Efficient 3D Reconstruction for Urban Scenes

Recently, researchers working in the fields of computer graphics and computer vision have shown tremendous interests in reconstructing urban scenes. For this task, the acquisition of the 3D point clouds is the first step, for which scans are usually widely utilized. Nevertheless, on seeing the potential drawbacks of scans, in this paper, we propose a novel urban scene reconstruction system based on the Multi-View Stereo (MVS). Given a set of calibrated photographs, we first generate point clouds using an existing MVS algorithm, and then reconstruct the sub-structures that often regularly repeat in urban buildings. Finally, we recover the entire architectural models through an automatic growing algorithm of the sub-structures in dominant directions. Experimental results on regular urban buildings show the practicality and high efficiency of the proposed reconstruction method.