)04
the
ere
me
PS
out
ure
the
Ire
ce
re
71
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B6. Istanbul 2004
RMSE Number of GCPs
l 3 6 10
| Horizontal 0.959 0.979 0.992 0.951
Vertical 1.319 1.263 1.284 1.302
Table 2. Accuracy obtained from different number of GCPs.
It can be seen that there is no significant difference in the
RMSE with varying numbers of GCPs. In all cases it can be
seen that IKONOS imagery with bias-corrected RPCs produces
RMS error of around 1m in planimetry and 1.3m in height. The
approach yielded accuracy surpassing specifications for the
much more expensive Pro and Precision products. The
accuracy obtained is around the same as Precision Plus, which
is the most accurate product provided by Space Imaging.
Further investigation was undertaken by fixing the number of
GCPs at three and varying their location, as shown in Figure 5.
Three different cases were investigated: i) all points at the
center, ii) all points at different corners, and iii) all points in the
same corner. From table 3, case III produces less desirable
results as compared with the other two, but such a GCP
arrangement may be necessary in particular cases such as
mapping over the national border where GCPs can only be
established on one side.
Figure 5. Three different configurations of GCPs.
RMSE Case I Case II Case III
Horizontal 0.951 0.902 1.247
Vertical 1.302 1.229 1.305
Table 3. Accuracy obtained from the cases in Figure 5.
S5. ALTERNATIVE SENSOR MODEL FOR RPC
In instances where RPCs are not provided, eg with standard
IKONOS Geo image products, further alternative orientation
models need to be considered. Once such model is based on
affine projection (Fraser et al, 2003). The general form of the
model describing transformation from 3D object space (X, Y, Z)
to 2D space (x, y) for a given point / within an image is given as
3
Yymds X4 de; d Z; d, $
The affine model can be considered as a special case of the
more general rational function model where the polynomial
degree is one and the denominator is also one. The coefficients
Az and As can be interpreted as translation parameters. The A,
A» As, As, Ag and A, coefficients contain combined effects of
rotation and scale. Despite its simplicity, which may raise a
81
degree of suspicion, recent research by Fraser et al (2002) has
shown that this model can produce high geopositioning
accuracy comparable to that produced via RPCs for IKONOS
Geo imagery..
The attractive feature of this approach is that no RPCs need to
be purchased and this leads to an even more economical means
of acquiring and using HRSI. The tradeoff is that more GCPs
are required. As can be seen in (3), at least four GCPs are
required versus one in the case of bias-correction described in
Section 2. However, in the case of mapping remote areas, the
major portion of cost of the ground measurement is to get there.
Since GPS measurement techniques are now very efficient,
establishing a few more control points does not necessarily
generate substantially more cost. Another favourable feature of
this approach, especially when using Barista software, is that
whenever RPCs are required they can easily be generated.
Three different sets of 6 GCPs with 75 check points were tested
with the affine model and all yielded almost identical results.
The planimetric accuracy was just above 1m while vertical was
again around 1.3m. These results are effectively the same as
obtained from the bias-correction method in Table 2.
6. CONCLUDING REMARKS
This paper has highlighted that lowest-cost HRSI products can
meet large-scale mapping needs of developing countries. The
experimental results clearly demonstrate the capability of HRSI
for large-scale mapping. In metric terms, a 1:10,000 scale
topographic map can be derived from IKONOS imagery and
even up to a 1:5,000 scale line map can also be produced. In the
test reported, however, the resulting RMS errors were only just
below the threshold and thus could be considered a little too
close to allowable tolerance limits for 1:5,000 mapping.
Apart from metric quality, factors such as whether the desired
features at the required scale can be clearly identified on the
source images need to be taken into consideration. Though this
topic was not addressed in the current research, it is worthwhile
to note that this user requirement is no less important than the
metric accuracy. It should also be emphasised that results
shown in this paper are from the use of only one stereo-pair.
Further evaluation of both metric and non-metric aspects of
HRSI is warranted, particularly on the larger blocks of
overlapping imagery that are often the case in real-world
mapping projects.
ACKNOWLEDGEMENTS
This research has been supported by a grant from the Geo-
Information and Space Technology Development Agency
(GISTDA) of Thailand. In particular, the authors would like to
thank Dr. Suvit Vibulsresth, GISTDA director, and his staff
including Dr. Chaowalit Silphatong and Dr.Pakorn'Apaphan for
the acquisition of the IKONOS stereo pairs and for other
assistance throughout the project. The support of Dr.Wicha
Jiwalai, GISTDA chairman, during the initial project phase is
also much appreciated.
REFERENCES
Ager, T.P., 2003. Evaluation of the Geometric Accuracy of
IKONOS imagery. SPIE 2003 AeroSense Conference, Orlando,
Florida.
