Automated 3D Building Reconstruction using ALS and Digital Photogrammetry

Automated 3D Building Reconstruction using ALS and Digital Photogrammetry


Sudhagar Nagarajan
Sudhagar Nagarajan
Stuttgart University of Applied Sciences, Germany

Dr. Eberhard Gülch, Dr. Michael Hahn
Stuttgart University of Applied Sciences, Germany

The extraction of 3D buildings from Airborne Laser Scanning (ALS) data and digital aerial images is an existing research field. There have been ample of researches performed and to derive cost effective and resource effective automatic building extraction methods using airborne laser scanning. Some of them have managed to bring favorable results with some limitations. The limitation could be either the involvement of a manual operator to help the algorithm (semi-automatic) or the requirement of other resources like multi-spectral images, building footprints, etc. Building extractions can be achieved either by model driven (top down approach) or data driven (bottom up approach).

In this research, a data driven approach is developed to derive 3D building information from ALS data. Complementary nature of ALS and digital photogrammetry has been exploited to derive 3D building models. For ALS data classification, simple geometric algorithms are used. Possible building planes are detected using uniform surface normal. Convex hull rectangle of each robustly fitted plane is extracted. This algorithm is limited only to the buildings made of rectangular planes. A detected building model is transferred to the image space using collinearity equations. Aerial images are used to adjust the outline of the buildings derived from ALS data. The edge search and linking algorithm is used to detect the outline of buildings. Matching the edges and the building outline is based on the Hough transform. For the experimental investigations the test site Hermanni from Finnish Geodetic Institute is used.

1. Introduction
3D building models are prominently used in 3D city models. The requirement for 3D city models is increasing over the last few years in many applications like urban planning, noise simulation, disaster management, tourism, defense and security, navigation, telecommunication, architecture and entertainment. The important issue on 3D city model creation is the unavailability of 3D building information. Some cities have only building footprints and many cities have no information about the buildings. Also, accurate 3D building models are essential for true orthophoto generation. There have been few options tried for 3D building extraction. It has been proven by many researchers that ALS, photogrammetry, remote sensing and tacheometric survey are the best options. Methods using photogrammetry alone, ALS alone and terrestrial surveying are time consuming. Automatic 3D building reconstruction is the aim of this research by utilizing the potentials of airborne laser scanning and digital photogrammetry.

Related research
Among the several researches accomplished for automatic building extraction, only some closely related researches are discussed below.

Elaksher and Bethel (2002), Morgan and Habib (2001), Rottensteiner and Briese (2002), Maas and Vosselman (1999) adopted different techniques to extract 3D buildings from ALS data.

Arefi and Hahn (2005), use morphological operations and surface properties to demarcate the building and vegetation areas if both first and last pulse data are available.

Kim and Nevatia (2004) present an approach for detecting and describing complex buildings with flat or complex roof tops by using multiple overlapping images.

Schenk and Csatho (2002), assert that ALS data and aerial images deliver complementary surface information and Kaartinen et al (2005), EuroSDR research group confirm it for building extraction. Sohn and Dowman (2004) demonstrated building extraction using Lidar and IKONOS images.

This research is aimed to take the advantages of both ALS data and aerial images to derive a complete and accurate 3D building model. ALS gives accurate height information and aerial images give accurate horizontal breakline position of building outlines. ALS data is classified into ground and non-ground points. Then building planes (points) are detected from the non-ground points. Building outline is derived from the detected ALS building points. Finally, the building outline is refined using aerial images and the 3D building model is reconstructed.