di
(PA).
k 09,
:349.
Week
1000 e
. Boll.
IFET,
ito. In:
) 1996,
1 area.
Thomas Vógtle
3D MODELLING OF BUILDINGS USING LASER SCANNING AND SPECTRAL INFORMATION
Thomas VOGTLE, Eberhard STEINLE
University of Karlsruhe, Germany
Institute of Photogrammetry and Remote Sensing
voegtle @ipf.uni-karlsruhe.de
steinle @ipf.uni-karlsruhe.de
KEY WORDS: Building Reconstruction, Urban Objects, Laser Scanner, Data Fusion
ABSTRACT
In this paper an approach for 3D building modelling based on laser scanning data combined with spectral information is
presented. Airborne laser scanners offer a new efficient method for 3D data acquisition (DEMs) in urban environment.
However, for automatic building recognition and reconstruction, other 3D objects like vegetation or vehicles have to be
excluded from the acquired DEM. Therefore, a strategy for stepwise elimination of areas and objects which are not
buildings will be explained. These procedures are based on spectral as well as shape and size parameters. After a
separation of the remaining 3D objects as building hypotheses, a generic approach for modelling of building shapes by
planes will be described. As results of an intersection process wireframe models are obtained. Finally, first experiences
with/on this approach in a test area are presented and the achieved accuracy is discussed.
1 INTRODUCTION
In the recent past, an increasing demand of 3D city models for various purposes has been observed, e.g. for planning,
administration, facility management or visualization. At first, traditional methods like photogrammetric image analysis
by an operator had been applied for this which was a very extensive and time consuming task. Therefore, semi-
automatic and automatic approaches based on digital image processing were investigated, e.g. (Griin & Baltsavias,
1997), (Ebner et al., 1999) and (Weindorf & Vogtle, 1998). However, the results from these methods could not totally
satisfy the requirements because they had to cope with a lot of difficulties during extraction of 3D objects from stereo or
multiple images. First of all, lots of occlusions caused by high buildings or vegetation occur outside the principle point
of the images (especially in urban areas) which disturb or prevent stereoscopic analysis. Also irrelevant details of
buildings like dormer windows, balconies or chimneys disturb automatic edge detection and matching in aerial images.
Furthermore, in most cases no 3D information can be determined in shadow areas or other unstructured parts of the
image (e.g. flat roofs).
One of the latest developments in sensor technology, airborne laser scanning, offers a new efficient data acquisition
method for measuring urban objects directly in three dimensions and storing the results digitally which shortens
postprocessing time enormously. In most cases, the determination of XYZ co-ordinates is based on pulsed laser
scanning, i.e. the measurement of the runtime of a laser pulse which is reflected on the surface of the Earth or objects on
it (slant distance). By measuring the exterior orientation of the sensor during flight (by means of differential GPS and
INS systems) the resulting 3D co-ordinates of the object points can be calculated (DEM). These sensors provide a high
point density (commonly 0.5 to 10 points per square meter) and a suitable point accuracy (up to +/- 0.10m in elevation,
+/- 0.20m in position) (Baltsavias, 1999). Additional advantages are low dependency on weather and light conditions,
few occlusions and point measurements in unstructured areas.
2 OBJECTIVES
In the following an approach for 3D modelling of buildings will be presented combining height data of airborne laser
scanning with spectral information. As experiences have shown automatic building detection cannot be done with high
reliability when based only on elevation data. Therefore, additional spectral information has to be integrated in data
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 927