gital Images 
posium, IAP 
f Corner and 
ol. 2031, pp. 
ol structures 
n in digital 
d Extraction 
S Com. III 
art 3/1. 
s. SPIE Vol. 
of Regions 
lata Layers. 
grammetry , 
87. Snakes: 
nference on 
L.G., 1995, 
ng multiple 
June. 
Boundaries 
derstanding 
]. IT. 
51903 A 
. SPIE Vol. 
Kübler, O., 
. Automatic 
and Space 
ser Verlag, 
roving the 
Algorithms. 
derstanding 
ner-Fritsch- 
ns, Herbert 
for Active 
n,. 13:2. pp. 
)ynamics of 
at SPIE'93, 
Diego, CA, 
4. A Road 
rties of the 
of damping 
8-246. 
cally-Based 
SPIE Vol. 
deling with 
26,2, July, 
D., 1993, 
sed Surface 
tic Feature 
Man-Made 
ler-Agouris 
. 95-104. 
FUSION OF 2D-GIS AND IMAGE DATA FOR 3D BUILDING RECONSTRUCTION 
Norbert Haala and Karl-Heinrich Anders 
Institute of Photogrammetry, Stuttgart University 
Keplerstrafe 11, 70174 Stuttgart 
Phone: 0711-121-3383 
e-mail: norbert.haala@ifp.uni-stuttgart.de 
Commission III/3 
KEYWORDS: Three-dimensional, Reconstruction, Human Settlement, Fusion, GIS, Aerial, Image 
ABSTRACT 
This paper addresses the issue of extending preexisting two-dimensional geometric data representing the ground plans 
of buildings to three-dimensional descriptions by the combination of this data with the analysis of aerial images. Fre- 
quently, ground plans have already been registr 
ated and represented either analog in maps and plans or digital in a Geo 
Information System (GIS). Within the article the utilized ground plans are exemplary provided by a digital cadastral 
map, which is assembled as an area covering data base and at the moment is available for 40 % of Germany. Even 
though this information is very valuable 
for the 3D building reconstruction, the available data structures contained in 
the GIS have to be expanded considerably until they can be used to support the interpretation of images. In general 
the available information has to be transformed into more comp 
on how to use this information has to be provided. This implies steps 
lex structures, which are more suitable and knowledge 
like the extraction of relations which are only 
contained implicitly in the available data, the elimination of unnecessary information (e.g. details not visible in aerial 
images) and the generation 
of hypotheses on the missing 3. dimension of the represented building. These hypotheses 
the can be verified and the unknown parameters can be determined by the analysis of the aerial images. 
1 INTRODUCTION 
At present most GIS products and also authoritative GIS 
use a flat surface (planimetry) as spatial reference and 
therefore only provide 2D data structures. Nevertheless 
there is a clear trend towards systems which are able to 
represent, manage and analyze 3D information. An ex- 
ample for the growing number of applications of this kind 
of data are 3D urban models; simulations and visualiza- 
tions demanding three-dimensional descriptions of build- 
up areas have become standard applications for planning 
purposes in urban regions. Objects mainly relevant for 
the description of these areas are Digital Terrain Mod- 
els (DTM) and man-made objects like buildings, roads or 
other supply facilities. This article will concentrate on the 
acquisition of three-dimensional building reconstructions. 
The data capture for that purpose is frequently done by 
analyzing aerial imagery, since photogrammetric data is 
three-dimensional, exact, largely complete and up-to-date. 
Because manual interpretation is very time consuming, a 
lot of effort has been spend to speed up this process by au- 
tomatic or semi-automatic procedures. Even though the 
use of aerial imagery as mere database for automatic re- 
construction of the three-dimensional building geometry is 
sufficient in principle, especially while aiming on fully au- 
tomatic procedures considerable progress can be achieved 
if additional information is used. One reason for the actual 
problems of automatic image interpretation is the great 
amount of (partly irrelevant) information contained in im- 
ages. Grey-values are influenced by a large number of fac- 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
tors, e.g. object geometry, shadows, texture and reflec- 
tions. This frequently makes it difficult to separate be- 
tween irrelevant details and important information, which 
is required for a task like the automatic reconstruction of 
objects from images. Nevertheless, the three-dimensional 
reconstruction of quite complex objects like buildings can 
be improved considerably if image data is combined with 
other data sources. One example is the additional use of 
height data. Since height data is independent from illumi- 
nation or surface material the extraction of the required 
geometric information and therefore the interpretation of 
this kind of data is easier compared to image interpreta- 
tion. By a combination of both data sources their specific 
strengths — relatively simple interpretation of height data 
on the one hand, great content of information, richness 
of detail, high resolution and accuracy of image data on 
the other hand — can be utilized (Haala 1995). This arti- 
cle aims on the combination of image data and preexist- 
ing two-dimensional ground plans of buildings to capture 
three-dimensional reconstructions of buildings. The use of 
information, which is supplied by a GIS is also motivated 
by the fact that due to the growing number of already ex- 
isting databases, apart from the initial data acquisition, 
the upgrade, completion and extension of these existing 
databases will become an aspect of growing importance. 
In the following chapter available data sources, which can 
be used for 3D building reconstruction will be described. 
In order to integrate this 2D data into the automatic im- 
age interpretation, in the first processing step the used 
data has to be expanded e.g. by the extraction of rela- 
285