Precise reconstruction of geometric primitives in built environments

Lingling Wang, Hanbin Luo, Ying Zhou, Cheng Zhou
School of Civil Engineering and Mechanics, Huazhong Univ. of Science and Technology, China, People's Republic of

DOI: 10.35490/EC3.2019.165
Abstract: Precise reconstruction of the built environment is very useful for the management of the construction site. As far as the reconstruction of large-scale built environments is concerned, the reconstruction effect still needs to be further improved. Considering that most of the structures are piece-wise planar/linear in the built environment, this paper proposes a method for reconstructing the geometric structure of the scene that display its appearance precisely. This method focuses on reconstructing objects with plane and edge structures, such as buildings, to achieve the reproduction of their geometry. The paper introduces a new dense reconstruction algorithm, the patch based stereo matching algorithm to refine a sparse point cloud to produce a dense point cloud. This method further merges three-dimensional (3D) line into the dense point cloud to optimize the geometric line of the model. The experiment demonstrates that the improved method has a flawless reconstruction effect on the geometric primitives of buildings.
Keywords: 3D Reconstruction, Geometric Primitives, Built Environments, Depth-Map Propagation
Pages: 85 - 91
Paper:
http://ec-3.org/conf2019/contribution_165_final/

Presentation Video: