The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
1141
Scene
Number accepted
points (Mio.)
Height difference to
ICC DTM
Mean (m) a (m)
Cat
21.5
-0.80
3.41
115/207
23.9
-1.74
4.14
115/208
24.1
-2.14
4.64
116/207
27.9
-3.02
4.84
116/208
26.3
-2.82
5.40
116/209
15.8
-3.70
5.43
117/207
18.0
-3.26
4.98
117/208
21.2
-3.31
5.59
117/209
7.5
-4.83
6.42
Table 3: Accepted object points statistics
3.4 DSM and Orthoimage creation
After the outlier removal step, a DSM on a 10 meter grid is
interpolated for all scenes. Figure 4 shows the difference
between the ICC DTM and the generated DSM for the CAT
scene. The consistency of the produced Aft/Fore orthoimages is
assessed by image matching. Highly accurate tie points
between each ortho pair have been established using LSM. The
mean shift varies between 0.28 and 0.30 pixels, the
corresponding standard deviation between 0.15 and 0.18 pixels.
Figure 3 shows that the remaining subpixel shift between the
Aft and Fore orthoimages is slightly regular.
x (pixel)
Figure 3: Shifts (averaged on a 150x150 pixel grid) between
orthorectified Aft and Fore image of scene 116/208
3.5 Mosaicing
After each scene has been processed independently, DSM and
Aft orthoimages of scenes (115-117)/(207-209) are combined
into orthoimage and DSM mosaics. No further radiometric or
geometric adjustment has been applied. To assess the
consistency of the resulting mosaic, shifts between the
neighbouring orthoimages are computed by LSM matching and
shown in Table 4. This is a good result, considering that no
bundle block adjustment was used, and all scenes were
processed independently. The Ortho and DSM mosaics are
shown in
Figure 4: Height difference (meter) obtained by subtracting
the CARTOSAT-1 DSM from the ICC DTM.
Path/Row
Lateral shift (m)
mean a
(115,116) / 207
1.66
1.46
(116,117)/207
0.92
0.71
(115,116)/208
1.12
0.64
(115,116)/209
3.12
1.31
Table 4: Lateral shift in overlapping areas of neighbouring
orthoimages. Except for the overlap between scene
115/209 and 116/209, subpixel shifts are found.
4. CONCLUSIONS
A highly automated DSM processor for CARTOSAT-1 scenes
is presented. Key features include the automatic collection of
reference data from Landsat ETM+ Geocover mosaic and the
SRTM DSM, a novel affine RPC correction estimation
approach which leads to ortho and DSM products with much
higher lateral accuracy than the ETM+ Geocover mosaic.
Comparison with reference data in Catalonia indicates a
location accuracy of 3-4 meter, and a height accuracy of 3-4 m
in terrain with good pattern matching characteristics. In
mountainous terrain and with rather low sun elevations the
normal problems of too steep slopes and large shadow areas
without enough matching targets lead to coarse reconstruction
and local artefacts in the DSM.
The ability to produce quality DSM and ortho products without
manually measured, highly accurate ground control points is
especially valuable for emergency applications, mapping of
remote areas and large scale DSM production. Seamless
mosaics can be formed, even if all scenes were processed
independently. Further work includes development of
comfortable manual quality inspection and editing tools, as well
as improved computational efficiency and better handling of
occluded and shadowed areas in the stereo matching module.
ACKNOWLEDGEMENTS
EUROMAP provided most of the CARTOSAT-1 scenes used
for this analysis, with the exception of the Cat scene, which was
provided by ISRO/SAC as part of the CARTOSAT-1 Scientific
Figure 5.