Automated Matching of Segmented Point Clouds to As-built Plans

Abstract

Terrestrial laser scanning (TLS) is seeing an increase use for surveying and engineering applications. As such, there is much on-going research into automating the process for segmentation and feature extraction. This paper presents a simple method for segmenting the interior of a building and comparing it to as-built plans. The method is based on analysing the local point attributes such as curvature, surface normal direction and underlying geometric structure. Random sampling consensus (RANSAC), region growing and voting techniques are applied to identify the predominant salient surface feature to extract wall and vertical segments. This information is used to generate a 2D plan of the interior space. A distance weighted method then automatically locates the corresponding vertices between the different datasets to transform them into a common coordinate system.A traditional survey was performed alongside the 3D point cloudcapture to compare and validate the generated 2D plans and the comparison to the existingdrawings. The accuracy of such generated plans from 3D point clouds will be explored.