Towards automated quality assessment of construction elements

Abstract

Construction progress monitoring has gained increasing interest in the recent decade due to the implementation of Building Information Modeling and affordable and efficient Reality Capture technologies. The latter include Laser scanning (Bosché and Haas, 2008) as well as photogrammetric methods (Golparvar-Fard et al., 2009). Scan-vs-BIM methods allow an as-planned vs. as-built comparison to make inferences on the presence of individual construction elements. With the incorporation of 4D data, statements on the construction progress are possible (Turkan et al., 2012). However, point clouds do not always provide sufficient or adequate information for quality assessment. Thus, recent research has been focussing on image-based methods and deep learning to solve this problem. For example, several researchers effectively detect cracks in asphalt or concrete elements using images instead of 3D point clouds (NhatDuc, Nguyen and Tran, 2018). The authors propose to incorporate photogrammetry-based Scan-vs-BIM workflows with image-based processing enhancements to make detailed inferences on construction quality as well as providing continuous and semanticallyclassified image data for QA personnel.