repeated in various locations within laser scanning strips, 
internal deviations of laser data become visible. 
2. MATERIAL 
The test site in the Espoonlahti was flown with TopoSys Falcon 
in May 2003 from the altitude of 400 m resulting in more than 
10 measurements per m? (Figure 1). The data was pre-processed 
by TopoSys. Five of the strips (numbers 2, 3, 4, 5 and 6) were 
overlapping almost completely. The flight direction was almost 
from southeast to northwest for the strips 2, 4 and 6. Two strips, 
3 and 5, were flown to opposite direction. 
  
Figure 1. TopoSys Falcon laser scanner provides dense point 
sampling at the flight direction. However, there is a 
gap between scanning strings causing uncertainty in 
local  planimetric registration in  across-track 
directions. 
  
Figure 2. Thirty-nine small orientation sites cover an area of 
1500 m x 100 m. Each site is visible from five 
different laser strips. Aerial image courtesy to FM- 
Kartta Ltd. 
   
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
Thirty-nine circular test sites, with radius from 12 to 15 meters, 
were selected from the overlapping area of five laser strips. The 
sites were chosen in the way that some buildings or part of the 
buildings could be seen in each site. The buildings were 
expected to be the most robust features for relative orientation 
between the laser point clouds. Only the first pulse was used 
from the laser scanning data. 
The whole test area and small test sites can be seen in Figure 2. 
The buildings in the test areas had both saddle roofs and flat 
roofs. The size of the building varied from small one-storied 
building to high apartment houses. The orientation of many 
buildings in the test sites was unfortunately either parallel or 
perpendicular to the flight direction, which caused some 
problems when the across-track direction was inspected. 
3. METHODS 
The interactive orientation method (Rónnholm et al, 2003) was 
used to find the direct relative orientation between two laser 
point clouds. The interactive orientation method was originally 
designed to be a tool for solving direct orientation between an 
image and 3D reference data, like in the case of Figure 3. The 
reference data for orientations can be 3-D control points, 
vectors, objects or even laser point clouds, for example. 
The interactive orientation method is based on visual 
interpretation of superimposed 3-D data in the image. The 
superimposing is done using the collinearity equations 
zo UT A. rU EZ TZ x 
7 (XX Yr (PY =-Yo V+ rm (ZZ) 7 a) 
ee ar X= Todt tm fa ; 
Rh KFZ =Z 1 
  
  
where c = camera constant 
x, y = 2-D image coordinates 
X, Yo, Zo ^ coordinates of projection center 
X, Y, Z = 3-D ground point 
Xo, Yo = principle point 
Kilos 7 elements of 3-D rotation matrix 
After superimposing laser point cloud with some initial 
orientation parameter values an operator is able to see, whether 
the data is fitting correctly or not. If not, the image orientation is 
not correct. The image orientation parameters contain three 
independent shifts and rotations. With tools presented in 
Rónnholm et al. (2003), these six parameters can be 
interactively modified. After every correction, the laser point 
cloud is superimposed again in the image, with the new 
orientation parameters. The method leads to an iterative 
process, until the orientations cannot be improved any more. 
One disadvantage of the interactive method is that there is no 
automation involved. On the other hand, this is as well an 
advance, because human intelligence can understand and handle 
quite complex data sets. For example, there is no need to filter 
laser point clouds before orientations, because an operator can 
interpret and fit the entity, even if some details do not seem to 
correspond to each other. However, sometimes even small 
details, if identifiable from both data sets, can be used as a tie 
features. Actually, more important than filtering, is to improve 
visual interpretability of laser point clouds with color-coding. 
  
    
   
   
       
    
    
     
    
   
    
    
   
    
       
    
   
      
  
   
  
  
  
   
   
   
  
  
  
      
  
  
    
   
  
   
   
     
    
  
Interne 
Typica 
the car 
The in 
image 
fences 
images 
may c 
orienta 
angle a 
within 
mosaic 
of mos 
(2000) 
The in 
relative 
point c 
the situ 
image. 
to find 
cloud a 
With 
perspe« 
chosen 
referen 
perpen 
directic 
to six c 
two la: 
system 
forestry 
  
Figure . 
Compar 
small te 
referenc 
Because 
between 
shift (e. 
orientati 
differen