Urban scene 3D reconstruction optimization leveraged by line information

Abstract

Recently, great progress has been made in large scale scene reconstruction based on structure from motion and multi-view stereo approaches. However, the accuracy of this kind model still needs to be improved in the urban scene reconstruction. In this study, a novel method was proposed to optimize the 3D reconstruction result from multi-view images with 3D line information. The joint estimation of the line and dense points information are used to refine all the existing edges in scenes. The dense points and 3D line information were firstly reconstructed from the structure and motion result. Then, the wrong lines were removed based on the dense points and 3D line information. The incorrect points are removed based on the structural line segments and the mesh curvature. Finally, all the point position information are fused with normal information. The strategy is illustrated on several real sets of images of urban buildings and the experiments demonstrate the performance of this method is convincing and the reconstruction results is definitely better than the original ones.