The International Archives ofthe Photoprammetrv. Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
(sensor orientation) and Event (SPICE) library provided by 
NASA’s Navigation Ancillary Information Facility (NAIF). The 
navigation data is stored as kernels and can be accessed by time 
(Yoon and Shan, 2005). The acquisition time and number of 
image lines can be retrieved from each image header. Plugging 
them into the SPICE kernel will generate the line-by-line 
telemetry trajectory data needed in this study. 
Different from the situation on earth, no absolute ground truth is 
available on the Martian surface. Therefore, the performance of 
the bundle adjustment was evaluated in terms of back-projection 
residuals in image space. The refinement of EO parameters was 
also analyzed. 
Bundle adjustment aims at removing the inconsistencies between 
HiRISE stereo images by adjusting their EO parameters. The 
refined EO parameters are compared with those obtained from 
telemetry data. Figure 2 presents their differences in graphic 
format. The horizontal axis of Figure 2 is the image row index 
and the vertical axis is the difference. 
Change of Xs(red) Ys(green) Zs(blue) 
in meters. 
<r seconds 
13 * ‘ ; • • •• ^ ; f 
c «æ icon issi .T«; ja* ms m tm tssa 
Change of Omega(red), phai(green) and 
Kappa(biue) in seconds 
(a) Victoria Crater 
Change of Xs(red) Ys(green) Zs(biue) 
in meters. 
Change of Omega(red), Phai(green) and 
Kappa(blue) in seconds. 
(b) Husband Hill 
Figure2. Difference between telemetry and refined EO 
Besides the tie points, a comparable number of evenly distributed 
check points independent from the bundle adjustment were also 
evenly selected for evaluation. The differences between check 
point measurements and ground check point back-projection on 
the image, called back-projection residuals, represent the 
inconsistencies between HiRISE stereo images. Figure 3 shows 
the back-projection residuals on the images covering the two 
sites before and after bundle adjustment; table 2 shows the 
corresponding statistics. 
Before BA 10 times exaggeration After BA 100 times exaggeration 
(a) Victoria Crater 
Before BA 100 times exaggeration After BA 300 times exaggeration 
(b) Husband Hill 
Figure3. Check point back projection residuals on the images 
Site 
Victoria Crater 
Husband Hill 
Status 
Before BA 
After BA 
Before BA 
After BA 
Mean(pix) 
-26.4 
-0.3 
-0.5 
0.0 
Max(pix) 
-29.2 
-1.4 
4.1 
0.5 
Std(pix) 
0.81 
0.3 
2.0 
0.2 
No. of 
Points 
135 check points 
180 check points 
Table 2. Statistics on back-projection residuals at both sites 
Results show that along-track error is more significant than 
cross-track error before adjustment. A possible explanation is that 
telemetry EO greatly depends on the accurate time measurement. 
Any error in time measurement will more greatly affect on the 
along-track direction in which the satellite is moving. After 
bundle adjustment, along-track error greatly reduced. This 
corresponds to the larger correction of (p and Z s compared 
to other pointing angles and position elements as is showed in 
Figure 2. It is noted that Z s is mostly along track because both 
sites are in equatorial area. Finally, comparing the two data sets 
studied, Husband Hill images shows smaller back projection 
residuals before BA and smaller EO refinement after BA. This 
indicates that telemetry EO might have been improved during the 
period between taking the two stereo pairs.