SEMI-AUTOMATIC ROAD EXTRACTION 
FROM DIGITAL AERIAL PHOTOGRAPHS 
O. Eker**, D. Z. Seker® 
* General Command of Mapping, Photogrammetry Department, 06100 Dikimevi Ankara, Turkey - oeker@hgk.mil.tr 
° ITU, Civil Engineering Faculty, 34469 Maslak Istanbul, Turkey - seker@itu.edu.tr 
KEY WORDS: Photogrammetry, Extraction, Aerial, Vector, Method 
ABSTRACT: 
By means of the progress in photogrammetry; it is now possible to produce the maps of huge areas more quickly and economically 
but the production steps are still expensive and time consuming. In recent years, parallel to the development in photogrammetry, 
more effective methods are researched to make available digitizing line features like roads full automatically or semi-automatically 
from digital aerial photographs. Because it is not still possible to designate the full automatic feature extraction methods clearly, the 
semi-automatic feature extraction methods, which combine experienced interpretation of an operator with the speed of an computer 
algorithm, are the best solutions. There are different semi-automatic approaches for digitizing features from digital aerial 
photographs. Nowadays, suitable methods for each specific mission are chosen in the literature, instead of one exact method which 
carries out all alternative algorithms. In this paper, roads are chosen as line features for digitizing from digital aerial photographs, 
because roads are the most time consuming features to digitize in photogrammetry. Two road extraction algorithms are applied to 
digitize roads from mono digital aerial photographs and the results are compared. 
1. INTRODUCTION 2.1 Characteristics of the roads 
By means of the progress in photogrammetry; it is now possible The characteristics of the roads can be classified in five groups 
to produce the maps of huge areas more quickly and as below: 
economically but the production steps are still expensive and 
time consuming. In recent years, parallel to the development in — Geometry 
photogrammetry, more effective methods are researched to — Radiometry 
make available digitizing line features like roads full — Topology 
automatically or semi-automatically from digital aerial — Functionality 
photographs. Because it is not still possible to designate the full ~ Contextual 
automatic feature extraction methods clearly, the semi- 
automatic feature extraction methods, which combine 
experienced interpretation of an operator with the speed of a 
computer algorithm, are the best solutions. 
An operator uses these characteristics to recognize the roads 
during the evaluation and mapping processes from the imagery. 
Both of the two semi-automatic road extraction methods focus 
on radiometric and geometric characteristics because the other 
characteristics require more intelligence in order to exploit them 
automatically from the imagery so complex situations are left to 
the operator. Examples of these characteristics can be seen 
below: 
In this paper, roads are chosen as line features for digitizing 
from digital aerial photographs, because roads are the most time 
consuming features to digitize in photogrammetry. There are 
several semi-automatic road extraction methods but two of 
them are chosen for the application. First one is based on edge 
and correlation analyses, developed by Aluir Porfirio Dal Poz. 
The other one is a road tracing algorithm by profile matching 
and Kalman filtering developed by George Vosselman and 
Jurrien de Knecht. In section 2 the road characteristics and the 
principles of two algorithms are described. Section 3 contains 
the experiments which are carried out by a software while 
conclusions are presented in section 4. 
— Roads are elongated (geometry) 
— Roads have a maximum curvature (geometry) 
— The road surface usually is homogeneous (radiometry) 
— The road surface often has a good contrast with the 
adjacent arias (radiometry) 
— Roads do not stop without a reason (topology) 
— Roads intersect and build a network (topology) 
— Roads connect cities (functional) 
2. ALGORITHMS OF METHODS — Higher roads (fly-overs) may cast a shadow (contextual) 
— Trees may occlude the road surface, but, on the other 
In this section the algorithms of two methods are described but hand, an array of trees may also indicate a road 
first of all the road characteristics have to be investigated (contextual) (Vosselman and Knecht, 19953), 
because the algorithms are based on these characteristics. 
  
* Corresponding author. 
518 
Inte 
2.2 
An 
tha 
val 
mo 
sec 
maf 
esti 
pos 
of 1 
pro 
pro 
pro 
par 
foll 
the 
and 
unt 
Kn 
In | 
ma 
shi: 
is 
trai 
squ 
esti 
wic 
The 
par 
apr 
cas 
par 
The 
19€ 
If t 
dyr 
it 
pro 
In 1 
usii 
up 
epc 
ma 
tim 
ttc 
The 
pro 
dif 
gec 
cor 
mil 
bet 
Af 
pre