Markus Niederöst 
  
2 DATA 
2.1 Image data 
The used test region is the village Hedingen south of Zürich. The color images of one stereo model (mean flight heigl 
above ground 4'800 m, focal length - 300 mm) were scanned with a resolution of 28 microns. At an image scale of 
1:15'800 this resulted in a ground resolution of 0.45 m. 
2.2 Digital terrain model 
The digital terrain model (DTM) was provided by the Swiss Federal Institute of Topography (L+T). This so-calle 
DHM35 is derived from the contour lines of topographic maps (scale 1:25’000) through interpolation. It is available for 
the whole area of Switzerland with a rasterwidth of 25 m. The accuracy of the DHMO5 is around 1.5 m for the Swiss 
Plateau and approximatively 5 to 8 m for the alpine area. 
2.3 Digital surface model 
Commercial software (Phodis by Zeiss) was used to orient the stereomodel and to generate a dig- 
ital surface model (DSM). Several tests have been made to find an optimum parameter set which 
preserves the surface shape including buildings as accurate and detailed as possible (Fig. 1). A 
comparison of DSMs derived from 14 pum images and from 28 jum images and the choice of a 
rasterwidth of 2 m or 1 m respectively showed no significant difference of the results. The better 
resolution and smaller gridwidth even resulted in disturbing small details and height errors. For 
reasons of saving processing time and disk space the 28 ptm images have been used and a DSM 
rasterwidth of 2 m was chosen. In addition not the whole overlapping region was used but the 
test area to be processed was reduced to an extent of 500 m x 440 m. 
  
  
  
  
Fig. 1: Part of DSM 
2.4 Orthophoto 
A color orthophoto of the test area (Fig. 3) with ground resolution 0.25 m was calculated using one image of th: 
stereomodel and the previously produced DSM. The image data for multichannel classification and house reconstruction 
was derived from this orthophoto. 
2.5 Approximate 2-D vector data 
The vector data set (VECTOR25) was produced at L+T by semi-automatic vectorization from digital maps 1:25°000 
Characteristics of the initial vector data relevant for this project are: 
* Level of detail is according to the content of the 1:25'000 map (generalized) 
* 2-D (no height values) 
* New buildings are not yet included 
* Old buildings which don't exist anymore are still included 
3 DETERMINATION OF APPROXIMATE BUILDING LOCATIONS 
The initial location of buildings is determined both by approximate vector data as well as by two procedures for building 
detection. The detection is essential because houses that are not yet included in the vector data set have to be added. Al 
three resulting data sets - VECTOR25, result from blob detection, classification result - are used as initial data for the 
building reconstruction. 
3.1 Vector data 
Each building in the approximation vector data set provides one building approximation. 
3.2 Blob detection 
The blob detection from height data is done in 3 steps carried out either using the DSM or the normalized DSM (differ 
ence between DSM and DHM25) as input data. The existence of the DHM25 (see '2.2 Digital terrain model") would als 
  
636 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
allow sin 
also appl 
3231 € 
the form 
height bi 
The chos 
tion of he 
lap betw« 
steps cau 
respectivi 
nate lies 
value is 0 
3.2.2 Fir 
group of 
detected : 
well as tl 
one list o: 
3.2.3 Tr: 
elements 
bin. If thu 
centres oi 
rectangle 
Main pro 
standing « 
The blob « 
3.3 Mul 
The two « 
(brightnes 
several ch 
RGB colo 
«Ch 
«Ch 
* Ch 
8n 
Ch: 
his 
wh 
+ Ch: 
All chann 
described 
* Rer 
+. Un: 
cha 
* Um 
veg 
exc 
* “Pos