Mohammad Saadat Seresht 
AUTOMATIC BUILDING RECOGNITION 
FROM DIGITAL AERIAL IMAGES 
Mohammad Saadat Seresht, Ali Azizi 
University of Tehran, Iran 
Dept. of Geomatic-Surveying Engineering 
SAADAT @ncc.neda.net.ir 
KEYWORDS: City models, Building reconstruction, Feature extraction, Data fusion 
ABSTRACT 
Automatic building extraction from large-scale aerial images is at beginning of the way, but after recent 
developments in related techniques, it can be automated partly and hence can be economized time and cost, 
From 1988, various groups in whole world have been studied on recognition and reconstruction of 
buildings from aerial images. From working on buildings with simple and flat roofs to reconstruction of 
buildings with complicated roofs and also using of texture analysis, color analysis, shadow analysis and 
height analysis are the range of various techniques today. In this article a method of hierarchical building 
recognition is studied. 
1 INTRODUCTION 
Automatic extraction of altimetric information and linear features has been come of the main areas of 
research in digital photogrammetry since last decade. Due to the more or less deterministic nature of 
altimetric information inherent in multiple images, the extraction of this information, apart from the areas 
with low and periodic signals, has been carried out successfully. However, the extraction of linear features 
such as building, roads, etc. is much more complicated. 
This dissertation is concerned with the altimetric extraction of buildings using large scale multiple aerial 
images. This stage may be regarded as an initial step for automatic reconstruction of a so called "city 
model" for which auxiliary data such as close range photographs may also be used. In order to extract 
buildings in this city model, they should first be recognized and then reconstructed. In this article different 
methods of automatic recognition and reconstruction of buildings are implemented and compared and a 
technique for practical semi automatic recognition of buildings from large-scale aerial photographs is 
proposed. The building extraction is based on initial coarse recognition of building regions followed by fine 
recognition of building details. 
2 IMPORTANT CRITERIA FOR BUILDING RECOGNITION AND RECONSTRUCTION 
METHODS 
This section describes totally conventional strategies in building recognition and reconstruction from large 
scale aerial images. These different strategies use some techniques such as texture analysis, color analysis, 
shadow analysis and height analysis that are exerted to various range of buildings from flat and simple (0 
complicated roofs. The flowing techniques may be used by those strategies: 
Color images. Using color images instead of gray-level images in order to spectral analysis of RGB 
spectrums after image segmentation, can be leaded to more attributed data for each segment. It influences 
on performance of image matching and grouping of image segments that redounding to better automation 
of flow-work. 
  
792 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
  
DSN 
builc 
imag 
3D ı 
for n 
rules 
imag 
Shac 
Opel 
builc 
of m 
Oper 
com] 
impr 
Spec 
gable 
roof 
Flat 
maps 
flatn: 
Rect 
estim 
3 C( 
Basic 
infor 
us fo 
comp 
DTM 
abov 
The f 
Then 
exist« 
There 
Reco: 
recog 
build; 
recog 
due tc