CONTRIBUTION OF EXTERNAL DATA TO AERIAL IMAGE ANALYSIS
Ghislaine BORDES, Philippe GUERIN
LG.N. - DT/ SR/ Matis - Saint-Mandé - FRANCE
email : bordes, guerin(a)matis.ign.fr
Gérard GIRAUDON
LN.R.LA. - Projet PASTIS - Sophia Antipolis - FRANCE
Henri MAiTRE
E.N.S.T. - Département image - Paris - FRANCE
ISPRS Commission IV, Working Group 1 : GIS Data and Applications
KEY WORDS : Cartography, Database, Aerial image analysis, Road extraction
ABSTRACT :
To face the difficulties of a fully automatic aerial image analysis, we study the potential contribution of external
data. Different methods to integrate external data in image analysis are presented. Two examples of image
analysis which use external data are described more precisely. Both of them are taken in the domain of
cartography and concern the road network extraction in aerial imagery. The external data are provided by a
cartographic database in the first case and by a scanned map in the second one.
RÉSUMÉ :
Pour surmonter les obstacles rencontrés lors de l'élaboration d'un systéme entiérement automatique d'analyse
d'images aériennes, une solution souvent adoptée consiste à utiliser des données externes. Différentes méthodes
d'utilisation de ces données sont présentées. Deux exemples précis d'utilisation de données externes pour
l'analyse d'image sont décrits. Ces deux exemples, issus du domaine de la cartographie, concernent plus
particuliérement l'extraction de routes dans les images aériennes. Les données externes qu'ils utilisent sont
fournies par une base de données cartographiques dans le premier exemple et par une carte scannée dans le
second.
1. INTRODUCTION
The French IGN (National Institute of Geography)
intends to automate the acquisition of the BD Topo®, a
topographic database which content fits with 1:25 000
scale maps, with a metric accuracy for the three
coordinates X, Y and Z. Its acquisition is based on
stereoplotting of 1:30 000 aerial images. In this context
of automation, one of the solutions to reach an automatic
and reliable aerial image analysis is to use external data.
That is to say a priori data about the scene which has to
be analyzed.
In parts 2 and 3 we describe the advantages and
shortcomings of using external data for image analysis. In
part 4, the different methods to integrate external data in
image analysis processes are presented. Finally, in part 5
and 6, two examples of the external data contribution to
road extraction in aerial images are described. In the first
case, the external data source is a cartographic database,
in the second one it is a scanned map.
134
2. MAIN DIFFICULTIES OF IMAGE
ANALYSIS
Automatic interpretation of aerial images is known as a
difficult task which has to face important obstacles. The
complexity of the geographical reality is coupled with the
diversity of the contexts to handle (rural or urban areas,
coastal or mountainous landscape,...). The global
consistency between the objects has also to be taken into
account to explain the structure, the positionning and the
shape of these objects. As all understanding system is a
processing sequence of operators, each step produces
errors that are difficult to be handled for the next step.
Thus, the control of the global system is a very complex
task, requiring the most higher level of mechanisms.
Finally, all the information necessary to understand the
underlying structure of the scene is not a visual one (for
example, the administrative boundaries). For all these
reasons, existing algorithms often fail to analyze aerial
images with reliable results. An attractive solution to
reach an exhaustive analysis is to use symbolic external
data that include a part of the information missing to
perform the image understanding.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
