3D Plane-Based Maps Simplification for RGB-D SLAM Systems
Abstract
RGB-D sensors offer new prospects to significantly develop robotic navigation and interaction capabilities. For applications requiring a high level of precision such as Simultaneous Localization and Mapping (SLAM), using the observed geometry can be a good solution to better constrain the problem and help improve indoor 3D reconstruction. This paper describes an RGB-D SLAM system benefiting from planes segmentation to generate lightweight 3D plane-based maps. Our aim here is to produce global 3D maps composed only by 3D planes unlike existing representations with millions of 3D points. Besides real-time trajectory estimation, the proposed method segments each input depth image into several planes, and then merges the obtained planes into a 3D plane-based reconstruction. This allows to avoid the high cost 3D point-based maps as RGB-D data contain a large number of points and significant redundant information. Our algorithm guarantees a geometric representation of the environment so these kinds of maps can be useful for indoor robot navigation as well as augmented reality applications.