ua-
sin
Cts
:cts
ese
nto
eas
the
can
; of
of
he
ned
ital
ion
ni-
hat
th
age
ala,
n't
mi-
Recognition of projecting objects
Determination of orthoimage areas
affected by projecting objects
|
Replacement of the affected areas
Figure 3: Enhancement of orthoimage texture
fitting the position of 3D objects interactively and fixing the
shape automatically afterwards).
4.2. Determination of Areas Affected by 3D Objects
The areas of the orthoimage affected by the projecting
objects are geometrically restricted by their polygons lying
on the DEM surface. The polygons can be calculated using
ray-tracing. The rays are defined by the center of projection
in the case of displacements or the sun position in the case
of shadows and the outlines of the 3D objects.
As the shape of the 3D objects, especially of trees or hedges,
is usually not known well enough the polygons of shadows
or displacements will not be geometrically exact. Therefore
combining the ray-tracing method with image processing
methods like region growing or edge detection is suggested.
The profit of image processing are geometrically exact
polygons. The problem is the determination of the right
polygons. One way to solve it for the case of shadows is to
illuminate a automatically derived DEM with the given
sun-angle, to segment the shadowed areas by thresholding
and to use this regions as seeds for region growing in the
image (Eckstein/Steger, 1996).
4.3. Replacement of the Affected Areas
After the derivation of their outlines, shadowed or occluded
regions are replaced by more or less situation-independent
image information. For the case of occlusion this informa-
tion is taken from another image of the stripe or block,
where the occluded area is visible. To fill formerly shad-
975
owed areas, information from adjacent regions of the same
object is used.
Orthoimage pixels that are both shadowed and occluded can
not be replaced using pixels of another image, because the
shadow will cover the same region considering a short time
delay between capturing of the two aerial photographs.
Thus first occlusions are replaced, filling shadowed areas
afterwards.
The result of this processing is called an enhanced or
situation-independent orthoimage. Note that the effects of
shading also caused by the position of the sun at the time of
exposure are not taken into account. This enhanced ortho-
image is used as texture information for all objects lying on
the DEM surface. The 3D objects have to be modeled
separately.
5. PHOTOREALISTIC PRESENTATION
OF 3D OBJECTS
A great number of different 3D objects is present in real
landscapes. Nearly everything is a projecting object, e.g.
grass, stones etc. For photorealism, the biggest and most
eyecatching objects have to be handled first. Smaller objects
are tackled later. In rural landscapes big objects are farms,
woods, hedges, trees, fields of cereals or corn.
Experiences have been made modeling farm-buildings and
trees. Mapping textures from terrestrial photographs to ob-
jects with simple geometry already results in realistic rep-
resentations of buildings (e.g. farm buildings consisting of
six plane surfaces as used for the scene in Fig. 2). In general
problems arise similar to that using digital orthoimages. For
example trees or bushes in front of the buildings occlude
parts of the walls or shadows are caused by projecting roofs
or details of the buildings like chimneys (Gruber et al,
1995). The same methods as described in section 4 can be
used to get rid of these problems.
For trees sophisticated algorithms like l-trees (Prusinkie-
wicz,1994) lead to a detailed geometrical description giving
photorealistic presentations. However, considering the
available computer graphics facilities, they are not practi-
cable for landscapes with numerous trees, hedges and
woods.
Therefore, the method of texture mapping was improved
using photographs of the vegetation (photo textures). This
method is based on the simple geometrical description of a
vertical rectangle always facing the viewing position. It is
called the billboard method and generates quite realistic
trees or avenues (see Fig. 4). The attribute of alpha-transpa-
rency is used as additional information within the texture.
Transparent pixels can be defined around the outlines of the
trees and within the foliage. For its separation from the
terrestrial photographs of trees it is only necessary to gua-
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996