Eberhard Steinle 
  
The differences between these results are quite obvious. Using first pulse mode, the building contour lines approximate 
quite well the shape of the roof planes. For last pulse mode, poorer results are obtained. The building is modelled much 
smaller and the contour lines do not fit very well to the shape of the house. This is caused by the kind of measurement, 
as shown in section 2.3. 
By means of the second test building another phenomenon can be outlined. In Figure 9 (left side) the results are 
presented for first pulse measurement without elimination of vegetation (due to experimental purposes). The example 
shows that this can cause serious problems: a tree is standing very close to the building; because the tree and the roof 
have similar elevations, parts of the tree were considered during segmentation as belonging to one of the roof planes, 
Therefore, the lower left building corner was set at the end of the tree, and not at the end of the building. 
However, as mentioned above, the modelling in this approach is based on a vegetation reduced DEM. This leads to 
another result in the reconstruction process (Figure 9, right side). In automatic processing it is important to extract 3D 
objects as building hypotheses, i.e. regions for further processing. For the second example, significant differences 
between both versions occur. Because the tree was nearly eliminated, no building hypotheses has been established at 
this position (see Figure 7). Therefore, the plane extraction process did not include the tree canopy and the roof plane 
was not extended outside the house borders, ie. the reconstructed house model coincides much better with 
corresponding CAD model (Figure 9, right side). 
  
  
J 
Figure 9: Reconstruction process for test building 2; left to right: first pulse DEM (not vegetation reduced), segmented 
planes (including disturbing objects), wireframe model (vertical view), same (olique view), 
segmented planes from vegetation reduced DEM, resulting wireframe model (vertical view) 
  
  
  
  
  
  
  
  
  
  
Figure 10 shows a vertical view of the superimposition of the wireframe models derived from vegetation reduced first 
pulse and last pulse DEM. The latter is again much smaller than the other one, but contains an additional small plane in 
the upper left part. The correct shape is of type hip roof, therefore none of the data sets lead to a correct description, but 
with last pulse DEM a better coincidence concerning the shape can be achieved, not concerning the dimensions of the 
building. 
  
  
  
  
  
  
  
  
  
Figure 10: Superimposition of the reconstructed buildings as wireframe models, 
derived from first pulse (dashed lines) and last pulse mode (solid lines) 
  
864 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
[47] 
+ 
A^ Un