Detection of Surfaces and Edges in Large Point-Clouds Using Region-Growing on Spherical Space

Abstract

Engineering facilities can be digitized as large-scale point-clouds by using the state-of-art mid-range laser scanners. For utilizing captured data in CAD systems, it is important to convert point-clouds to parametric surfaces. In this paper, we describe a method for robustly extract cylindrical faces and planar faces. Edges and silhouette lines have to be generated to construct bounded faces, but unfortunately points on silhouettes are very noisy in the case of mid-range laser scanners. Our method applies region-growing method on spherical space and improves the robustness. In addition, we enhance the region-growing so that surface regions can be propagated to disconnect points using multiple overlapping point-clouds.