Abstract

3D laser scanners provide accurate as-built conditions for the surrounding environment in the form of 3D point cloud data. Although this technology has had high attention from the construction industry for the as-built documentation of buildings, the reconstruction process, especially identification and segmentation of the building elements, still has manual and labor-intensive tasks leading to time-consuming and human errors. In addition, it has not reconstructed the building elements successfully yet in multi-level building spaces. In an effort to address these issues, this research proposes an automatic 3D reconstruction framework that identifies, segments, and reconstructs vertical and horizontal building elements from the point clouds of multi-level building spaces. The proposed framework composes of: (1) identifying locations, diameters, lengths and the number of vertical building elements using Hough line and circle transform; (2) comparing the dimensions of the walls to determine single- or multi-level building spaces; (3) developing the region of interest defined by the building codes; (4) implementing plane RANSAC for not only segmentation of the vertical building elements but also identification and segmentation of horizontal building elements; and (5) reconstructing the segmented building elements into simple forms. The effectiveness of the proposed methodology has been validated with high accuracy and low deviation in three different building spaces at Concordia University, Montreal, Canada.