3D Structure from 2D Dimensional Images Using Structure from Motion Algorithms

Abstract

Natural disasters and human interference have endangered heritage structures around the world. Therefore, 3D modeling of buildings is important for historical preservation, particularly in low-income and war-affected countries. The majority of 3D structure surveying acquisition approaches, terrestrial laser scanning (TLS), total station measurements, or traditional photogrammetry require either high-cost technologies or professional user supervision. Structure from motion (SfM) approaches address both of these issues by allowing a non-expert user to produce a dense point cloud for real structures by taking a few 2D photographs with a digital camera and processing them with highly automated and freely available data processing tools. The state of the art for the SfM technique is presented in this paper. Agisoft Metashape, VisualSFM, and Regard3D, three well-known types of SfM software, were examined and compared. The 3D point cloud was scaled and transformed into a local coordinates system using total station instruments that were used to obtain some ground control points (GCPs). Ninety-six 2D digital photographs for the historical Emara Palace in Najran, Saudi Arabia, were obtained as data input, and the image matching, bundle adjustment (BA), completeness, and accuracy of three used packages were calculated and compared.