International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
(Terrasolid, 2004a). As a first approach to the terrain model, a 
triangulated irregular network (TIN) was computed taking into 
account only the ground points with TerraModeler (Terrasolid, 
2004b), from the same company. As most of the computer 
programs usually used in terrain modelling TerraModeler builds 
2.5D surface models. The name 2.5D is applied in computer 
graphics to those special kinds of surfaces were each point in 
the horizontal domain has only one corresponding elevation. 
Therefore, the elevation in these surfaces is a function of the 
planimetric coordinates (x,y). 
  
Figure 1. Spike artefacts in an overhang area 
This surface model is not appropriate to represent overhang 
areas where a single (x,y) point can have three corresponding 
elevation values and in these regions characteristic spike 
artefacts appeared (Fig. 1). 
Usually, after automatic 
classification some editing is 
required to remove residual 
vegetation that the automatic 
classification has wrongly 
classified and that has been 
included in the terrain model. 
The classification algorithm 
employed by this program is 
based in a combination of the 
opening filter from 
mathematical morphology 
(Serra, 1982) and a filter 
similar to the slope filter 
(Vosselmann, 2000). The 
presence of vegetation in this 
very steep terrain confused 
the program very often and an intensive editing work was 
required. The tops of many hills had also to be checked during 
the editing phase (all those hills with a width smaller than the 
kernel size of the opening filter). 
   
Figure 2. Lidar points in 
an overhang 
With the 2.5D model an approximation to the real surface was 
done replacing the overhang areas with almost vertical walls. 
  
Figure 3. Editing of the hills. 
The surface was edited to remove the spikes that appeared in 
that areas (Fig. 1 and 2). The editing operations do not remove 
any point from the data set, only the class labels of the points 
involved in the editing operation are changed from one class to 
another and the total number of points remains unchanged. The 
editing process continued until the resulting 2.5D model was 
considered to be an acceptable representation of the bare earth 
surface (without vegetation), within the limitations of 2.5D 
surface models. This intermediate surface (Fig. 4) was 
employed for two different purposes: The first one was to detect 
the areas where the density of aerial data was too low or where 
data gaps appeared due to occlusions (Fig. 5). A terrestrial lidar 
survey campaign was carried out to cover these areas. The 
second use of the intermediate 2.5D surface was to improve the 
orientation of the terrestrial lidar data. 
  
Figure 4. Slope map of the 2.5D surface model. The arrows 
show the location of the railway track. 
Interne 
— 
  
A loca 
Fontal 
from | 
Llivia 
Five s 
of the 
terrest 
5% an 
coordi 
knowr 
appro: 
as the 
accura 
prelim 
surfac 
anothe 
groun 
the re 
scene 
from | 
cloud 
the di 
surfac 
the co 
Once 
they h 
to proi 
that tl 
increa 
to bel 
in this 
applie 
terrain 
appro» 
possib 
fast e 
Terras 
data g 
editing