Extracting structural components of concrete buildings from laser scanning point clouds from construction sites. (January 2022)

• Record Type: Journal Article
• Authors: Truong-Hong, L.; Lindenbergh, Roderik
• Is Part Of: Advanced engineering informatics. Volume 51(2022)
• Publication Date: 2022-01
• Subjects: BIM Building Information Modelling -- 3D Three dimensional -- 2D Two dimensional -- HT Hough Transform -- RANSAC RANdom Sample Consensus -- CRG Cell-based region growing segmentation -- VRG Voxel-based region growing segmentation -- PPRG Patch-point region growing -- CSC Connected surface component -- SbF Surface-based filtering -- PCA Principal component analysis -- rPCA Robust principal component analysis -- KDE Kernel density estimation -- PDS Probability density shape -- mbb Minimum bounding box -- PM Proposed method -- GT Ground truth -- TP True possitive -- TN True negative -- FN False negative -- Comp. Completeness -- Corr. Correctness -- Qual. Quality; Point cloud -- Object extraction -- Structure extraction -- Segmentation -- Inspection -- Scan to BIM; Computer-aided engineering -- Periodicals; Engineering -- Data processing -- Periodicals; 620.00285
• DOI: 10.1016/j.aei.2021.101490
• Languages: English
• ISSNs: 1474-0346

Abstract

Abstract: In construction projects, inspection of structural components mostly relies on classical measurements obtained by measuring tapes, levelling, or total stations. With those methods, only a few points on the structure can be measured, and the resulting inspection may not fully reflect the actual, detailed condition of the complete object. Laser scanning is an emerging remote sensing technology to accurately and quickly capture surfaces of structures in high details. However, because of the complex, massive point cloud data acquired at a construction project, in practice, data processing is still manual work with computer aided programs. To improve upon current workflows, this paper proposes a method to automatically extract point clouds of individual surfaces of structural components of a concrete building, which subsequently can be used to inspect construction quality based on geometric information of the surfaces. The proposed method explores both spatial point cloud information and contextual knowledge of structures (e.g., orientation or shape) derived from building design specifications and practice. For extracting point clouds of surfaces of each structural component, the proposed method consists of 4 consecutive steps for extracting: (1) floors, ceiling slabs, and walls, (2) columns, and (3) primary and (4) secondary beams. Each step consists of two ingredients: (i) rough extracting the candidate points of the component and (ii) fine filtering of the surface points of the components via cell-based and voxel-based region growing segmentation (CRG and VRG) incorporating contextual knowledge of the structural members. Experimental tests on two different types of concrete buildings showed that the proposed method successfully extracts the structural elements, in which the completeness, correctness, and quality from the point-based evaluation are larger than 96.0%, 96.9%, and 92.0%, respectively. Moreover, the evaluation based on a shape similarity showed that the extracted floor, ceiling slab and wall overlap to the ground truth more than 92.5%.