stanbul 2004 
considering 
wing criteria 
concept of 
xed distance 
ered on the 
on wouldn't 
intended to 
: duplicated 
an adaptive 
center is the 
> ray of this 
the length of 
1gth edge by 
je 59^ of its 
longs to the 
">, then both 
are updated. 
means non- 
d: the facets 
‘we consider 
n have data 
layers where 
other one all 
is important 
ecific tiling. 
facet groups, 
for all next 
connexity of 
ct, there is at 
vertex or an 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
  
After this first stage, the identified objects exactly and strictly 
fulfill the geometrical criterion of continuity. [It appears 
significant to increase the given definition of an object with a 
semantic dimension. The Figure 4 presents an object where the 
continuity criterion is not cnough to fulfill the semantic 
definition of a building: we have got three different objects 
instead of a single edifice! Therefore, we need to group all 
intersected objects into a single meta-object. 
Finally, two initial objects having each one at least a facet 
intersecting the other are then merged to form a single object. 
The Figure 6 presents an example of produced object. 
3.3.3.3 Baseline detection 
The objects defined at the previous stage must be merged into a 
traditional DEM. This fusion step requires the detection of the 
part of each object that must be geometrically in contact with 
the DEM. We call baseline the part of the object that we should 
merge into the DEM in order to increase its accuracy. 
The detection stage of the object baseline has to identify: 
e The main set of vertices identifying all local minima 
according to the ‘y’ (vertical) axis, 
e All edges connecting these vertices. 
  
Figure 7. Detected baseline for example object 
A vertex is a local minimum according to the ‘y’ axis if we 
cannot found any other point in any other facet for which its ‘y’ 
coordinates is lower than the considered vertex. 
The whole set of edges connecting these vertices constitutes the 
baseline of the considered object. The Figure 7 represents the 
baseline extracted from the example object. 
The constrained Delaunay algorithm that we use makes it 
possible to guarantee the respect of contact edges between the 
ground and the model (Mostafavi, 2003). 
3.3.3.4 Arch suppression 
The arches can be considered as specific interpretations of part 
of the detected baseline coming from the previous stage: the 
arch defines a concave object where the detected baseline, 
according to local minima's criteria, should neither be merged 
nor linked with the ground represented by the DEM. 
The choice to take into account or not an arch probability can 
only be taken by a human operator. Actually, it is impossible to 
automatically distinguish an arch that should not be merged, 
from an inclined baseline that should be. 
Therefore, if the detection mechanism is activated by the 
operator, it proceeds in three phases: 
* Classification of the edges constituting the baseline, 
e Detection of the intersections between classes, 
e Elimination of the identified baseline classes. 
Classification 
This stage constitutes the first step for the arches identification 
in the object baseline. 
Classification consists in gathering edges forming the baseline 
in order to define "altitude classes". An altitude class is made 
up of the baseline elements (i.c. edges set), each one covering 
an altitude range (minimum and maximum) intersected bv at 
least the altitude range of another element. 
Another way to define this intersection is to consider the 
projection on the 'y' axis of each baseline edge. These 
projections produce several continuous areas on the ^y' axis. 
Any discontinuity on this projected area defines a new altitude 
class. 
Intersections ' 
Once the classification of the baseline completed. we should 
consider it projected on the ‘xz’ (horizontal) plane. For each 
altitude class, we define on this plane a single polygon 
enclosing all edges of the object baseline forming the given 
class. This process produces several closed areas, each one 
covering distinct altitude ranges. 
The intersections detection is then made between each polygon. 
Deletion of arch classes 
For all the edge classes identified at the previous step. we 
determine their level i.e. the lower points of the class 
considering a minimum ‘y’ criteria. We choose in the class set 
all intersected edge groups that do not have the local minimum 
level. Finally, we remove all edges of the identified classes from 
the list constituting the baseline. 
This process is applied for all edges defining each baseline of 
every used object. 
The Figure 8 shows the final baseline built for the given object 
example. 
   
Figure 8. Final baseline for the object 
3.3.3.5 Tiling process 
Before being modified by the various objects of the 3D model, 
the DEM over which we wish to navigate is meshed. This 
allows us to handle it like a simple 3D object. 
Once the various elements merged into the DEM. we have a 
single 3D model describing the whole scene, ground and 
surface. 
In order to guarantee a freely virtual navigation. the improved 
3D model is cut out in geometrical tiles (see Figure 9). These 
tiles are saved separately in order to be dynamically loaded 
during the virtual flight.