2004
y the
both
Id be
with
n the
to 8)
1).
pace
ts of
um
e (m)
11.361
1.553
1.328
10.650
10.624
10.955
10.429
ct
uim
e (m)
13.219
17.786
2.757
12.872
13.006
12.787
14.505
3.860
rove
sing
NOS
12.3
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
m in the vertical; while for QuickBird images, RMSE is less than
| m in horizontal and 1.2 m in the vertical. The level of accuracy
has no apparent relationship to the location of the GCP used.
However, since only one GCP is used, the quality of the GCP is
critical to the final result.
Maximum
Method I DC itn RMSE (m Difference (m)
s X Y Z X Y Z
Translation| 1 1 0.61910.669/0.425/0.983/0.960/0.726
irs:
© 8 1-710.942/0.689/0.494/1.39611.26610.811
Zr
2 m
Scale and $ 5|,
mr E 8 |3-6/0.555|1.045|0.445|0.943|1.357|0.735
4 0-3-5-7 .0.291/0.591/0.378/0.39810.700/0.585
6 ]|5-6-3-7-2-0/0.451/0.241/0.175/0.451|0.24110.178
Aff 4 0-3-5-7 ||0.284/0.789/0.362|0.327/0.895/0.539
ME [ e [5-6-3-7-2-0]0.401]0.463]0.174]0.401 0.463]0.174
Second-
Order 6 15-6-3-7-1-0/0.530/0.706/0.342/0.53010.70610.342.
Polynomial
Table 8. Accuracy of ground points improved by four image
space-based methods in QuickBird stereo images
Scale and Translation Model: The scale and translation model
has additional scaling factors in the coordinate axis directions. At
least two GCPs are necessary for this model. In the object space,
if these two GCPs used are distributed in the cross-track direction,
the computed RMSEs of the ground points are relatively smaller
than if they are distributed in the along-track direction. This trend
is consistent with the results achieved using simulated IKONOS
images (Zhou and Li, 2000). The RMSEs calculated by using the
model in image space with two GCPs shows similar results
associated with the GCP distribution. In order to increase
redundancy, more GCPs should be used. With four evenly
distributed GCPs (see 1357 in Table 5 and 0357 in Table 6 — close
to four corners), the result is improved (for QuickBird images, the
RMSE is less than 62 cm in the horizontal and 53 cm in the
vertical) in both object and image spaces. With six GCPs, a more
consistent and better result (for QuickBird images, RMSE is less
than 50 cm in the horizontal and 63 cm in vertical directions) is
shown using the method in both object and image spaces.
Affine Model: The affine model offers the capability of
considering affinity. However, the additional affine parameters
and GCPs do not generate an improvement over the result from
the scale and translation model when used in the object space. In
the image space, however, a comparable result is obtained by
using six GCPs (for QuickBird images, to less than 50 cm in the
horizontal and 20 cm in vertical).
Second-Order Polynomial: The addition of the second-order
parameters requires the use of a larger number of GCPs.
Therefore, it is only applied to image space, where the model uses
six GCPs. No significant improvements are found in comparison
to the other three models. In general, high-order polynomials are
693
very sensitive and require a large number of GCPs and a very
even GCP distribution. The second-order polynomial model does
not exhibit convincing advantages over other models.
The accuracy of the U.S. Geological Survey (USGS) 1:24,000
scale topographic map is approximately 12.2 meters. The
National Oceanic and Atmospheric Administration /National
Geodetic Survey (NOAA/NGS) 1:5,000 Coastal Topographic
Survey Sheet (T-Sheet) is accurate to within approximately 2.5
meters (Li et al, 2001). Thus, ground points derived from
IKONOS one-meter and QuickBird sub-meter panchromatic
stereo images are appropriate for updating features in both the
USGS 1:24,000 topographic maps and the NOAA/NGS T-Sheets.
Furthermore, both kinds of stereo images can be used for high-
resolution coastal mapping.
APPLICATION: 3-D SHORELINE EXTRACTION
A semi-automatic method is applied to extract a 3-D shoreline
from QuickBird stereo images. To extract the 3-D shoreline,
shoreline image coordinates must be obtained in both images of a
stereo pair. If performed manually, this process is very labor
intensive and time consuming. It is difficult to find conjugate
points on the shoreline in areas where the shoreline does not have
much change in shape and/or the background does not have
sufficient texture information. In this research, a semiautomatic
method was developed to solve these problems.
Figure 3. 3-D shoreline overlapping with QuickBird
orthophoto
Since QuickBird stereo images are not resampled by the vendor
using epipolar geometry, a second-order polynomial relationship
between two stereo images was set up first. Through this
relationship, the position of the conjugate point of a shoreline
vertex in one stereo image can be approximately located in the
other stereo image within a close neighborhood. The closeness of
the transformed point in this experiment is around 4 pixels in both
the x and y directions. Then, an area-based matching using
