International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004 
  
MA 
  
Figure 4: Projection of the linear SAR primitives on the optical 
image for a height of Hrmin = 3m (in green) and Hrmaz = 40m 
(in red) which are the bounds of S, for this scene. 
  
Figure 5: Projection on the optical image of the three best 
matches of linear SAR primitives. 
4 GROUND ELEVATION 
Most of the SAR primitives of S; are due to the response of 
wall/ground corners to the line detector. It means that their asso- 
ciated heights correspond to the ground height. Therefore a first 
step is the ground estimation. In a quite restrictive approach, we 
suppose that the ground is rather flat and that all the wall/ground 
corners have a relatively close height. Therefore, an height his- 
togram is built, filtered with a Gaussian kernel and its principal 
mode A, is computed. The height A; is supposed to be close to 
the mean height of the ground and is used to select the primitives 
of S] which lie on the ground using a new search step around hg 
with a more restrictive height interval (h € [hg — 6n; hg + 6n]) 
and a lower threshold th; (< thy). Figure 6 shows the match 
obtained with this approach. 
  
Figure 6: On the left the three best matches using the whole 
S, interval; on the right, the match given by the refined search 
around the ground height. 
The segments which have been matched with this refined ap- 
proach around the supposed ground height are then put in a sub- 
set S? and will not be considered in the following. They can be 
used to build a DTM for instance using a Delaunay triangulation 
using the heights of the segment extremities. 
As for the primitives of (S, U S;) \ S?, they should not lie on the 
ground. They are thus supposed to belong to internal structures 
of the buildings and should give information on their elevation. 
5 BUILDING RECONSTRUCTION 
For this preliminary study an external knowledge is used which 
is a map of the buildings present in the scene. It is used to select 
the more often detected height inside the building footprint. This 
is of course a very basic method to detect the building height but 
the aim here is to analyze the potential of SAR / optical matching 
for 3D reconstruction. 
Let us remark here that no prior knowledge about the building 
shape is introduced. Specially, the detection of the segments ly- 
ing on the ground has been done without this map. Particularly, 
it means that some segments have been falsely suppressed during 
the ground elevation estimation step if by coincidence they have 
been matched with an edge in the optical image with an associ- 
ated height close to hy: it is especially the case for rooftops with 
a periodical structure (see Figure 5). 
6 RESULT EVALUATION 
The results are evaluated using an height map which has been 
computed by classical stereovision using two optical images. The 
ground height is well evaluated with 8.5m. The results for the 
DEM reconstruction highly depend on the kind of building and 
specially on the structures lying on the roof. 
  
   
Intern 
  
Figure 
itives : 
to groi 
the rig 
For in: 
of 20n 
11.5m 
of the 
feature 
This is 
tial of: 
ence tl 
the int 
about 
and ar 
Conce 
tion: 
d 
te 
Conce