(b) DOM generated from CE-1 CCD images
Figure 5. Partial DEM and DOM generated using CE-1 CCD
images (take No.0562 orbit for example)
4.2 Experiment on CE-2 CCD Images
(a) Forward-looking (b) Backward-looking
Figure 6. Partial forward- and backward-looking lunar images
taken by CE-2 CCD camera from No.0581 orbit (7m resolution)
By now, all the images of CE-2 CCD camera with a resolution
of 7m can be applied from China Lunar Exploration and Aero-
space Engineering Center. We also choose 3 orbits (No.0579,
No.0580 and No.0581) of CE-2 CCD images located at Sinus
Iridum. Figure 6 shows partial forward- and backward-looking
images (15,000 rows, after grayscale stretching) of No. 0581
orbit.
In our experiment, we use 60,000 rows by 6,144 columns for-
ward- and backward-looking images for each orbit. After fea-
ture point extraction, matching and gross error elimination (us-
ing RANSAC procedure), around 60,000 evenly distributed
conjugate points for each stereo pair are obtained and used for
space intersection and back-projection. Table 5 shows the re-
siduals between the feature points and the back-projected points
correspondently in image space which indicate that the residu-
als mainly exist in the column direction.
Orbit Column direction Row direction
NO Image Mean RMS Mean RMS
: (pixel) (pixel) (pixel) (pixel)
0579 F -22.77 2.22 0.00 0.00
B 22.73 2.21 0.00 0.00
F -23.46 1.93 0.00 0.00
0550 B 23.42 1.92 0.00 0.00
0581 F -22.34 1.81 0.00 0.00
B 22.30 1.81 0.00 0.00
Table 5. Back-projection residuals in image space for CE-2
CCD images before adjustment
Comparing with Table 2, it is obvious that the back-projection
residuals of CE-2 images are much larger than those of CE-1
images. As analyzed early, this is mainly because the two linear
arrays of the CE-2 camera are separately assembled on the focal
plane, while the three line CE-1 camera is realized by one area
array sensor. Thus, the large residuals in column direction are
not mainly caused by attitude angle bias but by errors of interior
orientation model. We fitted LSLs of column residuals versus
column position for forward- and backward-looking images of
these three orbits. The coefficients a and b of them are listed in
Table 6.
Orbit Coefficient a Coefficient b
No. F B F B
0579 -0.0011020 | 0.0011000 -23.02 22.97
0580 -0.0008601 | 0.0008583 -23.46 2342
0581 -0.0009592 | 0.0009575 -22.19 22.15
Table 6. Coefficients a and 5 of the fitted lines of column resid-
uals for forward- and backward-looking images of CE-2
Mean values of coefficients a and b of three orbits in Table 6
are converted to scale and offset and then used in Equation (8)
to refine the interior orientation model. Back-projection residu-
als are calculated again after the refinement. Table 7 shows the
results which indicate that residuals in column direction are re-
duced significantly from over 20 pixels to 0.02 pixel .
Orbit Column direction Row direction
ral Image Mean RMS Mean RMS
: (pixel) (pixel) (pixel) (pixel)
F -0.02 1.22 0.00 0.00
0379 B 0.02 1.22 0.00 0.00
F -0.02 1.22 0.00 0.00
0330 B 0.02 1.22 0.00 0.00
0581 F -0.02 0.81 0.00 0.00
B 0.02 0.80 0.00 0.00
Table 7. Back-projection residuals in image space for CE-2
CCD images after interior orientation model refinement
The residuals distributed in column direction before and after
adjustment (interior orientation model refinement) as well as
the LSLs are shown in Figure 7 which visually demonstrated
the effectiveness of this method.
We can also observe that the RMS residuals are relatively larger
than the mean residuals, e.g., up to 1.22 pixels. This indicates
that there is still room for further improvement, for example, fit
ting a 2™ order polynomial may generate better result than fit-
ting
tí
cokmn residusis
xt
3
column resituais
=
Figu
and
exan
DEN
enta
Figur
imag
In thi
of CI
objec
By ar
ods t
Expe
attitu
Very i
quent
cisior
In the
furthe
