2008 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
called 
of 
nDEM 
DTM 
like 
6 the 
nDEM 
nDEM 
DTM 
DSM 
.0 500.0 550.0 
the 
-Values 
original 
it 
undefined. Such positions get filled with also a value marking 
them as undefined in the true orthophoto. In the case an 
orthophoto pixel is seen from only one image of the stereo pair 
it is filled with this value, in the case it is visible in both stereo 
images it got filled with the average color of both pixels to 
reduce specular reflections. 
Figure 8. Pan sharpened orthophoto based on the left and right 
stereo image and the DSM from the Munich scene 
3.6 Classification 
Applying a height-threshold to the nDEM and a vegetation- 
threshold to the NDVI derived from the true orthophoto two 
binary masks are derived which allow a coarse classification 
into the classes: low level and no vegetation, low level with 
vegetation, high level and no vegetation and high level with 
vegetation. 
The “high objects mask” is derived from the nDEM by 
applying a threshold of “high” (about 4 m) as shown in Figure 
9. 
Figure 9. High objects mask calculated from the nDEM 
applying a height threshold of 4 m (section 600 m x 400 m) 
The “vegetation mask” is derived from the normalized 
difference vegetation index (NDVI) which is calculated from 
the red and near infrared channels of the pansharpened 
multispectral true orthophoto by applying a suitable 
vegetation-threshold (Figure 10). 
NDVI = ( NIR - Red ) / ( NIR + Red ) 
Figure 10. Vegetation mask based on the thresholded NDVI 
calculated from the orthophoto (600 m x 400 m) 
Combining these two binary masks leads to four classes: 
• ■ low and no vegetation: streets, plain soil,. .. 
• ■ high and no vegetation: buildings,... 
• ■ low and vegetation: meadows, grass,.. . 
• ■ high and vegetation: trees, bushes,... 
Figure 11 shows these classifications for the section used 
from the Munich scene: 
Figure 11. Classification of the Munich scene using a height 
mask derived from nDEM and a vegetation mask based on the 
NDVI from the pan-sharpened orthophotos 
3.7 Object extraction 
For extracting objects the nDEM and the orthophoto will be 
masked with one or more of the derived classes. Extracting the 
“high vegetation” class yields trees and bushes. The “high non 
vegetation” class will result mostly in man made buildings. 
Extracting all “low” objects will result in a ground plane. Each 
of these binary “one class masks” will undergo a 
morphological erosion to filter streaking effects and separate 
loosely connected parts. Afterwards the masks get separated 
into single connected areas - the derived objects. All of these 
separated objects will be fed to a process which extracts the 
outline and model parameters.