In practise, one of two approaches 10 orthorectification was
applied by contractors, i.c. space resection using cither a
rigorous model or the RPC-based model with refinement by a
few (usually) GCPs. Both approaches were used also at JRC.
As software, Erdas Imagine, PCI Geomatica, Socet Set
orthorectification modules were used by contractors, and PCI
Geomatica and Erdas Imagine for the work done at JRC.
Figure 1: The VHR satellite imagery sites
4.0 Geometric quality assessment
[n order to have both results (i.e. delivered by contractors and
produced in JRC) comparable the uniform procedure for
geometric evaluation of orthoimage products was applied:
* common guidelines, recommendations for all ortho
producers,
+ final accuracy check by one user in one software
environment,
* independent check points (not GCPs) used, according to a
standardized protocol.
The geometric accuracy of each orthorectified product was
checked in a customised ArcView based application. Two types
of output reports were generated from the application:
- a text report from checking procedure listing the check
point X,Y discrepancies and final RMSE as ID (ie.
separately for X and Y) and 2D (overall for XY),
* a diagram report form, which illustrates the vector
discrepancies, useful in detailed considerations.
5. RESULTS AND ANALYSIS
5.1 Ikonos
The majoritv of Ikonos images were corrected within the
specification of 2.5m RMSE;p. The scatterplot of RMSEx,y
(Figure 2) shows that results based on GCPs taken from the
maps are more heterogenous when compared with those based
on GPS measurements. However, there is need for further
evidence to conclude definitively the reason of such error value
distribution. It can be noted from Figure 3 that the highest
values of RMSE belong to the same site (Koza). For this site the
control points were taken from orthophotomap (1:5,000) and
DTM has resolution of 40m which might be not enough for
some parts of terrain with more heterogeneous relief. A 1:5,000
scale orthophotomap was also source of GCPs for the ‘Xant’
site, where the DTM has resolution of 30m.
There is also significant difference between RMSEx or y
values for ‘Koza’ (not present for ‘Xant’) that may be caused
by a probably poor DTM for this site, taking into account the
scanning direction (west to east, which corresponds to y) for
both sites.
For the rest of sites the RMSE values present acceptable
levels, i.e. 2.5m and better, including those with image View
102
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
A
angles up to 20.15°. Further analysis of the Figure 3, in which
the sites are ordered according to increasing values of image
view angle, shows that for this data set there are no clear
relationships between the RMSE values and view angle change.
There is also no clear evidence that RMSE values may be
sensitive to number of GCPs or to the variation in the site
elevation values (an approximate relief descriptor).
5.2 QuickBird
All the imagery checked met the 2.5m RMSE;p specification.
The RMSEy,y values presented on Figure 4 are very
homogenous and quite low. This may however be due to the
better quality (compared to many Ikonos sites) of ancillary data,
supported through the analysis of Figure 5, in which the sites
are ordered according to increasing values of image view angle,
RMSE values generally seem to be not correlated with view
angle change, or at least it is too weak to be detected by visual
inspection. Some disproportion in value of RMSE for x and y is
visible for ‘Lefk’, ‘Pige’, and *Cson' sites, but assuming the
good quality of GCPs measured by GPS the reason can be
rather related to DTM itself or its georeferencing.
For ‘Keda’ and ‘Krui’ (almost flat sites, no DTM), average
heights were used for modelling; nevertheless the results are
very good. It can also be noted that for these sites and image
view angle range, neither the number of GCPs nor relief
characteristics have a significant influence on the RMSE
values.
3 EROS
Un
Only three sites were covered by EROS images. All
orthorectification was done with parametric modelling using
different software. The GCPs were obtained as follows: for GPS
surveyed ‘Fred’, for ‘Char’ from digital vector map (ref. scale
1:2500), and for ‘Alc3’ from orthophotomaps. For ‘Fred’
(almost flat) the corrected EROS basic scene results in RMSE
values in an acceptable range (Table 3). For *Alc3' and 'Char
sites, the so-called “vector scenes” (up to 32 km length) were
acquired. The larger scene size, more heterogeneous relief,
higher view angle together with some complexity of image
(related to asynchronous acquisition mode) limits the possibility
of achieving good geometric accuracy of orthorectified product.
Nevertheless, in part due to the image resolution (GSD =
1.8m) the results for ‘Char’ are quite good. For ‘Alc3’ the
higher values of RMSE (comparing to ‘Char’) can be explained
by the source of GCPs; for which only an orthophotomap with
2m pixel (resampled from Im) was only accessible.
SITE DEM gsd | No. of GCP | View. Angle RMSE_X| RMSE_Y
FRED | 50 (dH=10) 9 14,97 2,45 2,43
FRED | 50 (dH=10) 9 * 14,97 2:74 2,13
FRED | 50 (dH=10) 45 ** 14,97 1,37 2,23
ALC3 | 50 (dH=134) 21 21,90 4,47 3,32
CHAR | 50 (dH=180) 18 20,18 2,78 2,69
CHAR | 50 (dH=180) 18: 20,18 2,56 3,93
Table 3. RMSE yy for EROS: after DTM resolution, the dH
values mean difference between extreme values of heights
(relief) measured on GCPs and check points. Software used: *
Socet Set, ** SipOrtho, PCI Geomatica in other cases.
D
dm
Intel
——
RMSE Y
n
Figu
Add
(ph)
map
Figu
for 1
for 2
