The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
Adaptive ATE also works for satellite images with good 
geometry (e.g, after triangulation with control points). Table 2 
shows some results. 
Sensor Type 
Accuracy 
Check # 
RMSE (meter) 
alos 
565 
11.87 
cartosat 
255 
7.36 
Eros 
127 
2.99 
ikonos 
147 
3.8 
quickbird 
95 
3.54 
worldview 
700 
9.15 
Table 2. ATE accuracy for satellite sensors 
Fig. 9 shows an IKONOS image overlapped with extracted 
points and contours. Since satellite images have much lower 
resolution, point density from ATE is also very low, and the 
performance of object filtering is not as good as on frame 
images. 
For frame images, adaptive ATE is slower than traditional ATE 
and ranges from 2 times to 16 times depending on terrain type. 
However, the quality from adaptive ATE is normally better in 
terms of point density, distribution, and accuracy. Adaptive 
ATE needs more computer time but less human time because it 
reduces overhead and post-editing work. 
Figure 9. IKONOS image overlapped with extracted points (red: 
points on objects; blue: points on ground) and 
contours, before (above) and after (below) object 
filtering 
The converge angles of some satellite stereo pairs are very 
small, so space intersection from sensor model can be very 
sensitive to small changes of parallax: at high pyramid levels, 
image coordinates of correctly-matched points are not accurate 
(at pyramid n, the uncertainty of a pixel is 2 n times bigger than 
a pixel at pyramid 0 or original resolution) , so calculated z 
values from space intersection may be far out of valid range and 
be subject to elimination as blunders. It is possible to see that