C, WAOSS
resolution for
mmetric data
1ases: First, a
matching. In
s obtained by
solution. The
rame camera.
g mehrerer 3-
Teil der pho-
zsansatz bein-
uordnung auf
ebene — wird
Punkte durch
terrestrischen
'sts zeigen die
onstruction of
he generation
ntribution an
he image ori-
e 3-line CCD
ed nearly si-
that 3 linear
he same time,
rrain at differ-
SC consists of
nts each, and
84 active sen-
near periapsis
rbital height),
ant orbit seg-
ground track
Figure 1. A
operations is
given in Albertz et al. (1993) and Ebner et al. (1994).
WAOSS
[ ]
Forward HRSC
Stereo ECO
Far red LE
Photometry CC
104 km
68 km 86 km
Blue [721777] %18 km
L Stereo 1l += 1 ]
Nadir Green CF
143 km Photometry 7)
Infrared C11
Stereo (I1:
[ ]
Backward
62 km
n 519 km i
Figure 1: Ground track of HRSC and WAOSS at 300 km
orbital height (not true to scale)
The image matching procedure of HRSC and WAOSS data
has to consider several factors:
e the highly elliptic orbit, which causes varying
ground pixel sizes in the image strips;
e the different ground pixel size of HRSC and WAOSS
(normally factor 8);
e the macropixel formation due to the limited onboard
storage capacity: the gray values of 4 or 16 pixels
are averaged to 1 macropixel;
e changing illumination conditions during the mission;
e poor texture in parts of the image data to be ac-
quired.
In order to improve the accuracy of the subsequent bundle
block adjustment, the matching approach has to work with
any combination of HRSC and WAOSS imagery taken dur-
ing the mission. Comprehensive computer simulations on
local, regional and global block triangulation have shown,
that most accurate results can be achieved, if image strips
from many orbital arcs are processed simultaneously in a
block of high geometric strength based on a combination
of HRSC, WAOSS, orbit, attitude and ground control data
(Ohlhof 1996).
The matching approach has to cope with different sets of
input parameters. In the most general mode, only a rough
estimation of the overlapping area is known. In the nom-
inal case, preprocessed position and attitude parameters
and image coordinates of a few ground control points mea-
sured interactively are available.
In the next section the matching approach is described in
detail. The approach was tested using terrestrial 3-line
imagery of the MEOSS and MOMS-02/D2 projects and
extraterrestrial frame imagery of the Viking mission. The
results of the tests are presented and discussed. Finally,
conclusions are drawn.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
2 DESCRIPTION OF THE MATCHING
APPROACH
In this section we describe the new image matching ap-
proach. It mainly consists of a combination of feature
based matching and least squares matching. Moreover, im-
age pyramids are incorporated into the strategy. Some spe-
cial adaptations are necessary to properly deal with several
overlapping images recorded by HRSC and WAOSS.
In the first strip! one image! is selected as the reference
image. Then each other image of this strip is matched
with the reference image. The resulting pairs of conjugate
points receive the number of the point in the reference
image. Therefore identification of many-fold points is pos-
sible by a simple comparison of the numbers of the pairs.
In case of two or more strips, the concept developed by
Heipke et al. (1996) is used. For the subsequent block tri-
angulation, the strips have to be connected by tie points.
This task is solved using the concept of point trans-
fer which is well known from analytical photogrammetry.
There is one reference image in each strip. The reference
image of the first strip is matched with the reference image
of the second strip. For points in the reference image of
the second strip conjugate points are searched in the other
images of this strip. If there are regions with no points
transferred from the first reference image, we treat these
areas such as in the reference image of the first strip. Then,
the reference images of the second and the third strip are
matched and so forth. This concept is free regarding the
number of strips in the block.
The concept described above starts on a higher level (e.g. 4
or 5) of the image pyramid. If a sufficient number of con-
jugate points have been found, they are tracked up to
the original images. In Figure 2 the general work-flow
is shown, the single steps are explained in the following
sections.
2.1 Input Data
The input data consists of a set of images, some initial
information about their exterior orientation and a set of
control parameters.
Images. The images with smaller ground pixel size have
to be resampled to the largest common ground pixel size
of all images if there is a significant difference between the
image scales (factor 2 or more). For our task, this step
is necessary if images from both HRSC and WAOSS, or
if images with different macropixel formation are to be
matched. Starting from the common resolution level, an
image pyramid for each image is generated.
Initial Orientation Information. Information about
the exterior orientation of the images has to be provided by
control points, orbital data or both of them. For our task,
control points will be available from the existing ground
control net of Mars. Besides the control points orbital in-
formation, i.e. information about position and attitude of
the scanner for each recorded image line, will be available.
According to Montenbruck et al. (1994) this information is
expected to be very precise (~ 10 m, 4") within one flight
1One strip is normally acquired during one imaging sequence and
consists of at least 3 images (image strips) recorded by the fore-,
nadir-, and aft-looking CCD arrays.