Albert Baumgartner 
  
MULTI-SCALE ROAD EXTRACTION USING LOCAL AND GLOBAL GROUPING CRITERIA 
Albert Baumgartner, Stefan Hinz 
Chair for Photogrammetry and Remote Sensing 
Technische Universität München, D-80290 Munich, Germany 
E-mail: {albert}{hinz} @photo.verm.tu-muenchen.de 
URL: http://www.photo.verm.tu-muenchen.de 
KEY WORDS: Image Understanding, Road Extraction, Grouping. 
ABSTRACT 
In this paper we combine two approaches for road extraction. The first approach makes use of multiple scales to detect 
roads segments and employs local grouping criteria and also context information to extract the road network. This ap- 
proach is suitable for aerial imagery with a resolution of 0.2-0.5 m. The second approach was designed to extract roads 
from satellite imagery and can be applied to resolutions of 2-5 m. It fuses lines extracted from different channels for road 
extraction and exploits especially the connectivity properties of roads, i.e., global network criteria for road extraction. By 
combining the two approaches we can reduce the effort for selecting appropriate parameters, because both help each other 
to get rid of some individual deficiencies. In addition, the evaluation of the extracted road network showed significant 
improvements compared to the results we get by applying each approach on its own. 
1 INTRODUCTION 
There is a big economic desire to automate the extraction of objects from aerial and satellite imagery, and there is a lot of 
research in this field, too. Nevertheless, fully automatic extraction of objects like buildings or roads is still an unsolved 
problem. At our institute two different approaches for fully automatic road extraction have been developed during the 
past years. The first approach makes use of multiple scales to detect roads segments and employs local grouping criteria 
and also context information to extract the whole road network (Baumgartner et al., 1999). The second approach focuses 
on the connectivity properties of roads and is able to make use of the information derived from multi-spectral satellite 
imagery (Wiedemann and Hinz, 1999). Whereas the first approach is restricted to gray scale imagery with a resolution 
of 0.2 to 0.5 m, in which roads appear as homogeneous regions, the second approach models roads as lines and is able to 
fuse lines extracted in multiple channels. Both approaches show individually good results — within a limited scope. In the 
work presented in this paper we show how both approaches can be combined and how they benefit from the strengths of 
each other and help to overcome their deficiencies. 
Some of the basic ideas of our road extraction scheme are described in detail in earlier publications, e.g., in (Steger et al., 
1995, Baumgartner et al., 1997, Steger et al., 1997, Mayer and Steger, 1998). Work related to our local approach for road 
extraction has been carried out, by (McKeown Jr. and Denlinger, 1988, Ruskoné et al., 1994, Airault et al., 1994). E.g., 
(Ruskoné, 1996) proposed a fully automatic approach for the extraction of road networks from digital aerial imagery: 
Hypotheses for connections between automatically detected seed points are checked based on geometrical constraints. 
The influence of neighboring objects on road extraction has been investigated in (Bordes et al., 1997). For our second 
approach the relevant previous works are (Fischler et al., 1981, Vasudevan et al., 1988) 
Apart from the trend towards the integration of contextual information, there is a strong emphasis on defining evaluation 
criteria and developing methods to evaluate the results of automatic and semi-automatic approaches for road extraction, 
see e.g. (Heller et al., 1998, Heipke et al., 1998, Harvey, 1999). 
The model which serves as the basis of our road extraction scheme is outlined in Section 2. In Section 3 we describe 
the individual approaches for road extraction which are combined in Section 4. The benefits of this combination are 
documented by an external evaluation of the results (Section 5). In Section 6 we draw some conclusions. 
2 ROAD MODEL 
We base our road extraction scheme on the road model displayed in Fig. 1. The road model comprises multiple scales and 
describes the road network in three different levels. The real world level contains the road objects (e.g., road network, 
junction) and their relations. At the geometry and material level, the 3D-shape and the material of roads are represented. In 
  
58 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
  
ee ————————— nien — - LLL E 
  
contra: 
Depen 
solid d 
The us 
scales. 
be det: 
coarse 
clearly 
scale. 
In add 
erties 
This t 
Where 
the im 
open .? 
of the 
road e 
is sup]