hul 2004
varzkopf,
Applica-
/ models
'omputer
)3 23(6),
zenerator
VI. Sester
deler. In:
C Extrac-
ges (III),
of semi-
A. Grün,
fade Ob-
1blishers,
iutomatic
IAPRS,
n surface
E. Balt-
of Man-
rkhäuser,
>rojective
D thesis,
1stitut fiir
last ac-
ration of
of aerial
ein (eds),
construc-
en, Octo-
inar faces
onference
Imagery’,
p. 87-92.
model re-
n: M. A.
Land Sur-
etry, An-
mmission
ng recon-
| A. Grün,
Made Ob-
ublishers,
x traction.
Automatic
ce Images
1g Object-
n- Wesley.
AUTOMATIC MATCHING OF TERRESTRIAL SCAN DATA AS A BASIS FOR THE
GENERATION OF DETAILED 3D CITY MODELS
Christoph Dold! and Claus Brenner
Institute of Cartography and Geoinformatics, University of Hannover, Germany
{Christoph.Dold, Claus.Brenner} @ikg.uni-hannover.de
Working Group 111/6
KEY WORDS: LIDAR, Laser scanning, Fusion, Matching, Registration
ABSTRACT
The request for three-dimensional digital city models is increasing and
also the need to have more precise and realistic
models. In the past, 3D models have been relatively simple. The models were derived from aerial images or laser scanning
data and the extracted buildings were represented by simple shapes. However, for some applications, like navigation with
landmarks or virtual city tours, the level of details of such models is not high enough. The user demands more detailed
and realistic models. Nowadays, the generation of detailed city models includes usually a large amount of manual
work, since single buildings are often reconstructed using CAD software packages and the texture of facades is mapped
manually to the building primitives. Using terrestrial laser scanners, accurate and dense 3D point clouds can be obtained.
This data can be used to generate detailed 3D-models, which also include facade structures. Since the technology of laser
scanning in the field of terrestrial data acquisition for surveying purposes is new, the processing of the data is only poorly
conceived. This paper makes a contribution to the automatic registration of terrestrial laser scanning data recorded from
different viewpoints. Up to now, vendors of laser scanners mainly use manual registration mechanisms combined with
artificial targets such as retro-reflectors or balls to register single scans. Since these methods are not fully automated,
the registration of different scans is time consuming. Furthermore, the targets must be placed sensibly within the scan
volume, and often require extra detail scans of the targets in order to achieve accurate transformation parameters. In this
paper it is shown how to register different scans using only the measured point clouds themselves without the use of
special targets in the surveyed area.
1 INTRODUCTION
The fact that many research groups and institutions are en-
gaged in the field of 3D city modelling shows how impor-
tant it is to expand existing databases with the third dimen-
sion. There are a lot of applications for 3D city models,
but an area-wide use has been prevented by the lack of eco-
nomic and fast methods to obtain such models. Up to now
city models have only been created for single cities or even
parts of cities, despite the fact that many institutions and
companies are interested in them. City models are often
provided only in a low level of detail. For example in Ger-
many, Phoenics is stocked with a selection of digital city
models that cover more than 30,000 km? (Phoenics GmbH,
2004). The most important applications of city models in
the future are the integration of 3D data in car navigation
systems and in urban GIS systems for disaster manage-
ment, the support of the urban planning process, virtual
walkabouts in cities for tourists, simulation and analysis
purposes and the video game industry. This is a wide field
and the applications differ in the demands on the city mod-
els. On the one hand detailed models, for example for ur-
ban planning, are required. Facades and also small parts
of a building have to be modelled. On the other hand, for
example for navigation systems, a complete coverage is re-
quired. Detailed models are only needed for single points
of interest, standard buildings can be represented as a sim-
! Corresponding author
1091
ple model. To cope with such diverse requirements, several
levels of detail (LoD) have been introduced. A LoD con-
cept for 3D city models is discussed in Kolbe and Gróger
(Kolbe and Gróger, 2003).
In the past, several methods for the extraction of build-
ings for the generation of city models have been proposed.
They can be classified according to the data source they
use. In the following some previous work is mentioned
briefly. Brenner developed a semiautomatic approach and
uses ground plans of buildings and aerial laser scanning
data to derive building models (Brenner, 2000). The soft-
ware CyberCity Modeler can derive building models from
photogrammetric stereo images (CyberCity AG, 2004). An
early approach to combine DSM's with stereo images to
extract parametric buildings is presented by Haala (Haala,
1996).
Today, users of 3D city models also ask for detailed and
visually interesting models. Therefore, the airborne based
data is not sufficient. Detailed terrestrial data is required to
improve the quality of city models. Terrestrial laser scan-
ning is one possibility to collect the amount of informa-
tion which is needed for detailed building modelling. In
the next chapter it is described how terrestrial laser data is
currently acquired and the potential to simplify the mea-
surement procedure is exposed. In chapter 3, a method for
the registration of scans is presented using only the scan
data itself. The method is based on the identification of 3D