International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
A
A E
| — ODD
Frequency
M i
10 20 aü 40 50 50 70 &0 an 100
DN
Figure 3. Histograms of even and odd columns of the 3N band
*
2.3 Image orientation
GCPs were collected from topographic maps of scale 1:25,000.
The coordinate system used was the local map projection
(Gauss-Krueger) and the local datum (Datum Lisboa). PCI
software was configured in order to accept this reference
system. Figure 4 shows the location of the points on the 3N
image. All points were identified on both images. Heights were
interpolated from contours with 10 m contour interval.
"oy
eco
°
oN
eu
-
ol
e
eo
10 11
Figure 4. Location of the GCPs on image 3N
The root mean square (RMS) of the residuals after the bundle
adjustment in PCI software, of the residuals in these points, in x
and y directions, are listed in table 1.
Image RMS-x RMS-y
3N 0.81 0.86
3B 0.31 0.62
Table 1. RMS of residuals in image orientation with 11 points
Some experiments of reducing the number of GCPs were
carried out. Some points were kept and the others were taken as
check points (CP). Keeping the 4 points close to the image
corners (6, 8, 9, 10), very small residuals were found but for the
7 check points the RMS were large. Table 2 contains the results
obtained for both images on the GCPs and CPs.
GCPs CP
Image RMS- | RMS-y | RMS-x | RMS-
3N 0.02 0.04 1.78 1.81
3B 0.17 0.71 1.21 3.93
Table 2. RMS of residuals in image orientation with 4 GCPs
and RMS of 7 CPs
170
Some experiments with 3 GCPs only, gave very bad results, in
general with RMSs, both on GCPs and CPs, above 10 pixels.
After experimenting different combinations of GCPs it was
possible to conclude that 3 points are never enough to orientate
the image. The minimum would be always 4 points. In order to
have some redundancy for the least squares adjustment at least
some 5 or 6 will be needed.
Not much is known about internal procedures followed by the
software. The requirement of a large number of GCPs is a
limitation since the most interesting regions to apply satellite
images are those where GCP availability is more difficult.
Some alternatives would be possible, as for example, to fix the
orbital data given in the ancillary data and let only attitude to be
determined. That would be possible with 3 points only if the
initial orientation data is accurate.
Aster images are provided in the HDF format with ancillary
data, among which there is a set of simulated GCPs. For a grid
of points in image space the equations of projection rays,
obtained from the orbital and attitude data, are projected onto
the reference ellipsoid. These points are a total of 143 (13 by 11
grid) for the 3N image and 176 (16 by 11) for image 3B, and
completely cover the images.
The accuracy of these points depends on the accuracy of the
orientation data. In order to assess them they were used to
rectify the images using polynomials. Third order polynomials
presented residuals with RMS below 0.2 pixels for both images.
Once images were rectified on the local mapping system, vector
data (from 1:25,000 maps) was overlapped. Figure 5 shows
streets in the city of Porto on images 3N and 3B, respectively.
x»
ELS eh, quero ap i EU
Figure 5. Vector data superimposed on 3N and 3B images
rectified with a 3 degree polynomial.
8
The lack of coincidence depends on the inaccuracy of the orbit
and attitude data and on the height of the area. Height above the
ellipsoid in the area of the figure is of 150 meters. It can be
International .
observed that
around 35 me
meters, in opr
3B image due
there is a con
which is only
It would be pe
use of these d
possible alter
image space |
and Grodecki
2.4 DEM ex
Once images
and the DEM
resolution (3€
from 0 to 100
The matching
where the ma
extracted. Fig
can be identi
extracted, wi
gaps were fill
Figi
Failures also
uniform. Tha
image.
Fisure 7 .sl
interpolation
approximatel
There are tw«
be extracted.
Editing shou
knowledge of
In order to
extracted DE
from aerial pl
km and was
service. It is
Figure 8 show