anbul 2004
ion
ror
lilt angle
30°
and 1000
*80~
00*2000™
2000.
| 2000,
00m, the
gure 4. It
500pixel is
0.05 pixels,
00
riation are
rriation of
| 640m by
05 pixels,
Patch Size
(m*m)
i 80
160
320
— (40
$— 1280
+— 1500
9— 2000
Tain
2.3.3 Tilt angle model error
Given the variation of terrain in 1000m, the errors due to tilt
angle are shown in Figure 6. It indicates that when the patch
size is smaller than 640m by 640 m, the model error is smaller
than 0.05 pixels in different tilt angles.
Tilt Angle Model Error
0.15 Patch Size
(m*m)
E
Æ 0.10
=
Se
o
E
t
Be
& 0.05
=
0.00
0 5 10 15 20 25 30
Tilt Angle (Degree)
Figure 6. SPOT5 model error for tilt angle
3. EXPERIMENTAL RESULTS
The test data include a SPOT5 and a QuickBird images. The
GCPs and check points (CHKPs) were measured from 1:1000
scale topographic maps. The position accuracy is better than 50
centimeters. The distributions of those points are shown in
Figure 7 and Figure 9 respectively. In the figures, triangles
represent the GCPs while boxes are the CHKPs. The respective
DTMs are illustrated in Figure 8 and Figure 10 respectively.
The related information is shown in Table 2.
2800000
2790900-
2770000
2760000
2740000
000
On
Figure 8. The DTM used in SPOTS orthorectification
‘International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
2772000-
2768000
2768000
2784000
2782000
2760000
2758000
2758000
2754000
28500¢ © 288000
Figure 10. The DTM used in QuickBird orthorectification
274000 276000 278000 280000 282000 284000
Table 2. Related information of test images
SPOTS QuickBird
Location Taipei, Taiwan |YingKe, Taiwan
Date 2002/07/02 2002/05/26
GSD (meter) 2.5 (Supermode) |0.6
Test Area (km*km) 60 * 60 16 * I3
Image Size (pixel*pixel) |24000*24000 . |28764*27552
Tilt Angle (degree) 14.23 12.52
Number of GCP 9 9
Number of CHKP 40 14
CCP & ,CHNF. Data 1/1000 topographic maps
source
DTM 40m Topographic Data Base of
Taiwan
589
The experiment includes three parts: (1) patch size selection for
test data, (2) accuracy analysis, and (3) computation time
analysis. In order to get the adaptive patch size, we set a
tolerance of model error to do the analysis. The real ephemeris
data and simulate terrain are.consider in the analysis. When the
patch size is determined, we checked the accuracy of the orbit
modeling and examined the accuracy for the generated
orthoimage. The difference between orbit modeling and
orthoimage is corresponding to the model error that we setup.
The computation time is also compared in point-by-point, equal
patch backprojection, patch size optimized backprojection.
3.] Patch Size Selection
Considering the differences of the covering area, the elevation
ranges for SPOTS and QuickBird images are 80m to 2000m and
80m to 600m, respectively. Given tilt angles of 14.23? and
12.52? for the two satellites, the model errors behaved as shown
in Figure 11. It is expected that larger patch size will lead to