Restoring surfaces after removing objects in indoor 3D point clouds

Abstract

In this paper, we present fast approaches for object segmentation and surface restoration of indoor 3D point clouds, which are the results of 3D reconstruction methods or range scanners. These two problems are significant in constructing a augmented reality system using a range camera to build a virtual environment and provide the interaction mechanisms to the virtual model. For point-cloud segmentation, we apply a density-based clustering algorithm to extract the desired object after removing its ground planes. This low-complexity method gives stable results with high accuracy. After the segmented object has been removed, a restoration algorithm is proposed in such a case that the holes on the ground plane are revealed by removed objects. These holes are there because the corresponding surfaces are hidden by the segmented objects in the scanner phase. The process of filling the holes includes an object-ground-plane detection, a geometric restoration and a color fusion step. The newly added points are directly interpolated from the existing object points, which cover the holes in the original point clouds. Our approaches are experimented through a variety of test datasets and yield promising results.