QUALITY ASSESSMENT OF ROAD DATABASES USING AERIAL IMAGERY 
M. Gerke 
Institute of Photogrammetry and Geolnformation, University of Hannover 
Nienburger Str. 1, D-30167 Hannover, Germany - gerke@ipi.uni-hannover.de 
Theme Session 14 
KEY WORDS: Database, GIS, Imagery, Networks, Parameters, Quality, Reliability 
ABSTRACT 
Digital road databases are widely used in many facets of our daily life. Most of these dat 
quality indication, but often more detailed quality descriptions regarding possible errors, 
information on the completeness of the vector data are desirable. In this paper an approach 
abases come with a nominal 
the positional accuracy, and 
road data from the Authoritative Topographic Cartographic Information System (ATKIS) of Germany is introduced. The 
work is embedded in a project initiated by the German Federal Agency for Cartography and Geodesy (BKG), which is 
interested in an automation of the road data verification process. 
How existing road vectors from ATKIS can be assessed by combinin 
g the information coming from several object extrac- 
tion algorithms is investigated. These objects are modeled in the so called relationship model where the topologic and 
geometric relation between roads and other obj 
a minimum and a maximum distance from the carriageway. 
aerial imagery - may then support a given ATKIS road. If it does not coincide with the model it g 
the ATKIS road. The Hint-Theory is used which is derived from the Dempster- 
all information related to an ATKIS road segment. Example results show that th 
reliable information on the quality of ATKIS objects. 
1 INTRODUCTION 
Nowadays, large scale road vector data is available in many 
countries as part of the national geo-spatial core data. Ques- 
tions are starting to arise from the user’s side: is the data 
accurate enough for a particular application, is it up-to- 
date and are the attributes correct? In this paper a method 
for an automatic quality assessment for given road vector 
data using information automatically extracted from digital 
aerial images is developed. Quality comprises complete- 
ness, positional accuracy, attribute correctness and tempo- 
ral correctness for each object. The presented method is 
not designed to check the completeness as only objects 
contained in the database are considered (verification of 
existing data). However, a potential extension regarding 
the detection of new roads will be sketched in the outlook. 
In (Gerke et al., 2004) road objects from the Authoritative 
Topographic Cartographic Information System (ATKIS) of 
Germany are verified using automatic road extraction al- 
gorithms. The road extraction algorithm used in that work 
exploits knowledge on the appearance of roads in aerial 
or satellite imagery, but does not consider so called local 
context objects. These objects (such as rows of trees) may 
hamper the extraction of roads, as these may not be directly 
visible due to occlusion. The explicit modeling of the topo- 
logic and geometric relation which do exist in reality bet- 
ween such context objects and road objects helps to inter- 
prete gaps in road extraction and thus supports road ex- 
traction, e.g. see (Hinz and Baumgartner, 2000) and (Hinz, 
2003). 
In this work the topologic and geometric relations between 
local context objects, extracted roads and ATKIS road ob- 
jects are modeled in a so called relationship model. The 
802 
ects are given. For example a row of trees is often parallel to roads and has 
Every extracted object - such as rows of trees extracted from 
ives evidence against 
Shafer Theory of evidence to combine 
e introduced procedure is able to yield 
goal is to assess given ATKIS objects by means of ex- 
tracted objects (either local context objects or road objects). 
Every extracted object gives a certain portion of evidence 
regarding the hypothesis that a certain object from the AT- 
KIS database maintains the modeled relations. In order to 
balance the given evidences the Hint-Theory being an ap- 
proach to the Dempster-Shafer-Theory is applied. 
2 THE HINT-THEORY: AN APPROACH TO EVI- 
DENCE-THEORY 
The background of the Evidence-Theory (E-T) is the as- 
sessment of incomplete knowledge by means of degrees of 
belief (lower probability) and degrees of plausibility (up- 
per probability). The roots of E-T can be found in (Demp- 
ster, 1967), whereas the actual origin of E-T is known to 
be set by Shafer in his monograph (Shafer, 1976). The de- 
gree of belief (often called credibility) expresses to what 
extent information can be trusted. The degree of plausibil- 
ity specifies to what extent there is no disagreement regard- 
ing an information. Further information regarding E-T can 
be found in (Shafer and Pearl, 1990), an introduction to the 
Dempster-Shafer-Theory is given in (Gordon and Short- 
liffe, 1990). 
The Hint-Theory (H-T) is an approach to the E-T, its fun- 
damentals can be found in (Kohlas and Monney, 1995). 
The measure to what extent a hypothesis is proved by the 
Hint # is called support (degree of certitude). The ex- 
tent to what there is no disagreement to a hypothesis is 
called plausibility. The interpretations of support and plau- 
sibility are very close to Dempster's theory of upper and 
lower probability. Hints are combined applying Demp- 
ster's Rule. 
for the quality description of 
Intern 
In (K 
apprc 
tatior 
restin 
bility 
the e 
hypol 
to cal 
In E- 
there! 
quire 
In thi 
proac 
tribut 
objec 
tain tl 
a dist 
influe 
cann 
3 R 
The a 
objec 
a mo 
classe 
major 
positi 
ner ar 
jects 
given 
that th 
allow: 
matio 
The r: 
jor cl: 
road c 
lation: 
model 
lation: 
dersto 
as inti 
sists 1 
assess 
object 
Objec 
segme 
string 
contra 
tional 
repres 
quanti 
racy). 
is anal 
ment r 
an ass 
The r« 
Le ti 
This k 
ject ex