Void-growing: a novel Scan-to-BIM method for manhattan world buildings from point cloud

Abstract

The automated generation of 3D models of buildings from point clouds is still under heavy research. Currently, this Scan-to-BIM process requires high manual effort, and the previous research focuses on simple room structure under low occlusion level. We propose a novel “void-growing” approach that extracts walls, floors, and ceilings automatically. Different from the majority of current approaches starting with detecting surfaces of elements in point clouds, our approach grows the void volume space inside a room first. Our approach performs well in occluded environments. It can reconstruct simple cuboid rooms as well as complex rooms like L-shape and U-shape rooms. Different ceiling heights caused by suspended ceilings can also be represented properly.