ON 
formation 
this paper 
re masked 
, possible 
id objects 
airport of 
compared 
cnowledge 
agery on a 
eter data. 
lescription 
combining 
  
Data 
n of man- 
count, e.g. 
tions and 
epared for 
's edges of 
orders can 
materials. 
o regions, 
n-oriented 
le-filtered 
d in man- 
Ulrich Thoennessen 
  
made objects with constant height present in the scene like flat-roofed buildings. Hence, the height of one segmented 
object is set to the mean height over the segment, weighted with the intensity values [Soergel et al., 2000]. 
The derived depth map can for example be used to support an interpreter in map updating tasks. A map is compared 
with the depth map, which has to be geocoded for this purpose. A possible systematic bias of the height data, which 
might be present because of erroneous navigation data, can be corrected by incorporating tie points in the map. The 
depth map is split in two complementing depth maps by masking it with the given ground plans of the buildings in the 
map. The part containing the buildings is used to verify the buildings in the map and detect missing buildings. Large 
compact areas in the complementing depth map with significant height over ground are hints to recently built up 
buildings not represented in the map. 
First the interferometric principle is briefly recapitulated to point out the dependency of InSAR data and the signal to 
noise ratio (SNR). Then, the segmentation process in the intensity data and the smoothing of the height data are 
described. Afterwards, the incorporation of the evaluated smoothed depth data for the map updating task is suggested. 
Finally, results of the approach are presented and discussed. 
2 INTERFEROMETRIC PRINCIPLE 
In this paper we focus on interferograms derived from airborne single pass measurements. Figure 2 illustrates the basic 
principle of SAR interferometry. An aeroplane carries two SAR antennas which are displaced by a base-line B. One of 
the antennas illuminates the scene and both antennas receive the backscattered complex signals s; and s;. The 
interferogram S is calculated by a pixel by pixel complex multiplication of the two received signals: 
  
  
  
S =|s,|-|s.|-exp(jA@) (1) 
z 
The object height h can be expressed as a function of the S1 
phase difference Ag: > e 
A-r-cos(@ ae e k 
pan colt) nn (2) E id 
27 - B . up 
with parameters distance r, wavelength A, effective baseline TA Pen 
B and depression angle 6. S2 S Fr % 
SAR interferometry makes only sense in case of significant Pre "Ne 
correlation respectively coherence between the two complex el Ns " 
SAR images. The coherence y depends on the expectation n 1e 70e 2e 
values of the signals. It can be estimated from the data by Nea te) ES 
window-based computation of the magnitude of the complex ES RS 
cross-correlation coefficient of the SAR images [Hellwich, H y m VAS 
1999] = 
h 
(3) x 
  
  
  
Figure 2: Geometry of InSAR Measurement 
In case of a single pass airborne system temporal and geometrical contributions to the coherence coefficient can be 
neglected compared to the influence of noise. Assuming additive thermal noise, the complex signals s; can be modelled 
as consisting of a correlation part c and noise part 7; 
s, =c+n, (4) 
The absolute value of ycan be expressed as a function of the SNR [Rodriguez & Martin, 1992]: 
A 
  
  
  
1 ) lc 
pz with SNR = — (5) 
1+—— " 
SNR 
If no phase-unwrapping errors occur, the standard deviation o; of the height measurement is: 
A-r 
GE mu (6) 
2z:B-A SNR 
Thus, the height accuracy declines with decreasing SNR. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 329