For citywide orthophoto generation with thousands of
buildings a database system is necessary to manage all
these data. At the Institute for Photogrammetry and
Remote Sensing TOPDB has been developed (Loitsch,
Molnar, 1991). It is a relational database system
extended with topological data types and operators. The
communication is done by an SQL subset called
TOPSQL. TOPDB is well suited for the management of
a building model and will be used for this purpose.
2.2 Building Orthophoto Generation
In contrast to conventional orthophoto techniques, this
algorithm must consider hidden surface areas. Z-Buffer
algorithms are a well known solution for this problem but
require much memory. Therefore another solution will be
proposed.
With this algorithm all objects (buildings) of the area of
interest in the DBM will be processed sequentially. The
simple data structure of the building model enables fast
segmentation of an object into triangles. Figure 4 shows
the triangle F defined by F,, F, and F, representing a part
of a house. The corresponding triangle F' in the image
coordinate system can easily be computed. The area of
F' will be rastered and stored in the building mask and in
a local bitmap with reference to the image coordinate
system. Pixels covered by F' will be filled with grey value
0 in the building mask and bitmap. The building mask will
be required for the generation of the terrain orthophoto,
whereas the small bitmap is necessary to determine
visible pixels of F'.
SR
NY
|
4)
VN
Fig. 4: Simple building model
Next all objects intersecting with pyramid FF, FR
(projection center) have to be identified. Triangle G in
Figure 4 is a face of one of these objects. If there is an
intersection of G' (image of G) with F', G' will also be
rastered and the corresponding bitmap pixel will be set to
1. Thus only visible pixels of F' have grey value 0. The
algorithm continues within a loop for all other graphic
primitives and objects and stops if all intersecting objects
are processed or all pixels of F' are set invisible. In the
first case the rectification of the visible parts of F' will be
performed. Of course rectification will only be done for
roofs and not for walls. Suitable resampling functions are
86
found in (Kraus et al, 1996). Now the algorithm
continues processing the next face (in figure 4 it is H)
until all objects are rectified.
2.3 Terrain Orthophoto Generation
Terrain orthophoto computation requires building mask,
aerial image and DTM. Building mask and aerial image
are merged into a modified aerial image with blank pixel
(grey value 0) in building areas. Therefore the original
grey values of this photo must be scaled to the range
[1,255]. The modified aerial image will be used for the
computation of an orthophoto without buildings. For this
process any conventional orthophoto software can be
applied. The combination of the resulting terrain
orthophoto and building orthophoto can be done by raster
algebra and will be discussed in the next chapter.
For an improved solution it is not necessary to modify the
aerial image. The orthophoto software uses aerial image
and building mask simultaneously. Moreover it is also
possible to use the building orthophoto as input. The
rectified terrain surface will be automatically added to the
building orthophoto.
2.4 Raster Algebra
For the final orthophoto generated from one aerial image
the terrain orthophoto and the building orthophoto have
to be combined by raster algebra. Figure 5 shows input
grey values and resulting grey values of the orthophoto.
Fig. 5: Combining terrain and building orthophoto
by raster algebra
Grey value O0 represents buildings or hidden surface
areas in the terrain orthophoto. In the building
orthophoto pixels with grey value O identify areas imaging
terrain surface. g, and g, represent rectified surfaces in
the corresponding orthophoto.
Of course orthophotos generated from one image only
might contain patches without image contents (hidden
areas) represented by grey value O in figure 5. A final
orthophoto with no blank area can be obtained by
merging overlapping orthophotos derived from different
images by raster algebra. Figure 6 shows the required
raster algebra.
Fig. 6: Mosaicking two overlapping orthophotos
by raster algebra
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
For
rule
gre)
3.1
Cur
don
ster
acq
dev
sim
ster
gre:
mac
The
pur]
the:
add
etc.
ano
usir
pur|
Sin:
con
sets
3.2
For
are:
digi
(ea
eav
isn
Thr
gec
for
des
use
veri
