Multi-Criteria, Graph-Based Road Centerline
Vectorization Using Ordered Weighted
Averaging Operators
Fateme Ameri, Mohammad Javad Valadan Zoej, and Mehdi Mokhtarzade
Abstract
In this paper a novel road vectorization methodology based on
image space clustering technique and weighted graph theory is
presented. The proposed methodology describes a road as a set
of optimized points on the centerline which should be connected
by defining a number of appropriate criteria. The main contribu-
tion of this paper is to design a weighting scheme for combining
a small number of road identities using Ordered Weighted Aver-
aging (
OWA
) operators by defining appropriate decision strategy.
In this regard, a novel geometric criterion is introduced. Result
of the
OWA
aggregation specifies weight of each edge in the road
network graph. Comparing the proposed approach with two
state-of-the-art image space clustering-based road vectorization
methods proves its efficiency to deal with roads with different
widths, parallel roads with different distances, different types of
intersections, and also noise clusters. Obtaining improved quali-
ty measures for several high-resolution images, demonstrates the
successfulness of the vectorization approach.
Introduction
Identification of digital linear features from remote sensing
data in continuous objects such as roads is a complicated
procedure. Regarding the significance of vector representation
of the roads in many applications, such as automatic vehicle
navigation, traffic management, and updating geospatial data-
bases, health care accessibility planning, and even infrastruc-
ture management delineation of road centerlines from remote
sensing data has been attracted researchers over the last three
decades. (Mohammadzadeh
et al
., 2006)
Linear features have been investigated in various research
studies based on different models. The road model describes
the appearance of a road in the digital image and makes
the task programmable (Gruen and Li, 1995). Some models
consider roads as continuous lines (Gruen and Li, 1995;
Mena, 2006, Mohammadzadeh
et al
., 2006; Poullis and You,
2010). Most of researches model the roads using separate line
segments which should be connected based on defining an
appropriate connection hypothesis (Baumgartner
et al
., 1999;
Hinz and Baumgartner, 2003; Grote
et al
., 2012). Moreover, a
few road extraction methods define the road models based on
groups of points on the central axes of roads which should be
connected considering radiometric or geometric constrains
(Doucette
et al
., 2001; Ferchichi and Wang, 2005; Mokhtarza-
de
et al
., 2010). These point-based road centerline extraction
methods provide results with high accuracy defining appro-
priate geometric and radiometric criteria.
Road extraction researches concentrate on two subjects: (a)
Road detection which focuses on identification of raster road seg-
ments from other feature classes in the image (Zhu
et al
., 2004;
Mohammadzadeh
et al
., 2006; Youn
et al
., 2008; Grote
et al
.,
2012; Matkan
et al
., 2014; Poullis, 2014), and (b) Road vectoriza-
tion is related to extraction of road centerline as vector segment
(Doucette
et al
., 2001; Mena, 2006; Mokhtarzade
et al
., 2010;
Karaman
et al
., 2012; Zarrinpanjeh
et al
., 2013; Shanmugam and
Kaliaperumal, 2015). Among them, some existing road vector-
ization methods are concentrated on image space clustering
techniques to represent the road centerline by connecting initial
road candidates (Doucette
et al
., 2001; Mokhtarzade
et al
., 2010).
There are some problematic issues such as fixed number and
uniform distribution of the initial cluster centers in the Doucette
et al
. (2001) method as well as the complexity of computations in
Mokhtarzade
et al
. (2010) which have motivated us to establish a
novel point-based road centerline vectorization method.
The main contribution of this paper is to design a weighting
and fusion scheme for combining the outputs of the various
image analysis operators. Although most of the earlier road ex-
traction approaches do the same task by variety of techniques
such as artificial intelligence and fuzzy systems (Wiedemann
and Hinz, 1999; Hinz and Baumgartner, 2003; Mokhtarzade
et
al
., 2007 ), segmentation and classification (Bacher and Mayer,
2005; Mohammadzadeh
et al
., 2006; Poullis and You, 2010;
Grote
et al
., 2012), Dempster-Shafer approaches (Mena and
Malpica, 2003; Mena and Malpica, 2005), and Snakes and dy-
namic programming (Gruen and Li, 1995; Gruen and Li, 1997),
this paper uses an adopted weighting and ordering scheme
which needs only few parameters and has enough generality
to vectorize other linear features. In this regard, a new five-step
approach for road centerline vectorization based on image
space clustering and weighted graph theory is introduced. An
Ordered Weighted Averaging (
OWA
) based strategy is used to ag-
gregate the road geometric criteria into the cost of each edge in
the road network graph. Besides, an innovative road geometric
feature is proposed to enhance the results of the aggregation.
OWA
operators are defined as a nonlinear combination
of the ranked input data which was first proposed by Yager
(1988). In the short time since its first appearance, the
OWA
operators have been used in an amazingly wide range of
applications including neural networks (Yager, 1992; Yager,
1995), fuzzy logic controllers (Yager,1991; Yager and Filev,
1992, Chen
et al
., 2012), group decision-making and multi-cri-
teria decision analysis (Herrera
et al
., 1996; Liu
et al
, 2013;
Jelokhani and Malczewski, 2014), data mining (Torra, 2004),
GIS
(Rinner and Malczewki, 2002; Malczewski, 2006;Nadi and
Delavar, 2011), etc. This various applicability is due to the
flexibility of
OWA
to model variety of aggregators and different
Remote Sensing Department, Geomatics Engineering Faculty,
K.N. Toosi University of Technology, No. 1346, ValiAsr Street,
Mirdamad cross, Tehran, Iran, Postal code: 19967-15333(fam-
)
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 2, February 2016, pp. 107–118.
0099-1112/16/107–118
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.82.2.107
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
January 2016
107
