Reconstruction of Wall Surfaces Under Occlusion and Clutter in 3D Indoor Environments

Abstract

In this paper, we present a method for the reconstruction of interiors using a set of panoramic range data in scenes with clutter and occlusion. We specifically deal with the reconstruction of simply-shaped wide areas (such as walls, ceilings and floors) behind furniture and facility pieces in interiors. To date, little attention has been paid to this issue and only incomplete solutions in simple scenarios appear in literature. This document presents an integrated solution to this problem, ranging from the data collection to the restoration of missing 3D information. Our approach is based on a sequential updating labeling strategy in different data representation spaces. A volumetric representation is used to permit the labeling of the 3D space for different range data and the fusion of all the scene’s labels to obtain one single 2D labeling image for each of the simply-shaped wide areas. Based on this labeling process, our method is able both to identify occluded regions and, through an SVM learning technique, to recognize essential parts of the walls, such as doors and windows, so that labeling is continuously updated. Finally, the reconstruction of the wall is carried out in the last stage of the process by using an inpainting algorithm, which has been adapted to our particular application. The method was tested in real generic scenes under difficult clutter and occlusion conditions, and has yielded promising results.