i MODELS
as HRSC (High
experiment will
rived products
3 of the sensor
defined. Stan-
ts. Blunt errors
rmed onto the
pendent on the
be closed with
Ist be obtained
bit. This image
s the basis not
' maps, fly-bys
as HRSC (High
le Planet Mars
bgedeckt. Digi-
automatisierten
rdem muß eine
Jenschnitte mit
ke von Objekt-
innt und sofort
srenzkôrper, in
er Qualität der
smethoden ge-
ken angewandt
datei mit einer
uigkeit von 1m
ıdern auch für
processing line
eration, ortho-
f multispectral
production of
eloped for the
/ of Berlin, De-
iphy (Prof. Dr.-
o-Investigators
The Digital Terrain Model (DTM) is the standard form of the
discrete three-dimensional representation of terrains. The
DTM generation belongs to one of the key issues in the
chain of data processing. For many aspects of inter-
pretation of the image data a DTM is required. Also for en-
hancing the accuracy of 2D products like orthoimages
and orthoimage mosaics DTMs are essential. Finally de-
rived products like contour maps, colour-coded height
images and fly-by movies can be created with the help of
DTMs. Because of the wide range of the expected spatial
resolution of HRSC and WAOSS image data, a global DTM
derived from WAOSS data is expected as well as local
DTMs of high resolution using HRSC data.
First implementations of the photogrammetric software
were tested with Clementine images of the Moon. Al-
though the Clementine images are frame-grabber images,
it was possible to simulate line scanner imagery similar to
the ones from the HRSC and WAOSS camera.
The generation of a DTM requires roughly three steps. At
first the object points in space must be calculated from
image coordinates of conjugate points determined by the
digital image matching processes, secondly the object
points must be transformed into the desired map re-
ference system, and finally a regular grid of height points
has to be generated. For this method of generating DTMs,
which is going to be implemented for the Mars96 Mission,
the following input is necessary: results from the corre-
lation of corresponding images and the orientation of the
camera throughout the recording of the images. Further-
more a fixed reference body must be well defined. With
this information a DTM of the area covered by the images
can be computed.
2. COMPUTATION OF OBJECT POINTS
The input for the calculation of object points in space are
the results of the matching process of two or more
images. These results are discrete pixel positions of con-
jugating points in the images. Each pixel in the image data
corresponds with a unique ray, which is well-defined
through camera position and pointing, in space. Given
this information for at least two pixel positions a point in
space can be calculated through a standard ray inter-
section. Using more than two images an adequate least-
square adjustment can be applied. This is very helpful for
finding blunt errors in the foregoing correlation or in the
navigation data as well.
At least two images are needed for the determination of
conjugate points. But the HRSC/ WAOSS experiment with
its nine (HRSC) and three (WAOSS) linear CCD arrays will
provide multiple along-track stereo capability. Even con-
jugate points in cross-track overlapping can be intro-
duced to the system. These results together with the
pointing and position information of the camera for the
corresponding image lines define rays in space. Due to
the sensibility of the system towards the orientation of the
camera, especially its pointing, the navigation parameters
have to be recorded as accurate as possible and will be
improved through a photogrammetric bundle block adjust-
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
ment, which is the task of the Technical University of
Munich, Chair for Photogrammetry and Remote Sensing
(Prof. Dr.-Ing. H. Ebner) Besides the position and
pointing accuracy the quality of the matching process
also has influence on the definition of the object points.
The discrete object points can then be calculated through
a standard ray intersection with a least-squares adjust-
ment technique. Due to the possible number of involved
images and the possible high matching resolution, the
computation can be quite intensive. Utilizing least-
squares adjustment for each single point, blunt errors can
be detected and the affected points are eliminated.
The quality of the points depends on the following:
* quality of the navigation/orbit data
e quality of the correlation
e number of used images
The most important factor defining the quality scale is the
navigation/orbit data.
The result of this calculation process is a dense cloud of
irregularly distributed points in space.
3. TRANSFORMATION OF OBJECT POINTS
ONTO PLANET SURFACE
In order to generate a regular grid of DTM data the
calculated cluster of points in object space must first be
transformed into a geographical system using a reference
ellipsoid. Herefore a standard geodetic transformation is
utilized. The position and the height of each point can be
calculated referring to different ellipsoids. While plani-
metry will be defined on an oblate spheroid the basic
height reference system for the planet Mars is a triaxial
ellipsoid. Thereafter the points are transformed into a
given map projection which defines a rectangular line and
sample coordinate system. However, they still form an
irregular grid.
4. INTERPOLATION OF A REGULAR GRID
From the irregular grid of map projected object points a
regular grid has to be derived by means of adequate
interpolation methods. The following methods are imple-
mented in the software system: nearest neighbour, sector
based interpolation, sector based weighted average inter-
polation (Fig.1) and criging.
Figure 1: Effects of interpolation (gray values = heights).
Left: interpolation using nearest neighbor.
Right: interpolation by a sector based weighted approach
