The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1143 
the z values of same feature points are in range A at pyramid 
one and range B at pyramid two, yet range A and B do not 
overlap. ATE needs to address such issues by relaxing criteria 
for z values at high pyramid levels. 
The quality of adaptive ATE on satellite images is not 
significantly better than traditional ATE, because terrain 
variation is not significant either relative to flying height of 
satellites. The speed of adaptive ATE, however, can be faster 
than traditional ATE, because traditional ATE works best for 
epipolarimages where terrain variation mainly affects x-parallax, 
yet adaptive ATE does not assume epipolarimages, and most 
satellite images are not epipolarimages either. 
3.3 ADS40 Sensors 
Leica’s airborne ADS40 sensor provides multi-spectral 
pushbroom images with three looking angles (-14°, 0, +28°), 
high resolution (up to 5cm), and 5 bands (pan, rgb, and 
infrared). Normally, forward/backward configuration gives best 
configuration and highest accuracy, but it also has biggest 
distortion and is most difficult for matching. Normally customer 
uses forward/nadir or backward/nadir pair for ATE. The current 
version of ATE is handling images pair by pair. In next version, 
we will process three looking angles at the same time for better 
reliability and consistency. 
Figure 10. Contours overlapped with ADS40 images, before 
(above) and after (below) object filtering 
ADS40 normally has long strips with tremendous data flow and 
needs to be split into small blocks for efficient memory 
handling. Certain overlap between blocks is used to prevent 
block effect, which is discontinuity on block boundary and is 
very obvious on contour map. 
Fig. 10 shows the contours from ATE, draped on an urban 
scene. The contours without object filtering reveal shapes of 
buildings, and contours after object filtering shows general 
trend of terrain. 
Fig. 11 shows a DSM (without object filtering) and DTM (with 
object filtering) from ATE on ADS40 images. Most buildings 
and trees are removed and small variation of terrain is still 
preserved. 
Figure 11. DSM (above) and DTM (below) from ADS40 
4. SUMMARY 
This paper introduced ERDAS’s adaptive ATE and associated 
blunder detection techniques. The performance of ATE on 
frame, satellite, and ADS40 images are presented and discussed. 
Generally, adaptive ATE assumes piecewise continuity. During 
matching with different image pyramid levels, ATE iteratively 
applies terrain range to limit search range and uses matched 
points to update terrain range. It is suitable for natural terrain. 
For urban scene, it needs object filtering for good performance.