recent
d cost.
ion of
tion of
is and
rilding
'eas of
ure of
> areas
atures
aerial
] "city
extract
fferent
| and à
phs is
y fine
“TION
| large-
alysis,
nple to
RGB
uences
mation
Mohammad Saadat Seresht
DSM. Height information is used for automatically recognition of approximate location and region of
buildings. After that, we know each recognized region include only one building that with assistance of
image data, details of building can be extracted and modeled.
3D modeling. Buildings are modeled three dimensionally in this method that is more complicate but usable
for more various types of buildings. In this method, feature matching and grouping with help of topology
rules and Euler operators are done three dimensionally in model space instead of two dimensionally in
image space. Therefore we can reconstruct complicate building roofs by extraction and setting of 3D faces.
Shadows. Analyzing of image shadows are used to building recognition.
Operator interaction. In semi-automatic methods that are more operational, operator interaction for
building recognition and reconstruction causes to improvement of production reliability and reconstruction
of more types of buildings. Usually in this method, first, a set of building models has been grouped.
Operator after pointing to each building on images and related model in generic models prepares work for
computer interaction. At last, result is compared with fact and then reconstructed building is succeeded or
improved by operator.
Special building models. This method enables us to reconstruct buildings with specific shapes such as
gable, pyramidal or hemispherical roofs. In this method, we can reconstruct buildings after introducing of
roof model and related parameters.
Flat building roofs. Edges of building about are extracted in this method (for example with existence
maps) and only with one height point on its roof a 3D representation of building is created. Note to be that
flatness of building roofs is default.
Rectangular shaped buildings. In this method a rectangle is fitted to building plan based on least square
estimation and 2D modeling is done in image space.
3 COARSE RECOGNITION OF BUILDING REGIONS
Basically, for automatically recognition of building regions in images, have to be used altimetric
information in addition to planimetric information, because singly using of images can not completely lead
us for building recognition. It is due to occluded areas, types of texture and shadow, low contrast and
complexity of image regions. Also DSM is made from stereo image matching process that is selfsame of
DTM plus all features on and near surface of the earth. Therefore with help of DSM we can partly inhibit
above problems.
The first step in coarse recognition of building regions is detachment of 3D local ridges regions of DSM.
Then some filters are applied on selected regions based on their shape, size and content to reliability on
existence of a building in regions is raised and building hypotheses are made.
There are five methods for 3D-region segmentation:
Smoothed ridges extraction
3D-edges extraction of DSM
DTM from DSM differentiation
DSM classification with contour lines that is alleged MHB (multiple height bins)
DSM classification with region filling
Recognized ridges in DSM can be related to various 3D features. In order to decrease of mistakes in coarse
recognition of building areas, deletion of absurd ridges and problems are not occurred in recognition of
building details, a number of tests are needed to applied on these areas. Therefore default in step of fine
recognition is existence of one building at each specific region. A number of tests that are applied this step
due to 3D regions filtering are flowing:
Object space filtering includes:
Minimum height above the earth surface
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 793