The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
1288
either by mere shifts, by shifts and drifts, or by an affine
transformation, all in image space. It has been shown in the past
that the effect of the biases is almost constant in the images, so
that by using one well-defined GCP per image, the bulk of the
bias can be removed (Fraser and Hanley, 2003). Results can be
improved if a modest number of additional GCPs are available.
Third, a generic pushbroom sensor model can be initialized
from information contained in the metadata files and refined in
an orientation process by the use of well-distributed GCPs. If
the metadata of CARTOSAT 1 imagery contained information
about the orbit path and attitudes, the generic pushbroom
scanner model could also be used for direct georeferencing, and
the GCPs could be used for determining biases in the orbit path
and attitude information. As this is not the case, the orbit path
and attitudes need to be determined from GCPs alone. The
number of GCPs required will be slightly higher than would be
the case if orbit path and attitude information were available.
In this paper, the authors assess and compare the geopositioning
accuracy achievable with CARTOSAT 1 imagery via the 3D
affine, bias-corrected RPC and generic pushbroom sensor
models. A stereo pair of images covering Hobart, Australia,
distributed by ISRO as part of the Cartosat-1 Scientific
Assessment Programme (C-SAP), was processed using Barista.
Unencrypted RPC files were provided by ISRO for research
purposes in this case. General information about the satellite
can be found in CARTOSAT la (2006) and CARTOSAT lb
(2006). Scene related information is contained in the provided
metadata files. A set of GPS-surveyed 3D points for the area of
Hobart was available for the reported analysis. In order to
assess the georeferencing accuracy that can be achieved for
CARTOSAT 1 images, bundle adjustment was carried out,
applying the three sensor models accommodated within Barista
and using a subset of the available 3D points as GCPs. The
remaining 3D points served as independent check points.
2. THE HOBART DATA SET
2.1 Imagery
For the assessment of the georeferencing accuracy a stereo pair
of panchromatic CARTOSAT 1 images was used. The nominal
flying height of the satellite is 618 km and the two cameras are
mounted with a tilt of +26 degrees (fore) and -5 degrees (aft).
Table 1 shows some satellite specifications. Further information
can be found in CARTOSAT la (2006) and CARTOSAT lb,
(2006). The provided metadata files specified a slightly
different value of around 638 km for the flying height. No scene
specific across-track angle information is given. The forward
scene covers an area of roughly 30 km x 30 km around the city
of Hobart. The area covered by the aft view is about 27 km x 30
km. The elevations in this area range from sea level up to over
1260 m at the peak of Mt Wellington. The acquisition date of
the Hobart images was 1 October 2006.
was then assessed via the remaining 60 points, which served as
independent checkpoints. For the generic pushbroom sensor
model an adjustment was also performed with 15 GCPs and 54
checkpoints in order to investigate the influence of the number
of GCPs on the results. Figure 1 shows the CARTOSAT 1
forward-looking scene with the distribution of the 15 GCPs.
The 9 GCPs covered the same area but with a less dense point
distribution.
Orbital altitude
618 km
Swath
Fore
29.42 km
Aft
26.24 km
Along-track view angle
Fore
+26°
Aft
-5 0
Across-track resolution (at
Nadir)
Fore
2.452 m
Aft
2.187 m
Ground sampling distance (along-track)
2.54 m
B/H ratio
0.62
Image size
12 000 x 12 000
Pixel size
7x7 microns
Focal length
1945 mm
Table 1. CARTOSAT 1 specifications.
Figure 1. CARTOSAT 1 forward-looking scene of Hobart
with 15 GCPs.
3. SENSOR MODELS IN BARISTA
2.2 3D Points
In addition to the imagery, a test field of altogether 69 3D
points was utilised. This test field is described in Fraser and
Hanley (2005). The points were distributed all over Hobart and
covered the top left quarter of the scene. They were measured
with GPS and were mainly centres of road roundabouts,
determined as the centres of ellipses fitted to points around the
roundabout perimeter. A selection of 9 points was used to
provide GCPs for the bundle adjustment. The absolute accuracy
The processing of the CARTOSAT 1 data set of Hobart was
carried out with the software system Barista. Barista is
developed at the CRC-SI and has reached the status of a
commercially available product. It is a software system for the
generation of spatial information products from satellite
imagery. Data sets from high-resolution satellites such as
QuickBird, WorldView, IKONOS, SPOT 5, and ALOS/PRISM
have been processed successfully with Barista. Further details
about the software can be found in Barista (2008).