Andrew Bibitchev 
  
4.2 Auto Building Extraction 
If we have detailed DEM produced by stereo-correlation procedure, then we can try to use the DEM for definition of 
interest areas. 
First of all DEM of the ground can be calculated with the aid of dual-rang filter [W. Eckstein, 1996]. Then, difference 
between initial DEM and filtered DEM is used to localize high objects, such as trees, buildings and so on. This allows 
us to obtain areas of interest on the images. However, there are two problems. First calculation of detailed DEM 
requires a lot of time and resources. Secondly separation of interest area for one object from interest area for another 
object is often problematic (due to existence of trees and neighbouring buildings). 
Nevertheless, as soon as we produce areas of interest we can perform steps 4 and 5 of semi-auto algorithm. 
  
  
  
aa 
  
x 2 ARR ns: M 
disse. à a. Se : AS 
     
  
  
  
  
  
  
     
  
Figure 8. Example of building reconstruction: left image of stereo-pair with extracted roof edges (at the left), right 
image of stereo-pair with extracted roof edges (at the center) and reconstructed 3d model (at the right) 
5 | CONCLUSION 
It was shown that 3d features extraction can be performed via maximization of two functionals: integral intensity step 
along 2d edge and correlation between 2d edges on different images. Proposed approach is used in applications and has 
proved its advantages. Future work is supposed to be concerned with complicated models of buildings and development 
of model's pyramids. 
ACKNOWLEDGEMENTS 
I would like to thank S. Zheltov and U. Blokhinov for their valuable suggestions and supporting. 
REFERNENCES 
David M. McKeown, Jr., Chris McGlone, Steven Douglas Cochran, Yuan C. Hsieh, Michel Roux, Jefferey A. Shufelt. 
Automatic Cartographic Feature Extraction Using Photogrammetric Principles. Manual of Photogrammetry Addendum, 
Chapter 9: Feature Extraction and Object Recognition 
A. Singh, M. Shneier, 1989. Grey Level Corner Detection: A Generalization and a Robust Real Time Implementation. 
Computer Vision, Graphics, And Image Processing 51, 54-69 (1990) 
Poul S. Wu, Ming L,, 1997. Pyramid Adaptive Dynamic Hough transform to detect edges with arbitrary shapes. Opt. 
Eng. 36(5), Society of Photo-optical Instrumentation Engineers 
W. Eckstein, 1996. Segmentation and Texture Analysis. International Archives of Photogrammetry and Remote 
Sensing. Vol. XXXI, Part B3. 
  
78 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.