ata with
learners
lham at
an from
ings of
. Clas-
, Mon-
. Data
rigs and
lassifi-
lensity-
tabases
ry and
S. and
tic ap-
a. Pro-
lm and
edings
artially
trans-
n Arti-
Active
331.
>f Fea-
ng de
metric
2006.
Notes
on for
ars 24,
rithms
37(3),
msing
metry
ilding
SPRS
9.
Cooperative Research Center for Spatial Information, Department of Geomatics
University of Melbourne VIC 3010, Australia
[m.ravanbakhsh, c.fraser]@unimelb.edu.au
KEY WORDS: roundabout detection, feature extraction, topographic database, high resolution imagery, snake model, level sets
ABSTRACT:
Road roundabouts, as a class of road junctions, are generally not explicitly modelled in existing road extraction approaches. This
paper presents a new approach for the automatic extraction of roundabouts from aerial imagery through the use of prior knowledge
from an existing topographic database. The proposed snake-based approach makes use of ziplock snakes. The external force of the
ziplock snake, which is a combination of the Gradient Vector Flow force and the Balloon force, is modified based on the shape of
the roundabout central island to enable the roundabout border to be delineated. Fixed boundary conditions for the proposed snake are
provided by the existing road arms. A level set framework employing a hybrid evolution strategy is then exploited to extract the
central island. Black-and-white aerial images of 0.1 m ground resolution taken over suburban and rural areas have been used in
experimental tests which have demonstrated the validity of the proposed approach.
1. INTRODUCTION
The need for accurate spatial databases and their automatic
updating is increasing rapidly. Road networks form key
information layers in topographic databases since they are used
in such a wide variety of applications. As the extraction of
roads from images is still generally manual, costly and time-
consuming, there is a growing imperative to automate the
process. However, such a feature extraction task has longed
proved difficult to automate. The problem for automatic road
extraction lies mostly in the complex content of aerial images.
To ease the complexity of the image interpretation task, prior
information can be used (Gerke, 2006; Boichis et al., 2000;
Boichis et al., 1998; De Gunst, 1996). This often includes the
provision of data from an external topographic database.
Roundabouts, as a class of road junctions, are important
components of a road network and if modelled well can
improve the quality of road network extraction (Boichis et al.,
1998). However, there are only few approaches which are
dedicated to this task. Boichis et al. (2000) presented a
knowledge based system for the extraction of road junctions
and roundabouts. The method assumed that the description of
simple road junctions and roundabouts is the same in the
external database, so a previous detector has to certify the
presence of the circular form. A parametric Hough Transform is
used for this purpose. The roundabout is reconstructed after
straight parts of the connecting roads, curved parts including
splitter islands, and the circulating road are extracted.
These elements are connected using geometric and radiometric
continuities. In the approach, roads are treated as linear objects.
Thus, elements such as the central island and the roundabout
outline are not extracted, so kind of modelling does not always
reflect the required degree of detail. In Fig. 1, vector data is
superimposed on sample images to illustrate the problem. The
image resolution is such that the roundabout’s central area
covers the central island and the circulating roadway. In Fig.
lb, the roundabout is represented as point object neglecting the
central island and the circulating roadway. Thus, a detailed
modelling of roundabouts is needed for data acquisition
purposes at large scales.
The detailed modelling of road roundabouts area objects is
discussed in this paper, and an approach for their automatic
extraction is proposed. This uses an existing topographic
database leading to the extraction of refined roundabout data. In
the following section, a model for roundabouts is first
introduced. The stages of the proposed strategy are then
illustrated in Sect. 3. Results from the implementation of the
proposed approach using aerial imagery of 0.1 m ground
resolution are presented and discussed in Sect. 4, together with
an evaluation of their quality. Finally concluding remarks are
offered.
Figure 1. Superimposition of vector data on high resolution
aerial images of road roundabouts.
2. ROUNDABOUT MODEL
Illustrated in Fig. 2a is the conceptual two-part model of a
roundabout, the parts being the roundabout itself and the road
arms. The roundabout, where road arms are connected, is in
turn composed of the roundabout border and its central area
where a central island is located. A road arm is a rectilinear
object which is represented as a ribbon with a constant width
and two parallel road edges. Disturbances such as occlusions
and shadows are not explicitly included in the model.