ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
341 
SAR filtered data for classification (Lee et al., 1994), which is 
based on the complex Wishart distribution of the multilook 
covariance matrix. Based on the maximization of the Wishart 
density function, a simple measure is derived 
where the superscript * denotes the complex conjugate, ¡s 
the m th class, C m [j) is the feature covariance matrix from 
class tn , which is computed from a training area for each class, 
Z(j) is the covariance matrix for the j th frequency band, J is 
the number of bands, and T r ¡s the trace of a matrix. Fig. 3 
shows the image after classification of SAR data. 
5. CONSTRUCTION 3D OF ELEVATED OBJECTS 
In an extraction method of elevated objects, to judge the size and 
shape of objects by providing an impression of convexity and 
concavity, the shadows of the objects are very helpful. Elevated 
objects shadow and object pixels are directly connected in the 
SAR images. After filtering and classification SAR data, the sets of 
object and shadow pixels are searched along the range direction 
of the SAR image. A simple trigonometric models (Fig. 5) and 
knowledge of the SAR geometry are used to get the relative 
heights of elevated objects (Kirscht and Carsten, 1998). Fig. 6 
and Fig. 7 show the constructed 3D with contour map and VRML 
image. 
The object height ¡ s estimated as 
'obj 
= h 
2i_ 
D: 
yiobj = y3 
V 
J h °bj ^ 
/ 
v 
H 
y\obj ph 0 bj + .Vj h 0 bj 
where Hp is the flight height. 
(8) 
(9) 
(10) 
Fig. 5 Construction of Elevated Objects 
Fig. 7 The resulting 3D height map view (VRML) 
6. CONCLUSION 
In this paper we have presented the elevation extraction in an 
urban area using airborne SAR polarimetric data with a simple 
trigonometric model. In the processing of constructing 3D image 
from the airborne SAR, speckle noise filtering, segmentation and 
classification of SAR image Data are applied. Finally a contour 
map and VRML image were given using the result of the