The International Archives of the Photogrammetrw Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
4.2 Bundle Adjustment with Ground Control 
To demonstrate the possibility of integrated bundle-adjustment 
using both orbital and ground rover imagery, points measured 
from bundle-adjusted Spirit rover images were incorporated into 
the HiRISE bundle adjustment as control points using Husband 
Hill data set. During 4 years of rover localization operations at 
the Mapping and GIS Laboratory at OSU, a 10-kilometer 
bundle-adjusted traverse of the Spirit rover was generated. An 
localization accuracy of 0.2 percent of the whole traverse 
distance was achieved. Three dimensional (3D) ground 
coordinates of four distinguishable topographic features were 
measured from the bundle-adjusted rover images and 
corresponding features were identified on the HiRISE stereo 
images (Figure 4). Following equations in Wang 1990, the 3D 
coordinates of the four points were transformed from the Local 
Tangent Plane Coordinate System to the Mars Body-fixed 
Coordinate System and then used as control points in the bundle 
adjustment. Just as with the tie points, image coordinates of the 
control points were related to the ground control points and EO 
parameters using Equation 4. But unlike tie points, the control 
point ground positions do not vary during the iteration process. 
><■" 'Tf'C'f&il '4§lllill| 
dot O'-c-la; ch#ck ooints. 
Figure 4. Point distribution and correspondence between orbital 
and ground imagery (Red Circle: GCP; Red Dot: tie point; Blue 
Dot: check point) 
With rover imagery incorporated, the inconsistencies between 
orbital and ground imagery becomes another criterion of 
performance for the bundle adjustment. First, the 3D ground 
coordinates of a topographic feature are measured from rover 
stereo images. Afterwards, the ground coordinates of the same 
feature are intersected using HiRISE stereo images based on both 
telemetry and bundle-adjusted EO parameters. In order to 
remove the systematic shift between orbital and ground imagery, 
a 3D translation is conducted before BA so that the positions of 
the first point from both orbital and ground imageries are 
identical. The comparison of the positions derived from orbital 
imagery and rover imagery are showed in Table 3. 
Before BA 
After BA 
dX(m) 
dY(m) 
dZ(m) 
dX(m) 
dY(m) 
dZ(m) 
Point 1 
0 
0 
0 
-0.04 
-0.32 
-0.06 
Point2 
21.6 
-17.1 
-2.74 
0.67 
0.08 
0.18 
Point3 
49.2 
-32.3 
-10.4 
3.70 
1.85 
-0.06 
Point4 
61.0 
-7.62 
-51.6 
2.07 
-1.00 
1.33 
Table 3. The difference between orbital and ground based ground 
control points measurement before and after HiRISE BA with 
control from Spirit rover imagery. 
The results show that the inconsistencies between orbital and 
ground imageries can be reduced from dozens of meters to 
several meters or even sub-meter level by very simple integration 
without considering error accumulation in the rover traverse. 
Therefore, topographical accuracy could be further improved by 
integrating ground measurement into HiRISE bundle adjustment. 
5. INVESTIGATION OF JITTER 
Small motions of spacecraft around its nominal pointing, called 
jitter, will distort the images. This problem was identified for 
Mars Orbiter Camera (MOC) images, but it is more severe for 
HilRISE because of HiRISE’s higher resolution (R. Kirk, 2007). 
High-frequency jitter can be filtered out by subtracting the best 
fitting polynomial from the original telemetry HiRISE pointing 
angle data. An 80,000 line image in Gusev Crater was used in 
this study. Figure 5 shows the extracted jitter onO), (p, K, with 
the horizontal axis as image row index and vertical axis as jitter 
magnitude in arc-seconds. An analysis to the extracted residuals 
in spectral domain does not show any frequency significance. 
Therefore, it could be very difficult to incorporate this “jitter” 
into a mathematical model. 
Jitter, as is caused by unpredictable action forces on the camera, 
cannot be modeled by polynomials and therefore, it can not be 
solved by bundle adjustment. Thus, the topographic effect of 
orbital jitter must be evaluated for topographic capability