Semantic scan planning for indoor structural elements of buildings.

The objective of this paper is to propose a new semantic 3D data acquisition method which is focused on sensing data belonging to indoor structural elements of buildings. Our system uses and processes 3D information coming from a 3D laser scanner sensor. The presented approach deals with some essential key issues in the scanning world which are rarely dealt with in papers. These are: the final goal of the scanning process, the hypotheses about the scene, lack of dynamic spaces in the next-best-scan-based solutions and the quality evaluation of the data sensed. Whereas most of the Next Best Scan (NBS) based approaches do not discriminate between data and clutter, we propose a scanning process in which potential structural elements of building indoors are learned as a new scan arrives. Our workspace is not a priori hypothesized, but a dynamic space which is updated as a new scan is added. This allows us to deal with more complex shape scenarios (i.e. concave-shaped spaces). Through the so called Structural Element (SE) membership probability, we introduce the data-quality concept in the scanning process which highly reduces the point cloud to be processed. This system has been tested in inhabited indoors and has yielded promising results. An experimental comparison with three close techniques is presented in an extended and detailed experimental section. The results yielded from our experimental work demonstrate the quality and validity of the proposed method.