4.2 Preprocessing
The photogrammetric restitution of 3-line imagery requires
a large number of conjugate points. Digital image match-
ing is an appropriate technique to automatically determine
these points. Before starting the matching procedure, the
image strip of the nadir looking CCD array was resampled
by factor 3 to obtain the same pixel size in all 3 strips.
Using the least squares region-growing matching algorithm
(Heipke et al. 1994) about 14000 conjugate points were
found. The standard deviation of the image coordinates
was assumed to be 0.3 pixel.
In the area covered by the 3 image strips 79 DGPS-derived
ground control points (GCP) were available with a stan-
dard deviation of 0.1 m in X, Y and Z. 75 points were
identified and measured in the images (Baltsavias 1995).
During the D2 mission tracking was routinely performed
using the Tracking and Data Relay Satellite System
(TDRSS). The orbit determination for orbit #75B is based
on 900 S-Band Doppler measurements with a sampling
rate of 10s covering about 180 minutes. The force and
measurement modeling comprises the shuttle epoch state
vector, the drag coefficient and 5 parameters describing
perturbations of the attitude thruster system. The pure
statistical standard deviations of the epoch state vector
components were 30 m, whereas unmodeled effects of the
attitude thruster system contribute an additional error of
up to 50m (Braun, Reigber 1994).
A major problem arose from the fact that the image
recording times could in general not be related better than
0.5s to the time scale UTC. A time offset of 0.5s corre-
sponds to an along-track position offset of the space shuttle
of 0.5s- 7 km/s — 3.5 km (!). Since no parameter for the
time offset exists in the bundle adjustment algorithm, a
realistic weighting matrix for the epoch state vector com-
ponents has been derived to relax the orbital constraints
in the along-track direction (Gill et al. 1995). The force
model parameters p were treated as constants due to the
short time span (1 min) of image acquisition.
Attitude information was derived from gyro recordings of
the Inertial Measurement Units (IMU) of the shuttle Guid-
ance Navigation and Control System. Based on approxi-
mation tests the optimum distance between two orienta-
tion points was found to be 4615 rows, corresponding to a
flight distance of 62.3 km and a flight time of 9.1.
4.3 Combined Bundle Adjustment
The combined bundle adjustment was performed in the
geocentric coordinate system WGS84. The DGPS-derived
ground points were divided into two groups. The first
group consists of 12 GCP, where 3 points each are located
in the corners of the threefold overlapping area to ensure a
precise definition of the global datum. The second group
comprises 63 geometrically well distributed points which
were used as check points.
The following data were introduced as observations:
e [mage coordinates of 13959 conjugate points
(c —0.3 pixel)
e Image coordinates of 12 GCP (o —0.5 pixel)
e [mage coordinates of 63 check points (o —0.5 pixel)
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
e Object coordinates of 12 GCP (ox 2 ey — cz =
0.1 m)
e Epoch state vector components with associated 6x6
weighting matrix
e Attitude parameters (6,1,0) at 8 orientation points
AI Je qa pug
In Table 3 the rms values of the theoretical standard devi-
ations of the check point coordinates A ad Z and
the corresponding empirical values are presented. The
theoretical values were computed from the inverted nor-
mal equation matrix and the a posterior óo value of the
bundle adjustment. The empirical values were derived by
comparing the estimated object coordinates of the check
points and the known values.
MOMS #75B theor. | empir.
psg [m] |: 611.2 9.3
HT hal | 19.1 10.2
iz [m] 8.9 11.2
a [m]. (44.2: |; 108
Table 3: Rms values u$, py, uz and py; of the the-
oretical standard deviations derived from 63 check points
and corresponding empirical values
The good correspondence between the theoretical and the
empirical values proves the correctness of the stochastic
and the functional model. The empirical values show that
accuracies of about 10m (0.7 pixel) in X, Y and Z were
achieved. A graphical analysis of the residuals in the check
points showed that the results are not affected by system-
atic errors.
Due to the improper time synchronization between image
and orbit data, large corrections to the a priori state vec-
tor components occured, that contribute mainly to along-
track position errors. The theoretical accuracy of the six
components was improved compared to the a priori val-
ues by a factor of 6 to a level of 150 m, primarily due
to the availability of GCP. The orbit accuracy from the
pure TDRSS solution (50-70 m), however, could not be
improved. A high accuracy orbit determination from a
combined bundle adjustment thus requires a time synchro-
nization of 0.1 ms or better.
5 CONCLUSIONS AND OUTLOOK
In this paper a new concept for integrated orbit and
point determination in satellite photogrammetry is pre-
sented and verified by simulated Mars96 HRSC/WAOSS
and practical MOMS-02/D2 data. For the first time orbit
determination results are rigorously incorporated into the
bundle block adjustment, which is equivalent to a com-
bined adjustment of tracking and image data. The pro-
posed concept guarantees the proper utilization of orbit
information in the bundle adjustment and, vice-versa, en-
ables the use of image information to improve the orbit
determination and to support the estimation of scientific
parameters (e.g. Mars rotation parameters).
The ne
MOMS
station
MOMS
ronme:
height
tively.
consist
orbit a
imager
will en
justme
arcs wl
Anoth
based «
by a C
flyby 1
and at
opport
Dactyl
well as
I woul
tenbru
work p
of the
Albert.
(199
Miss
Baltsa:
Braun
minz
neuv
Mec
dard
Case J
lytic
25(5
Davies
B.H.
Tjuf
Wor!
tatio
Cele:
Davis
neou
sider
Air
tion
Ebner
stud:
Spac
7(1),
Ebner
Bild:
Mars
Fern:
