Comparative Study of Surface 3D Reconstruction Methods Applied in Construction Sites

The accurate and detailed 3D reconstruction of the construction sites plays a vital role in the digitalization of the construction domain, since accurate 3D models constitute the basis for the adaptation of advanced technologies from Industry 4.0 towards realizing Construction 4.0. This study provides a comprehensive assessment of key methodologies employed for 3D reconstruction in the construction sector. Our primary objective is to evaluate the performance of both monocular and binocular computer vision implementations in terms of their ability to accurately extract detailed 3D surfaces, while also considering their computational efficiency. Specifically, the monocular approaches of Neural Radiance Fields (NeRF), Instant Neural Graphic Primitives (Instant NGP) - NeRF, and Patch-based Multi-View Stereo (PMVS), along with the binocular methods of Kiss-ICP, Voxblox, and ICP - Poisson Surface Reconstruction (PSR), have been comprehensively assessed. The evaluation was carried out using a custom-made dataset acquired from challenging both indoor and outdoor construction areas. The findings of this study contribute to the advancement of 3D reconstruction techniques in the construction industry by providing valuable insights into the inherent strengths and limitations of these specific methods and pave the way for the adoption of adequate imaging systems and methods in the digitalization of construction sites.