3. Istanbul 2004
, Forstner, W.,
cting buildings
| in 2D and 3D.
3d city models:
cadastral maps.
., editors, Pho-
nich, Germany.
| du bâti en mi-
piques à grande
thesis, Univer-
'onstruction on
ching. In 2nd
;, Vienna, Aus-
Reconstruction
reometric Con-
708.
ogical and geo-
ting polyhedral
4. Amsterdam.
des bátiments
ses cadastrales
ennes à haute
upérieure des
tion von Gebu-
PhD thesis,
) Surface re-
of aerial digital
A probabilis-
g semantic la-
ce, pages 257-
on of 3D linear
odelisation. In
. A framework
ition. In ICIP,
AUTOMATIC ROAD EXTRACTION IN URBAN SCENES — AND BEYOND
Stefan Hinz
Remote Sensing Technology, TU München, Arcisstr. 21, 80 333 München — Stefan. Hinz@bv.tum.de
Commission II, Working Group 3
KEY WORDS: Image Understanding, Road Extraction, Fusion, Self-Diagnosis, Aerial Imagery
ABSTRACT
In this paper, we present work on automatic road extraction from high resolution aerial imagery taken over urban areas. In order to
deal with the high complexity of this type of scenes, we integrate detailed knowledge about roads and their context using explicitly
formulated scale-dependent models. The knowledge about how and when certain parts of the road and context model are optimally
exploited is condensed in the extraction strategy. To exploit information from multiple views, a fusion strategy for road objects (e.g.
lanes) has been developed. It is based on internally computed quality measures and embedded in the system's concept of self-diagnostic
extraction algorithms. The analysis of the final results shows benefits but also remaining deficiencies of this approach. We give an
outlook on the utilization of the approach in applications related with traffic monitoring in urban areas.
1 INTRODUCTION
From a practical point of view, research on automatic road extrac-
tion in urban areas is mainly motivated by the importance of geo-
graphic information systems (GIS) and the need for data acquisi-
tion and update for GIS. This demand is strikingly documented in
the survey on 3D city models initiated by the European Organiza-
tion for Experimental Photogrammetric Research (OEEPE, now
called EuroSDR) a few years ago (Fuchs et al., 1998). Applica-
tions of road data of urban areas include analyses and simulations
of traffic flow, estimation of air and noise pollution, street main-
tenance, etc.
From a scientific perspective, the extraction of roads in complex
environments is one of the challenging issues in photogrammetry
and computer vision, since many tasks related to automatic scene
interpretation are involved. Factors greatly influencing the scene
complexity are, for instance, the number of different objects, the
amount of their interrelations, and the variability of both. More-
over, each factor—and thus the scene complexity—is related to
à particular scale. To accommodate for such factors, techniques
like detailed semantic modelling, contextual reasoning, and self-
diagnosis have proven to be of great importance over the past
years. It is clear that these techniques must be integral parts of an
extraction system to attain reasonably good results over a variety
of scenes.
Before describing the details of how our approach tries to cope
with these challenges we briefly review work on automatic road
extraction with emphasis on approaches dealing with urban envi-
ronments (Section 2). In Section 3, we present underlying ideas
and basic components of our road and context model. The ex-
traction strategy is outlined in Section 4, illustrated by results of
intermediate steps. Special focus is thereby on fusing road ob-
jects extracted in multiple overlapping images. In Section 5, a
numerical evaluation of the results currently achievable with our
system is given followed by a discussion of the advantages and
remaining deficiencies of the proposed approach. We conclude
the paper with an outlook on future work (Sect. 6).
2 RELATED WORK
Compared to the relatively high number of research groups focus-
Ing their work on road extraction in rural areas, only a few groups
work on the automatic extraction of roads in urban environments
(see articles in (Grün et al., 1995, Grün et al., 1997, Agouris and
Stefanidis, 1999, Baltsavias et al., 2001)). Most of the past and
current efforts in road extraction rely on road models that describe
the appearance of roads in rural terrain rather than in settlements.
However, throughout all the different approaches, some issues
have proved to be essential: By integrating a flexible, detailed
road and context model one can capture the varying appearance
of roads and the influence of background objects such as trees,
buildings, and cars in complex scenes (Baumgartner et al., 1999,
Ruskone, 1996, Strat and Fischler, 1995). The fusion of different
scales helps to eliminate isolated disturbances on the road while
the fundamental structures are emphasized (Mayer and Steger,
1998). This can be supported by considering the function of roads
connecting different sites and thereby forming a fairly dense and
sometimes even regular network. Hence, exploiting the network
characteristics adds global information and, thus, the selection of
the correct hypotheses becomes easier (Fischler and Heller, 1998,
Price, 2000, Wiedemann and Ebner, 2000). Last but not least,
another important point is the integration of self-diagnosis tech-
niques. They are used to evaluate the reliability of hypotheses
of both low level features and higher level objects, which in turn
facilitates decisions that inherently appear during the extraction
process (Hinz and Baumgartner, 2002, Tupin et al., 1999, Tónjes
et al., 1999).
3 MODEL
3.1 Road Model
The road model illustrated in Fig. 1 a) compiles knowledge about
radiometric, geometric, and topological characteristics of urban
roads in form of a hierarchical semantic net. The model rep-
resents the standard case, i.e., the appearance of roads is not
affected by relations to other objects. It describes objects by
means of “concepts”, and is split into three levels defining dif-
ferent points of view. The real world level comprises the objects
to be extracted: The road network, its junctions and road links,
as well as their parts and specializations (road segments, lanes,
markings,..). These concepts are connected to the concepts of
the geometry and material level via concrete relations (Tónjes et
al, 1999), The geometry and material level is an intermediate
level which represents the 3D-shape of an object as well as its
