at appeared in 
do not remove 
Is of the points 
om one class to 
inchanged. The 
5D model was 
f the bare earth 
ations of 2.5D 
(Fig. 4) was 
1e was to detect 
o low or where 
terrestrial lidar 
ese areas. The 
to improve the 
  
  
Figure 5. Gaps in aerial lidar data. (colors according strip 
number). 
2.2 Terrestrial lidar data 
A local network was measured consisting of 13 points linked to 
Fontalba GPS point (290079002). This point was measured 
from Puig d'Estremera geodetic point (288080001) and 
Llívia GPS permanent station (284074001). 
Five sites were selected to station the terrestrial scanner in front 
of the areas showing important gaps in airborne data. The 
terrestrial lidar survey was done during two days, on September 
8^ and 9", 2003. Target reflectors were installed and their 
coordinates were measured with GPS and total station. The 
known coordinates of the targets allowed for a first 
approximation to the point cloud orientation of each scan but, 
as they were closer than the area to measure, the angular 
accuracy of this orientation was poor. In order to improve this 
preliminary orientation, surface matching was employed. A grid 
surface was computed for each terrestrial scan scene and 
another was computed from the aerial points classified as 
ground in the 2.5D model. This last surface was considered as 
the reference surface. The orientation of each terrestrial scan 
scene was adjusted to match the reference surface obtained 
from the airborne lidar points. For each terrestrial lidar point 
cloud a translation and a rotation were computed to minimise 
the distance between the corresponding grid and reference 
surfaces. This process was done with Polyworks software from 
the company Innovmetric. 
Once the orientation of the terrestrial points had been refined 
they had to be classified but the available software was not able 
to process data in almost vertical walls. The slope filter assumes 
that the terrain slope is not too high and those points that 
increase the surface slope over a certain threshold are supposed 
to belong to the vegetation. This assumption failed completely 
in this area. To circumvent this limitation a global rotation was 
applied to all the lidar points to reduce the average slope of the 
terrain. The point cloud was rotated by 30? around an axis 
approximately parallel to the railway track. After that, it was 
possible to add points to the previous set of ground points by a 
fast editing procedure using the standard tools available in 
TerraScan. The amount of available ground points in areas with 
data gaps increased and the model improved (Fig. 4). After 
editing, the inverse rotation was applied and all the points that 
   
  
  
   
  
  
  
  
  
   
  
    
   
     
   
    
    
  
  
   
    
   
  
    
  
  
  
  
  
   
  
  
  
  
  
  
  
   
   
   
  
   
   
  
   
  
   
  
   
   
  
     
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
had been classified as ground were used to build a 3D 
triangulated surface model. 
       
i d AL Uo 
2 es 2 A x nes ae 
Figure 6. Gaps covered with terrestrial lidar data. 
A dynamic survey was also carried out with the terrestrial laser 
to acquire some additional information about the rail path. Data 
was captured with the terrestrial LIDAR instrument integrated 
in the GeoMóbil, a Land Based Mobile Mapping System 
(Talaya et al. 2004a and Talaya et al. 2004b). The GeoMobil 
was mounted in a train platform that was driven by the train. In 
order to collect different parts of the track various paths were 
completed with the scanner mounted in different orientations. 
The GeoMobil system includes GPS/IMU sensors for the direct 
orientation of the terrestrial laser scanner and of two digital 
frame cameras. As the static laser campaign proved to be 
enough to fill the data gaps, the dynamic laser survey was not 
used in this project. 
3. RESULTS AND CONCLUSSIONS. 
Aerial and terrestrial lidar have been complementary in this 
project. Aerial lidar data has a high precision in height on flat 
areas and it is expected that its precision will decrease with 
slope due to the worse precision of angular measurements and 
footprint size. The usually achieved accuracy in elevation in flat 
areas is around 10 cm. In contrast, the standard deviation of the 
points in Easting and Northing was expected to be around 65 
cm. This figure was computed from the relation o-H/2000 were 
H is the height above ground according to system specifications 
from Optech. The footprint has a diameter of approximately 26 
cm from 1300 m above ground (Baltsavias, 1999). The largest 
error source for the terrestrial lidar is the footprint size. At a 
distance of 300 m, the beam divergence of 3 mrad corresponds 
to a footprint diameter of 90 cm. Angular errors are less 
important. The elevation angle is measured with a resolution of 
0.036? and the azimuth with 0.018?. At this distance, the 
precision in elevation is 9 cm while in azimuth it is twice that 
value. Precision in range is 2.5 cm. Both lidar systems had a 
better precision in the laser direction. The almost vertical 
mountain walls were scanned from sites in front of them. It is 
expected a high accuracy of terrestrial lidar because the laser 
ray direction was close to the surface normal. Combining aerial 
and terrestrial lidar it has been possible to obtain a product of 
better quality than achievable using only one of these 
techniques.