ıbul 2004
'esolution
matching.
'esolution
rocessing
of high-
15. Image
recovery
Opt. Soc.
'esolution
sequences
Int. Conf.
SP), San
f a high-
tion, and
IEEE Int
ation and
d Image
GENERATING TRUE ORTHOIMAGES WITHOUT A 3D SURFACE MODEL
J. Albertz, B. Wolf
Technical University of Berlin, EB 9, Strafe des 17. Juni, D-10623 Berlin — (albertz, zwercmea)@fpk.tu-berlin.de
Commission IH, WG I11/8
KEY WORDS: Matching, Rectification, Orthoimage, Three-dimensional, Tree-Line, Configuration
ABSTRACT:
It is the traditional procedure to generate orthophotos from aerial photographs by differential rectification. The effects of relief
displacement are corrected by considering a Digital Terrain Model (DTM). Due to the fact that DTM's do not describe the surface of
buildings, bridges, trees etc. such objects remain in perspective views in the resulting orthophotos. If a Digital Surface Model (DSM)
describes the mentioned objects, the caused displacements can also be corrected, and the results are called “True Orthoimages”.
However, recent developments of optoelectronic cameras like HRSC or ADX 40 allow for a totally new approach for orthoimage
generation. The nadir-looking channels of these cameras provide image data in parallel projection along the flight line, and in central
perspectives across. If a scene is imaged twice in flight lines perpendicular to each other, the first image strip provides correct
ground coordinates of any object in one direction, the second image strip in the other direction. The new approach takes advantage of
these particular geometric properties. It is based on the definition of corresponding points in the image data sets by means of
matching techniques or corresponding areas by means of segmentation procedures. Each conjugate point directly provides the
correct ground coordinates and can be assigned as a pixel to the orthoimage. Thus, a true orthoimage is generated without any
knowledge of the object heights or further calculations. The paper provides a detailed description of the approach, and its advantages
and limitations are discussed.
KURZFASSUNG:
In der Regel werden Orthophotos aus Luftbildern mittels Differentialentzerrung gewonnen. Durch das Gelände bedingte Lage-
versetzungen werden mit Hilfe eines Digitalen Gelándemodells (DGM) korrigiert. Da in einem DGM Objekte wie Gebäude,
Brücken und Báume nicht beschrieben werden, unterliegen diese im resultierenden Orthophoto weiterhin den zentralperspektiv
bedingten Lageversetzungen. Digitale Oberfláchenmodelle beschreiben die genannten Objekte, so dass die verursachten Lage-
versetzungen berücksichtigt werden können. Das Ergebnis wird „True Orthoimage* genannt.
Die Entwicklung von optoelektronischen Kamerasystemen in jüngster Zeit, wie HRSC oder ADX 40, ermöglicht nunmehr eine
völlig neue Verfahrensweise. Die senkrecht nach unten gerichteten Nadirkanäle liefern in Flugrichtung Bilddaten in Parallel-
projektion und senkrecht dazu in Zentralprojektion. Wird eine Gegend zweimal in senkrecht zueinander liegenden Bildstreifen
aufgenommien, liefert der erste Streifen lagetreue Koordinaten in einer Richtung und der zweite Streifen entsprechend in der anderen
Richtung. Der neue Ansatz nutzt die Vorteile dieser speziellen Aufnahmegeometrie. Er basiert auf dem Auffinden korrespondie-
render Punkte in den Bilddaten mittels Bildzuordnung oder dem Auffinden homologer Flächen durch Segmentierungsverfahren.
Jeder homologe Punkt liefert direkt lagetreue Koordinaten und kann als Pixel dem Orthoimage zugewiesen werden. Auf diese Weise
entsteht ein „True Orthoimage" ohne jegliches Wissen über Objekthöhen oder weitere Berechnungen. Der Beitrag gibt eine detail-
lierte Beschreibung des Ansatzes und behandelt seine Vor- und Nachteile.
This paper will provide a totally new approach for generating
orthoimages, which is based on digital airborne data recorded
with optoelectronic line scanners, and does not require any
information about the objects height or their geometry.
1. INTRODUCTION
The traditional generation of orthoimages uses digital elevation
models to consider the effects of relief displacements, which
occur through the central perspectivity of photographs. Ortho-
images, i.e pictures in parallel projection, are derived from
these photos by differential rectification methods. Objects
above the surface, like buildings and bridges, are mostly not
described in the used elevation model. Therefore such objects
are distorted from their true position, and the orthoimage shows
e.g. leaning buildings and bent bridges. Some interesting infor-
2. TRADITIONAL PRODUCTION
OF TRUE ORTHOIMAGES
Digital terrain models (DTM), which describe the topographic
surface geometrically, are so far generally used for the
mation from ground features like streets and other objects is
hidden for the user of the orthoimage. Furthermore the super-
imposition of vector data is nearly impossible. This is why
many attempts are made to generate so called "True ortho-
photos". All known approaches for this purpose require a de-
tailed three-dimensional description of buildings etc. in vector
format, which is difficult to achieve.
generation of orthoimages. It is well known, that objects not
modelled in the DTM, particularly man made objects like
buildings and bridges, are displaced in these procedures and are
shown in a wrong position. The results of these displacements
are e.g. leaning buildings and warped bridges. In order to
improve the situation, Digital Surface Models (DSM) or Digital
Object Models (DOM) describing the mentioned objects
