tanbul 2004
scribed in
the other
ie solution
dependent
extended
imined. In
he angular
meters for
is already
he centre
? rotations
GCPs are
eady been
d in this
ignificant
calibrated
ect to the
1(m)
).67
.74
).16
.68
on in the
> oriented
| solution
nds. The
tion. We
tal data is
points is
using the
S) This
to read
rea-based
no. The
yet to be
as: "The
is able to
age space,
ind point
tion) for
ASTER,
and SPOT. Generic modelling is based on the metadata
information about the sensor orbit." At this time this is the
only information for this model. All the following are the
conclusions from use of the model.
In the case of SPOT-HRS data the software reads the data as
SPOT-HRG, which is understandable because HRS images
are not a commercial product. Fortunately, the metadata
format for the HRS data is the same as HRG data thus the
navigation data are written and also computed correctly (The
interpolated position and the velocity vectors for the centre
line are compared with the values which are computed in UCL
model and they almost identical). Also all the inner orientation
parameters which can not be changed by the user are the same
except of one: The principal distance (focal length). The HRS
principal distance is 580mm while the HRG principal distance
is 1082mm. However, this discrepancy does not affect the
accuracy of solution nor, moreover, the accuracy of the DEM,
as will be shown later. Unfortunately it is not possible to solve
the model in a geocentric or in an inertial coordinate system.
The coordinate system should be geodetic (geographic is
possible but not convenient).
In order to check and compare the accuracy of the direct
solution of LPS orbital model with the UCL model the same
check points as in UCL model are used . The results are
shown in table 5.
ICP Dx(m) Dy(m) Dh(m)
Min -18.93 -0.15 -5.45
Max 19.39 20.51 7,35
MEAN -4.28 10.75 1.98
RMSE 9.64 10.99 3.35
Table 5. Accuracy of LPS solution
The following comments should be made:
* The UCL model provides us with almost the same
accurate results in the xy plane, but we should mention
here that in the LPS model there is an important
systematic error especially in y-axis. However it should
be mentioned that the rotation angles in UCL model are
computed using GCPs and if these GCPs are good, this
could provide a better solution.
» The rmse in heights is smaller in the LPS model.
However, having in mind that the HRG principal
distance is used instead of HRS, it is not expected to have
such good results.
* [tis not clear why we are able to obtain such a good
result with the incorrect principal distance. This may be
due to the way in which the LPS sensor model works,
for example that:
* The principal distance is used as an initial value in a
self calibration process where the correct principal
distance is computed, although it is very difficult to
do this without GCPs
* The principal distance is not used. A direct
transformation is computed between the object space
coordinate system and the image coordinate system as
it is defined in the specifications of this model.
5.3 DEM using LPS
5.3.1. Introduction. In the LPS all the strategy parameters
can be changed adaptively, which may improve the results of
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
the strategy application. Adaptive changes take place between
iterative pyramid layer processing.
In order to check the accuracy of the produced DEM the
following sources are used:
* The Ground Control Points
+ The IGN 50m DEM
+ The UCL PITKIN 30m DEM
The sensor model which was used in order to give reference to
DEMS is the direct model without any Ground Control Points.
The model is solved using the information from the metadata
file. The grid size of the DEMs is square (20mx20m) having
in mind that all the software can handle DEMs with only
square pixels. :
Figure 2. Extent of DEMs produced from HRS data
5.3.2. DEM quality. For the HRS data a lot of tests have been
made in order to define the best strategy. The DEM with the
best accuracy and detail was produced with the following
strategy:
«;- Search Size: 5 x5
* Correlation Size: 3x 3
* . Coefficient Limit: 0.85
+ Topographic Type: Mountains
« Object Type: Forest
+ - DTM Filtering: Moderate
The search size and correlation size was allowed to change
adaptively.
The general mass point quality is described by reference to a
DEM covering the west area of the IGN reference DEM
(DEMI in red borders in image 1), which covers almost half
of the area that HRS data covers. 92% of the points were of
excellent or good quality, 8% were suspicious.
5.3.3. DEM accuracy. Accuracy of DEMI. The accuracy of
the DEMI covering an area of 1509.12 sq. km is described in
table 6.
Min diff -262.39m
Max diff 286.35m
Mean diff -0.48m
RMSE 16.16m
413
Table 6. Accuracy of DEMI
The ICP accuracy information is in table 7.