camera was slowly tilted within the orbit plane. The stereo
images for the second area (Test 2) were selected from a
combination of nadir images (orbit 333), and tilted images
(orbits 334 and 338). The time difference between images taken
on adjacent orbits is about five hours. The time difference
between orbit 333 and orbit 338 is about 25 hours. Image data
information is listed in Table 9.
in the adjustment. The a priori variances are presented in the
Table 12.
The precisions of the navigation unknowns, and the precisions
of the object coordinate unknowns are listed in Table 13. We
also list the precisions of initial object coordinates in the first
step and the object coordinates in the bundle block adjustment.
The precisions of the object coordinates after the adjustment are
essentially improved.
Control Number |Conjugate |Image
points of images [points points Test 1 Test 2
initial estimated [initial estimated
Test 1 0 171 50 1785 values values values values
image (mm)
Test 2 6 96 417 1403 coordinates — [0.01 0.015 0.01 0.015
control - -
Table 9: Image data points 150 48
position for |(m) - -
The results in Test 2 show that offset and drift of the navigation orbit 332 1200 430 = =
data exist for orbit 388 (Table 10). Because the systematic 333 > - 400 141
errors are very large, we have used these adjusted parameters 334 5 = 400 195
(the offset=4392 m, the drift= 2.004 m/km ) as observations in 338 . ]. : 400 134
the further bundle block adjustment. orientation (deg) - 4
for orbit 332 0.02 0.043 - -
C3(z0) 43(z0) 333 - - 0.04 0.020
(m) (m/km) 334 | - 0.04 0.040
times 1 3620 1.977 338 |- - 0.04 0.014
+/-107 +/-0.266
results 4392 2.004 Table 12: A priori variances for the Clementine data
+/-61 +/-0.155
Table 10: Systematic errors for orbit 338 in Test 2 in the Z
direction
The Sequent method is used for searching for gross errors in the
Clementine data during the first step of the adjustment, the
Robust, or the Baarda method, is used in the second step. The
gross errors found in the observations, are given in Table 11.
Two gross errors in the position observations, and one gross
error in the orientation observations were found in Test 1 using
the Baarda method. We did not eliminate the "bad" navigation
observations, but low weights for these are applied (see section
4. 1).
The bundle block adjustment fails if the initial values of the
object coordinates involve very large gross errors, and are not
eliminated by the Sequent method in the adjustment.
Test 1 Test 2
Conjugate |Image Conjugate |Image
points points points points
Sequent
(20 km) 3 29 31 35
Robust 0 21 1 7
Baarda
(5) 0 35 1 5
Table 11: Elimination of errors for the Clementine data
The Clementine image data have three observation groups for
Test 1: image coordinates, position and orientation
observations. In comparison, the image data for Test 2 have
three orbits and control points. Hence they have altogether eight
groups of observations i.e., eight a priori variances are estimated
Test 1 Test 2
mean precisions for (m)
adjusted positions x 1207 100
y 379 134
Z 309 108
mean precisions for (deg)
adjusted orientations ¢ 10.036 0.017
œ 0.031 0.013
K 0.041 0.019
mean precisions for initial (m)
object coordinates x (230 2044
y 1103 2702
Z 547 987
mean precisions for adjusted (m)
object coordinates x |45 96
y 130 150
Z 80 85
Table 13: Adjusted results for the Clementine data
Using the adjusted navigation parameters and the large number
of image coordinates, which are found from the matching
process, the object coordinates are calculated for all conjugate
points. The Sequent method is used again for eliminating gross
errors. Two digital Terrain Models have been produced using
these adjusted object coordinates (Oberst et al. 1996).
The adjusted navigation parameters can also be used in the
production of accurate mosaic images and orthophoto images.
To demonstrate this, we produced two mosaic images using
original navigation parameters (Fig. 1). adjusted navigation
parameters (Fig. 2). The result is a clear improvement in the
quality of the mosaic.
1008
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Fig. 1 : Mosa
Fig. 2 : Mos:
