575 
SPATIAL ROAD NETWORK EXTRACTION FROM MULTI SPECTRAL REMOTE 
SENSING IMAGERY WITH FCD 
Yemin Fan 
Department of Geography, University at Buffalo, State University of New York, Amherst, NY 14261, USA; e-mail: 
yfan2@buffalo.edu 
KEYWORDS: Local Moran’s I statistics, Floating Car Data (FCD), Monte Carlo Approach, Multi spectral Remote Sensing 
Imagery. 
ABSTRACT: 
This paper presents an automatic methodology for spatial road network extraction from multi spectral remote sensing images with 
floating car data. Basically, it can be divided into two steps. In the first step, a spatial local statistics is carried out to extract nodes of 
road segments. Based on local Moran’s I statistics, a new statistic is defined to detect local clusters. Significance is assessed by using 
a Monte Carlo approach to determine the probability of observing that many data under the null hypothesis of no pattern. When all 
necessary nodes are detected, spatial road segments then can be organized by linking each two nodes, which are used as the candidate 
road segments in the second step. In the second step, a pre-processed multi spectral remote sensing image is prepared for testing 
those candidate road segments. Finally, road segments which are significant in the test are selected to construct the spatial road 
network. Methodology and experiments are respectively given in this paper. 
1. INTRODUCTION. 
Spatial road network, as the fundamental component of GIS 
especially in GIS-T (GIS for Transportation), is very important 
both in practical applications and theoretical studies. Topics 
related with spatial road network therefore are discussed by 
many researchers. In recent years, multi spectral remote sensing 
imagery, as a new data source to build the geographical 
information database, is used to provide more detail landscape 
information. Due to its characters of short cycle period and wide 
land coverage, many studies are carried out to extract spatial 
road network from multi spectral images, especially from high 
spatial resolution images. Object oriented methodology is 
introduced to model road objects and road network, in which 
road extraction is generally based on properties of roads and 
road net work (Peteri, Celle and Ranchin, 2003, Dal Poz, Zanin, 
do Vale, 2006). Shackelford and Davis (2003) combined pixel- 
based fuzzy and object-based together to extract road network 
from high-resolution multispectral satellite imagery. 
SkourikHine and Alexei (2005) proposed an image 
vectorization approach to the road network extraction from 
digital imagery which is based on proximity graph analysis. 
Knowledge-based methodologies are also very popular. Zhu, et 
al. (2005) extracts road network based on the binary and 
greyscale mathematical morphology and a line segment match 
method. However, no matter date-driven approaches or 
knowledge-driven approaches are used, they all, to some 
degree, largely depend on inherent character of greyscale 
images. That means radiometric information plays a vital role in 
feature extraction. Therefore, prior knowledge, even when some 
intellectual computation methods are employed, is always 
composed of greyscale character of features and shapes of 
features. In this sense, these methods all have their own 
limitations. 
Nowadays, floating car data (FCD), using dynamic sensor to 
collect spatial information, is developed rapidly. Each car is 
equipped with a GPS device and a wireless communication 
device. The instantaneous position of the car is transmitted at 
regular intervals to a data server centre. Data server centre 
collects and processes all the GPS data sets to facilitate decision 
making on traffic pattern. Nowadays, researches on FCD mainly 
focus on the application of FCD in traffic state detection 
(Kemer and Rehbom, 2001, Schafer, Strauch and Kelpin, 2001, 
Kemer et al. 2005, Kwella and Lehmann, 2000), FCD analysis 
(Fouladvand and Darooneh, 2005), updating road network in 
existing GIS database (Smartt, 2006) and traffic information 
publication (Fan and Liu, 2006). 
In this paper, taking the advantages of multi spectral remote 
sensing imagery and FCD, a new method to extract spatial road 
network is proposed. The significance of this method lies in not 
only helping us to extract the spatial road network using FCD 
and multi spectral RS imagery but also assisting us to divide the 
road network automatically into reasonable road segments 
which are compatible with FCD. Therefore, spatial road 
networks, especially those in the urban area where road 
networks are changing rapidly and difficult to construct, can be 
upgraded both automatically and dynamically. In section 2, the 
feasibility of integrating multi spectral remote sensing imagery 
data with FCD and the coordinate transformation between them 
are discussed. In section 3, based on local Moran’s I statistics, a 
new statistic to carry out a spatial cluster analysis is defined to 
detect nodes of road network. Monte Carlo simulation process 
is adopted to evaluate significance. In section 4, the strategy to 
construct spatial road network with nodes detected and pre- 
processed multi spectral imagery is given out. Experiment 
results are presented and discussed in section 5. Conclusions are 
provided in section 6. In figure 1, the relationship among topics 
discussed in this paper is given out. 
As described in figure 1, FCD database is the data source. 
Monte Carlo simulation process is employed to obtain critical 
values used to detect local clusters. In the candidate link 
selection process, multi spectral remote sensing imagery plays 
an important role to filter candidate road segments. Finally, 
spatial road network is decided by road segments filtered.