Layer-Based City Point Cloud Completion for Aerial Multi-View Reconstruction

EasyChair Preprint 8607

Abstract

3D reconstruction of large-scale scenes using drones is a widely used method in 3D urban modeling. The specific process is to acquire aerial multi-view images of a target city using a UAV (Unmanned Aerial Vehicle) and reconstruct a 3D city point cloud model by matching feature points. However, aerial multi-view images acquired in this way require sufficient overlap between adjacent images. Therefore, large-scale urban modeling is very time-consuming and labor-intensive because the UAV must be flown to capture as much scene information as possible to reduce the effects of occlusion and other factors. In this paper, we propose a hierarchical point cloud completion method that can extract the geometry of urban buildings using only a small number of aerial orthomosaic views and complement scenes where occlusions such as walls exist. The completion results are verified in accuracy and completeness by a ground truth urban model acquired in virtual reality space.

Keyphrases: 3D reconstruction, Multi-View Image Processing, point cloud completion

@booklet{EasyChair:8607,
  author    = {Haihan Zhang and Hisatoshi Toriya and Hidehiko Shishido and Itaru Kitahara},
  title     = {Layer-Based City Point Cloud Completion for Aerial Multi-View Reconstruction},
  howpublished = {EasyChair Preprint 8607},
  year      = {EasyChair, 2022}}