Slam and Multi-feature Map by Fusing 3D Laser and Camera Data

Indoor structured environments contain an important number of planar surfaces and line segments. Using these both features in a unique map gives a simplified way to represent man-made environments. Extracting planes and lines by a mobile robot requires more than one sensor: a 3D laser scanner and a camera can be a good equipment. The incremental construction of such a model is a Simultaneous Localisation And Mapping (SLAM) problem: while exploring the environment, the robot executes motions; from each position, it acquires sensory data, extracts perceptual features, and simultaneously, performs self-localisation and model update. First, the 3D range image is segmented into a set of planar faces which are used as landmarks. Next, we describe how to extract 2D line landmarks by fusing data from both sensors. Our stochastic map is of heterogeneous type and contains plane and 2D line landmarks. At first, The SLAM formalism is used to build a stochastic planar map, and results on the incremental construction of such a map are presented, further on, heterogeneous map will be constructed.