Modeling Façade Structures Using Point Clouds From Dense Image Matching

Abstract

The automated generation of 3D computer models for urban planning and disaster management is an ongoing activity since about two decades. For 10-15 years aerial LiDAR seemed to be a key technology to solve the data collection problem for this purpose. Today, aerial photographs with nadir and oblique viewing directions overcome many obstacles and can be processed by dense image matching algorithms do deliver very dense point clouds outnumbering LiDAR point clouds. For some applications terrestrial images seem to be a low-cost data collection method delivering as well very dense point clouds for further processing. This paper is focused on terrestrial image data collection and its processing using dense image matching algorithms. To solve the point cloud modeling problem a formal grammar processing pipeline delivers structured building information of Level-of-Detail 3 (LoD3). This information might contribute to refine existing FEM software of Civil Engineering to consider critical building parts, to be monitored during some underground activities. Furthermore, the LoD3 model will help to increase building comfort in cities by adavanced insulation and noise filter measures.