AN APPROACH TO BUILDING EXTRACTION FROM DIGITAL SURFACE MODELS
U. Weidner
Institut für Photogrammetrie
Rheinische Friedrich-Wilhelms-Universität Bonn
Germany
weidner@ipb.uni-bonn.de
Commission Ill, Working Group 2
KEY WORDS: Digital Surface Model, DEM, Building Extraction, Geometric Object Modeling
ABSTRACT
Motivated by the test data sets of ISPRS WG I11/3 on image understanding we investigate the feasibility of building extraction
using high-resolution Digital Surface Models (DSM) as input data, which do not only contain information about the topographic
surface like Digital Elevation Models (DEM), but also information about the buildings. The steps of the proposed procedure
increasingly use explicit domain knowledge, specifically geometric constraints in the form of parametric and prismatic building
models. The reconstruction of the prismatic models and the selection of the models are based on the principle of Minimum
Description Length (MDL). In addition, we also discuss the possible use of information from GIS or maps in our approach.
KURZFASSUNG
Unsere Arbeiten bezüglich der Gebäudeerfassung aus hochauflösenden Digitalen Oberflächenmodellen, die nicht nur Information
über die topographische Oberfläche wie Digitale Höhenmodelle, sondern auch über Gebäude enthalten, sind durch die von
der ISPRS WG I11/3 herausgegebenen Testdatensätze angeregt worden. Die Schritte des hier aufgezeigten Verfahrens nutzen
Wissen über Gebäude, welches durch die verwendeten parametrischen und prismatischen Gebäudemodelle repräsentiert wird.
Die Rekonstruktion der prismatischen Modelle und die Auswahl des für die Rekonstruktion anzuwendenden Modells basieren auf
dem Prinzip der Minimalen Beschreibunglänge. Desweiteren wird der Einsatz von GIS und Karten als zusätzliche Datenquelle
im Zusammenhang mit der Gebäudeerfassung aus Digitalen Oberflächenmodellen diskutiert.
1 INTRODUCTION
During the last recent years the need for 3D data describ-
ing urban areas increased. This data is needed for a variety
of applications such as town planning, architecture, micro-
climate investigations or transmitter placement for telecom-
munication. Our investigations concerning the feasibility of
building extraction using high resolution Digital Surface Mod-
els (DSM) as input data, which do not only contain infor-
mation about the topographic surface like Digital Elevation
Models (DEM), but also information about the buildings,
have been motivated by the test data sets of ISPRS WG
IH/3 on image understanding. In contrast to other authors,
who use digital imagery (e.g. [Lang and Schickler, 1993],
[McGlone and Shufelt, 1994]) or digital imagery and DSM
(e.g. [Haala, 1994], [Baltsavias et al., 1995]), we solely fo-
cus our investigations in a first step on DSM for two reasons:
First of all, we are interested in investigating the potential
which is inherent in the use of such DSM. The main advan-
tage of DSM is that they already provide a geometric de-
scription of the objects, although this description of buildings
shows some deficiencies with respect to building extraction.
These deficiencies are the representation of the object, the
resolution, and the discrimination of buildings and other ob-
jects. The representation of the objects is not sufficient in
all cases, e.g passages. The resolution of ground plan in-
formation is restricted to the resolution of the DSM. The
discrimination of buildings and other objects (e.g. trees) is
not always possible using solely a DSM. Despite these defi-
ciencies, DSM seem to be a good intermediate description
to link sensor data with high level knowledge about build-
ings. Secondly, using only DSM as input data enables us to
use DSM, which are generated by matching techniques using
digital images ([Krzystek, 1991], [Collins et al., 1995]), but
also those derived by other measurement devices like laser
924
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
scanners [Lohr and Eibert, 1995].
Our approach to building extraction from DSM consists of
two steps — the detection and the reconstruction of build-
ings. Both steps use high level knowledge about the build-
ings, which is introduced in the form of parametric and pris-
matic building models, and some specific knowledge about
buildings in the region of interest to fix some building rele-
vant thresholds. All this knowledge is object space related
making adaption to data of different density and resolution
simple and transparent. In [Weidner and Fôrstner, 1995] we
already described the general strategy of our approach, in-
cluding building detection and building reconstruction. The
main extensions of this contribution consists of the automatic
selection of the model for the reconstruction of buildings.
This selection is based on the principle of Minimum Descrip-
tion Length (MDL), as well as the reconstruction of prismatic
building models. Therefore, the next section shortly describes
the MDL-principle, followed by a summary of the general
strategy of our approach. Section 5 shows results for the IS-
PRS test data sets and other data. Section 6 describes, how
GIS or map information can be incorporated in our approach,
followed by the conclusions.
2 MINIMUM DESCRIPTION LENGTH PRINCIPLE
The shape reconstruction described in Section 4 is based on
the principle of Miniumum Description Length (MDL). This
principle provides means to select and estimate the param-
eters of the selected model in a common framework, and
enables us to reconstruct the shape of the buildings by inte-
grating data and model information. The description length
D L depends on the complexity of the used model and the de-
viation of the data from the model. The complexity depends
on the number of unknown parameters and the number of
gi
Le
