Planar Features for Accurate Laser-Based 3-D SLAM in Urban Environments

Simultaneous Localization and Mapping (SLAM) systems using 3-D laser data typically represent the map as an unstructured point cloud, which is inefficient in data association and does not allow one to use the map for reasoning about the observed scene. In this paper we describe a laser-based SLAM system that represents the map as a collection of 3-D planar and line segments, which provide a natural way of representing man-made environments. We demonstrate that this representation improves the accuracy of trajectory estimation and makes it possible to represent major objects as geometric shapes.