tanbul 2004
f 250 m in
hese points
determined
| backward
1 numerator
dir looking
e it has not
itor without
)
| alternative
matched in
rithm (see
the height
tes and the
€ listed in
uficant and
MOMS-02
otting with
rous model
? produced,
ach is also
S o
) 0.2
0.3
0.2
etween the
rived from
del
(Krzystek,
ng in object
and HRS2
ctions. The
rived from
meters and
t applying
an a priory
ye than 17
ts per mesh
is obtained
h proved to
y measures
SM
srences dh’
id) and the
rs dh listed
the DSM
G data the
in moderate
the results
lition to the
m. In the
deteriorate
ince of the
this DSM
> matching.
TS N Min Mean Max RMS o
#2 — 510531 66:5 0.7 49.7 4.5 4.4
#4 — 424112 -124.5 1:4 125.7 9.3 9.2
#5 478139 -106.5 1.6 95.6 4.6 4.3
HRS/HRG data, 9 estimated CFC
#2 507413 -60.8 8.8 1019 © 101 4.9
#4 413740 -151.9 — 9.3 280.5 : 172 14.4
#5 476494 -95.5 9.8 71.5 10.9 4.7
HRS data, no bundle adjustment
Table 10: Statistics on height differences dh' [m] between thé
DSM raster points (10m grid) obtained with region
growing image matching and the reference DTM
Figure 8 shows the DSM of the entire image scene at 45 m
grid size derived with ISAE. The corresponding statistics on
the height differences dh' is listed in table 11. Without
considering the HRG data the results deteriorate about 10%,
in the mountainous terrain of TS #4 even 45%. Compared to
the results in table 10 the results are 20-25% worse. The
standard deviation for the full scene is 8-9 m.
TS N Min Mean Max RMS G
#2 25025 414 07 7551 54 53
#4 21445 069" 0.2 1073 117 "1Ys
#5 24322 .33.0" 1.9 32.5 5.4 5.1
Full scene 1576146 -143.4 1.7 1854 84 8.3
HRS/HRG data, 9 estimated CFC
#2 25628 673 1.7 1304 SO S
#4 21445 '-1352^ 41 3336 172 167
#5 24322 -1449 30 31H199 65 5.8
Full scene 1579054 -423.0 3.0 457.44 9.7 9.2
HRS data, no bundle adjustment
Table 11: Statistics on height differences dh" [m] between the
DSM raster points (45 m grid) derived from ISAE
and the reference DTM
-
Figure 8: Color coded DSM represe : a 2
and HRS2 channels at 45 m grid size. The location of the four check areas is marked in red.
451
effort.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004
5. SUMMARY AND CONCLUSION
This report describes the DSM generation using SPOTS HRS
and HRG-supermode images. Correction polynomials for the
provided look angle values of each camera (interior
orientation) are estimated by bundle block adjustment using
17 ground control points in 4 control arcas located in the 4
corners of the covered surface. The application of correction
polynomials for position and attitude (exterior orientation)
proves to be not necessary. The burídle adjustment results in
a point accuracy of 2 m in Easting, Northing and Height,
which is demonstrated by 17 independent check points,
distributed in 4 check areas.
An automated region growing image matching algorithm is
applied to generate mass points in image space, which later
are transformed into object space. Without manual editing
and/or filtering of the resulting point cloud an RMS height
error of approximately 4 m (5 m in mountainous terrain) is
obtained for 3-ray-points matched in 3 combinations (nadir-
backward, nadir-forward, backward-forward). For 3-ray-
points matched in 2 combinations and for 2-ray-points the
RMS error is worse especially in mountainous areas. It turns
out, that the point cloud of the applied matching process can
be produced automatically with sufficient density in wide
parts, but not in all parts of the images. The algorithm fails in
areas with poor image contrast and/or homogenous texture
like forests, broad streets, large agricultural areas, etc. Here
manual interaction is required in order to avoid gaps in the
point cloud and, consequently, in the produced DSM. This
part of the work flow can be rather time consuming and
therefore has been excluded from that study. In other words,
the presented results reflect the accuracy potential of SPOT-5
HRS which can be achieved by largely automatic processing
and which still can be improved to some extent by manual
tation of the entire image scene (approx. 80 km x 60 km), generated by ISAE using the HRSI
