Save+Back 
Hoehenmessung 
Feineinpassung 
Back 
Figure 3: Semiautomatic building extraction 
— À union of simple wireframes for a complex build- with a program for semiautomatic building extraction 
ing normally results in many edge directions. [Lang et al., 1995]. Figure 3 shows an example of the 
— The form of the roof should be taken into ac- acquisition. The interactive process requires the fol- 
) count: A hip roof has more directions than a lowing steps: 
gable roof. A flat roof is the worst case concern- 
— First the operator chooses one of the image 
patches of the building. This patch is used for 
the interaction. lt is displayed as the left image 
on the screen (cf. figure 3). 
ing this aspect. 
4 THE SETUP OF THE DATABASE 
To setup the database of 3D-wireframes we are using image — Then he has to move a wireframe model aprox- 
patches (3 to 6) of old aerial photographs with a scale of 1:12 imately at the position of the real building. For 
500. Figure 2 gives an overview. this purpose he can choose one of the building 
The following remarks should be taken into consideration: models of the left menue. The length, height and 
width are adapted in order to fit the image. 
e Step 1: Choosing buildings. The choice is done at 
an analytical plotter P3 equipped with CCD-cameras 
and is based on the old database of roof points. The 
— The absolute height of the model is calculated 
automatically by a template matching between 
So built- old control points are taken over. New ones are also theichoseniparch andvalbothiers: 
close to- chosen by selecting small building groups. The image — A robust estimation optimizes the correspon- 
possible coordinates of the roof points are measured. Then dence between the image edges in all images si- 
th many image patches covering the buildings are scanned with multaneously with the parameters of the wire- 
ough. CCD-cameras. The main results of this process are frame model. 
f finding the image patches with known interior orientation and : ; i J 
ot many image coordinates of the roof points. Approximate s, Step 4 Controlling wireframes, Finally ah wireframes 
he wire- values are computed for the new points in object space gre being controlled by letting AMOR find the models, 
and for the exterior orientation of the aerial images. he e Son pots of the quin should 
e tested have to be already measured. em 
f models e Step 2: Bundle block adjustment. The bundle block which are not found in ey images are adapted again 
rrespon- adjustment is done to achieve good reference exterior by the semiautomatic building extraction. The opera- 
OR is the orientation values and object coordinates of the roof tor then tries to improve the model, e.g. by measuring 
edge. To points. The former measurement is controlled by this more houses of a building group. 
ivable: step also. The exterior orientation is a requirement for 
of simple the next step. The result of the process is a wireframe model optimally de- 
| are not e Step 3: Measurement of the wireframes. The wire- rived from the images. The coordinate data is stored in a 
frames are measured in the digital image patches database and could be used by AMOR. The image patches 
221 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996