International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
the improvement of the exterior orientation by a spatial resec- 
tion. Based on this information, the mapping between image 
and model can be refined and image texture can be extracted 
automatically for the visible facades. 
Within our model based refinement of direct georeferencing, the 
localization is based on the overall shape of the buildings. This 
shape is of course only depicted, if a sufficient distance between 
camera station and the depicted building is available. Despite 
these limitations, this technique could be used as one compo- 
nent of a hybrid system for the area covering provision of ori- 
ented images. Based on complete sequences of images the rela- 
tive camera motion can be reconstructed (M. Pollefeys et al 
2000). Within such a sequence, the availability of the exterior 
orientation is only required for some images in order to relate 
the complete trajectory to the object coordinate system, which is 
defined by the building models. One application based on such 
image sequences and existing building models is the elimination 
of occlusions during facade texturing (Bóhm 2004). 
3.2 Automatic texture selection 
If a building model is depicted in multiple images, an automatic 
selection of the scene, which is most suitable to provide the re- 
quired texture information is feasible. In principle, this selection 
can be performed independently for each face of the building. 
  
Figure 8: Wire-frame model mapped to different images 
  
Figure 8 exemplarily demonstrates this situation by mapping a 
wire-frame model of a building to two different images. Based 
on this transformation, for each image the visibility and the cor- 
responding area can be computed for every facade of the 3D 
building model. If this information is put to a database, the most 
suitable surface texture, which provides the highest available 
568 
resolution can be selected during the visualisation process. The 
result of this step is depicted in Figure 9. 
  
Figure 9: Automatic texture selection from multiple images. 
Currently, the surface texture is only selected depending on the 
orientations and distances of the single building facades in rela- 
tion to the camera stations. In our future work, we are aiming on 
the selection of texture depending on the actual virtual view- 
point. For this reason, during visualisation texture will be se- 
lected from camera stations, which are as close as possible to 
the respective virtual viewpoint. By these means distortions re- 
sulting from the displacement of image texture i.e. due to pro- 
trusions or balconies, which are not modelled geometrically in 
the available building model can be minimised. 
4. MODEL BASED REGISTRATION OF 
TERRESTRIAL LIDAR 
For applications aiming on the highest visual quality the provi- 
sion of facade texture for existing building models might not be 
sufficient. In that case the geometry of the available building 
model has to be improved i.e. by terrestrial LIDAR. By these 
means a precise measurement of 3D point clouds at the facades 
of the building is feasible, which allows for an refinement of the 
model in the following steps. 
Figure 10: Terrestrial LIDAR measurement registrated to exist- 
ing CAD model. 
During data collection, the complete coverage of complex ob- 
jects like buildings can only be guaranteed, if data from differ- 
Int 
ent 
fro 
vid 
wh 
Op: 
tar 
loc 
for 
alte 
poi 
dat 
fea 
cor 
mo 
tec) 
dat 
If € 
airl 
Vis 
fine 
eva 
by 
pro 
was 
sets 
img 
imp 
pro 
Bal 
frac 
(111) 
Bos 
Geo 
11(¢ 
Bok 
text 
Friil 
Mer 
Gra 
Haa 
tion 
meti 
Hof 
Con 
ings 
pp. 1 
M. I 
Autc 
ages 
55(4