, we use these models in an ICP-like method in order to compute the robot movement. Once we know the movement performed by the robot, we can build a 3D environment map. Using ICP with planes is not a trivial task. It needs some adaptation from the original ICP. Some promising results in 6DoF are shown for both indoor and outdoor environment and for different kinds of 3D sensors.