Automated 3D reconstruction of buildings out of point clouds obtained from panoramic images

Report (2014)

Authors

J Chen External organisation

R. Sileryte Environmental Technology and Design - Architecture and the Built Environment

Kai Zhou External organisation

P. Nourian Design Informatics - Architecture and the Built Environment

S. Zlatanova OLD Department of GIS Technology

Research Group

Design Informatics (Architecture and the Built Environment) (TU Delft)

Point cloud 3D reconstruction Segmentation Panoramic photographs Polyhedral model

To reference this document use:

http://resolver.tudelft.nl/uuid:ca2c4e9e-6f09-448c-8e0d-feb0c6547f6b

More Info

expand_more

Published Date

2014

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Architecture and the Built Environment

Department

Architectural Engineering +Technology

Research Group

Design Informatics

Abstract

CycloMedia is a Dutch company specialized in the large-scale and systematic visualization of environments based on 360° panoramic photographs. The images are captured from public roads at street level, using a car to carry a panoramic camera system. High resolution spherical panoramas are geometrically correct and thus allow a user to perform photogrammetric 3D measurements and use imagery for 3D modelling and texture mapping. Drawing 3D point cloud (or rather pixel cloud) is a relatively new approach to use panoramic images. Currently there is no completely steady and consummate method to reconstruct digital building models from point clouds, thus this project aims to propose an innovative and practical framework of building reconstruction in particular. The developed method uses segmentation by region growing based on normals and curvature, constructing topology between segments by Voronoi partitioning, shape simplification by segmented least squares and principal component analysis methods. The method proves to have good results while tested on Lidar point clouds, however with some modifications is also applicable to pixel clouds obtained from panoramic imagery.