4 2004 
e non- 
n? and 
which 
photos 
photo 
o 13m 
leat 1; 
| small 
21:10, 
pically 
> taken 
Cyrax 
1g data 
north- 
m. The 
fferent 
2). An 
its and 
million 
ra and 
e. 
' set up 
stations 
  
ps and 
rdinate 
ind the 
station, 
> these 
casured 
on and 
1 and 
formed 
ta was 
id the 
tely, so 
ld be 
taset I) 
ig stage 
. of the 
rder to 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV. Part B5. Istanbul 2004 
  
perform interior orientation. Further to data processing. the 
relative and absolute orientation of the two blocks of four 
photos each, were performed. The orientations were carried out 
using 10 premarked targets (black and white squares) for the 
dataset I and 9 targets for the dataset II. Several attempts were 
made for performing triangulation adjustment, so as to allow 
estimation of the parameters with sufficient accuracy. The RMS 
of the resulting coordinates was less than 13mm (Kakli 2004). 
The next stage included the digital surface model (DSM) 
extraction. The models were produced by the Triangulated 
Irregular Network (TIN) method, as this is considered one of the 
best ways of surface representation. The DSMs were produced 
by two different approaches, automatically as well as manually. 
As it was expected, the automatic approach failed to fully 
‘describe the objects’ surfaces. Therefore, the only other 
approach was the manual editing. The manual creation of DSM 
involved using large number of points and a great number of 
breaklines, so as to define the surfaces in the best way. For data 
set I, which included a complex tiled roof, two experiments 
were made: one with a large number of points (5012 points) a 
small number of breaklines and one almost exclusively with 
breaklines (which outline all rows of tiles) and very few points. 
Due to the special characteristics of the object the second 
experiment gave much better results and the final TIN. On the 
contrary, the TIN for the data set ll is composed of large amount 
of points and only few main brcaklines. The points were 
carefully selected with a relative separation of approximately 
2cm at ground scale (17315 points). It should be mentioned, that 
both DSMs display some gaps, since there were some parts of 
the objects where the stereoscopic observation was very 
difficult, almost impossible, due to the geometry of the bundles 
(c.g. the right part of the roof and the upper right part of the 
eastern facade). 
32 Laser Scanning Data Processing 
The processing of the scanned data was performed with the 
Cyclone 4.0 software. The basic processes, which were 
accomplished in the acquired point cloud, were the tasks of 
registration and geo-referencing. Registration is the critical 
process of tying single scans with their own local coordinate 
system, defined by the individual scanner location and 
orientation, into a combined scan. The specific software 
provides the capability of performing registration by two 
methods; the so-called cloud constraints and target constraints, 
or using a combination of the two methods. For data set I, a 
combined registration was performed by making use of the 
acquired 17 special targets during scanning. The final 
registration produced an RMS of 0.016m. For data set II at the 
eastern facade, there were no special targets been acquired and 
therefore, registration was based on cloud constraints. The 
registration RMS was in the order of 0.006m. 
The next stage in processing included the geo-referencing or 
transformation of the scanned data to a common coordinate 
system. It is noted that the final registered point clouds were 
georeferenced to the same coordinate system defined by the 
surveying procedure and also used in the photogrammetric 
process. In particular, geo-referencing of the data from the north 
part was performed using the special targets accompanying the 
specific instrument. The resulting RMS for the coordinates was 
less than 7mm. The geo-referencing procedure for the data of 
the eastern facade was performed using distinct points of the 
Cloud with known coordinates. The RMS of the resulted 
coordinates was less than 1mm. 
469 
Figures 3a and 3b show snapshots from the merged point clouds 
of the two data sets. Clearly, there are more gaps in the merged 
point cloud of the north part of the church. These are due to the 
restricted window size of the scanner (only 40 degrees by 40 
degrees) and the inability of setting up the scanner at longer 
distances in order to capture more details. In the same point 
cloud, there are evident the Cyrax targets used for registration 
and georeferencing purposes (in blue). Also, in Figure 3b it can 
be seen that the areas with no overlapping scans present many 
gaps in the data such as lack of features at the top roof of the 
church. A higher scanner set up would have prevented so many 
gaps in the point cloud. 
  
  
  
  
  
Figure 3a. Snapshot of the merged point cloud from the north- 
western part of the church (data set I) 
  
Figure 3b. Snapshot of the merged point cloud from the 
eastern facade of the church (data set 11) 
Finally, the TINs of the surfaces were produced automatically 
from the merged and geo-referenced point clouds. However, due 
to the large volumes of data the resulted TIN files were difficult 
to manage. It was decided to implement decimation to the TIN 
of data set I. After a number of tests using different percentages 
of decimation at the initial TIN (approximately 485000 points), 
it was chosen to use a 15% decimation (total number of points 
73000) and 50% (almost 254000 points). For the creation of 
TIN of the eastern facades no decimation was considered 
necessary (total number of points 342000). 
4. ORTHOPHOTO PRODUCTION 
The production of orthophotographs was conducted at the 
digital workstation SoftPlotter. Provided the images are already