International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012 
XXII ISPRS Congress, 25 August - 01 September 2012, Melbourne, Australia 
304 
afterwards also in the overlapping areas. Again, the blunders are 
detected firstly in ray intersections only, and secondly with refer- 
enc to the MOLA DTM (Spiegel, 2007b). 
For the evaluation of the bundle adjustment results the mean inter 
section error is calculated only for point lying in an overlapping 
area. This reveals the improvement of the bundle adjustment of 
blocks more clearly than the usage of all points of a sub-block 
would do. 
4 RESULTS 
4.1 Systematic bundle adjustment of single orbit strips 
The method for systematic bundle adjustment of single orbit strips 
was used to process the HRSC data received in the almost 7 years 
from January 2004 until November 2010. During this time the 
Mars Express mission has surrounded the planet 8761 times. In 
about 2800 of these orbits the camera took imagery theoretically 
suitable for stereo processing. HRSC image data sets of 2535 
strips were succesfully processed. In 2329 strips (91.9%) the ex 
terior orientation was significantly improved by the bundle ad 
justment. For a number of 123 strips further improvements of the 
exterior orientation were achieved by modeling spacecraft oscil 
lations through a shorter orientation point distance (Bostclmann 
and Heipke, 2011). 
900 
800 
s 700 
В 600 
*S 500 
» 400 
I 300 
= 200 
100 
0 
mean intersection error per strip [m] 
Figure 5: Number of strips grouped by their mean intersection 
error before (light gray) and after (dark gray) the improvement of 
the exterior orientation by bundle adjustment 
In Fig. 5 the mean intersection error for a total of 2535 suc 
cessfully processed strips is presented. The strips are classified 
according to their intersection error in ranges of 10m. The light 
gray bars represent the quality of the exterior orientation data be 
fore, and the dark gray bars after the process of bundle adjust 
ment. It is shown that the number of strips with a mean intersec 
tion error < 10m distinctly increases after the bundle adjustment 
(more precisely from 47 to 437). Also the quantity of strips with 
a mean intersection error lying between 10m and 20m increases 
by a number of 200. On the other hand the number of strips with 
higher errors obviously decreases. E.g. the number with an er 
ror > 100m is reduced from 119 to 38 strips. Probably because 
there was no exclusion of strips with obvious image degradation, 
the mean intersection errors presented here are somewhat higher 
than in other statistics of HRSC DTM processing (e.g. Gwinner 
et al., 2010). In general the mean intersection error presented 
in this paper should not be taken as an absolute criterion for the 
quality of HRSC data, but it is a good indicator for a relative im 
provement. 
4.2 Multi orbit block processing 
The analysis of single orbit strips reveals the internal photogram- 
mctric accuracy of HRSC image data. To eonnect different strips 
to a stable block, in addition a sufficient number of accurate tie 
points in the overlapping area is needed. The previously de 
scribed method of adjusting the exterior orientation of a block 
was used for a set of 21 overlapping HRSC strips. The selected 
area is a region called Nepenthes Mensae between 1°S to 17°N 
and 111°E to 135°E. The footprints of the nadir channel images 
are shown as an overlay to the color coded MOLA DTM in Fig. 
6. 
Figure 6: Color coded MOLA DTM with footprints of 21 HRSC 
nadir channel images constituting a photogrammetric block 
The criteria for the selection of these strips were: 
• The 21 strips overlap to form a single block. 
• All five panchromatic channels are available. 
• The single strip bundle adjustment succeeded with a small 
mean intersection error (less than 20m). 
• The resolution of the nadir channel is better than 30m. 
• The image information allows tie points distributed over the 
whole image. 
• The overlapping area between neighboring strips is rela 
tively wide. 
Table 1 lists the selected strips. They arc ordered going from 
west to east. The data was aquired in four time frames. 14 of 
the strips were taken in the year 2005. Strip h3193 and h3160 
were taken 9 months later in 2006. Three strips (h5212, h5230 
and h5248) were taken in 2008. The newest two strips (h8362 
and h8383) from 2010 fill the remaining gaps. Fig. 7 shows the 
mean intersection error of the strips before and after the single 
strip bundle adjustment. The use of nominal exterior orientation 
results in a mean intersection error of 30m or higher in some 
strips. With the adjusted orientation data it is reduced for all strips 
to a mean of 11.3m.