- 113 - 
the single scans together. Since it was not permitted to put 
anything directly onto the reliefs, adhesive tape was fixed on 
both sides, bridging over the reliefs without touching them, with 
the reflecting targets on it. Twelve scans per relief were applied 
from different directions. Despite the large number of scans, 
there were still many hidden areas in each relief (see Fig. 6). 
This was not avoidable due to the very detailed reliefs with lots 
of projecting parts. 
A rough model of each relief can be processed directly on site. 
Due to the huge amount of data (approx. 150 Mbytes per relief) 
the polygonal models have to be reduced intelligently. 
Figure 6. Part of a relief scanned with GOM ATOS II 
Postprocessing: The first goal is to create a polygonal mesh 
using the registered Mensi point cloud. This will be used as a 
frame for the reliefs’ scans, which are inserted in order to create 
a detailed and geometrically correct model of the whole 
cenotaph. The processing of the scanned data is still in progress. 
4. SOFTWARE 
Overview: Currently various software products for the 
processing of 3D point clouds are on the market. Many 
producers of scanners provide their own software (like Mensi 
and Cyra), but there are stand-alone products, too. Mensi’s 
3Dipsos and Cyra’s Cyclone were mainly developed for 
engineering purposes. Cyclone allows the fitting of primitives to 
the point cloud, like planes, cylinders or spheres. 3Dipsos even 
provides a catalogue including different pipes, steelworks, 
flanges etc. which can be fitted to the point cloud. Additional 
catalogues can be imported to enable the user to create models 
for his own special application. As a result, a rough CAD model 
can be exported to MicroStation® or AutoCAD® for the final 
treatment. 
In contrast to the field of plant engineering, the modelling of 
irregular surfaces is still making great demands on both, 
software and hardware. The central element of a 3D processing 
software for irregular surfaces is the creation of triangulated 
meshes. 3Dipsos offers two options: automatic triangulation, 
which means a real three-dimensional triangulation, and the 
triangulation by projection. The latter uses geometric reference 
surfaces like planes or cylinders, which act as projection 
surfaces for a 2.5D triangulation. In addition, 3Dipsos allows to 
use simple editing tools like filling of flat holes, smoothing, and 
the deletion or creation of single triangles. The options offered 
by Cyclone are very limited. It is not possible to create 3D 
meshes. Instead, a simple TIN mesh creation can be applied. 
Demands on 3D modelling software: The crucial factor for the 
creation of accurate 3D models is the registration process. Both, 
3Dipsos and Cyclone, allow the registration of several 
viewpoints by using identical points, which can be special 
spheres, reflecting targets (only Cyrax) or any geometric object. 
The best solution is achieved, if all scans are adjusted in one 
step. This is what the Cyclone software does: All targets with 
identical ID’s are recognised automatically and an adjustment is 
computed. Additionally, different weights can be assigned to the 
targets. Mensi’s 3Dipsos only offers the registration of two 
scans in one step with at least three targets, which can lead to 
large errors, particularly for large objects with lots of 
viewpoints. To avoid this, it is possible to import topographical 
coordinates (e.g. determined by a total station) for the identical 
points. 
Another method is the registration by using overlapping areas of 
neighboured scans. This method is divided into two steps in 
3Dipsos: Firstly, an interactive consolidation has to be applied 
by manually translating and rotating one scan with respect to the 
other scan to obtain a good relative position. After this, the 
automatic consolidation adjusts both scans with regard to 
minimised distances between both point clouds. This function is 
not included in Cyclone. 
A satisfying solution would be a combination of an identical 
point adjustment and automatic registration of all point clouds. 
After registering single scans to one point cloud, the points have 
to be edited. Outliers have to be detected and deleted. Possible 
reasons for outliers are scanning errors, due to light sources, 
dusty air or other scanning effects like errors at sharp edges or 
heavily reflecting areas. Additionally, in most cases it is 
sensible to reduce the noise of the point cloud, resulting from 
the scanner’s inaccuracy. In case of large objects with huge data 
rates, the number of points should be reduced intelligently, 
depending on the object’s surface. 
If the point cloud is prepared sufficiently, a polygonal mesh can 
be created. For this purpose it should be possible to define 
breaklines which are considered during triangulation. After that, 
the software should be able to locate remaining errors like 
spikes or holes and repair them intelligently. Holes have to be 
filled with respect to the surrounding surface. 
Another function that should be integrated in a 3D modelling 
software is the creation of NURBS surfaces. Especially objects 
with few details (like car bodies) can be described very well 
with this method. 
A problem which appeared during the work with modelling 
software at i3mainz, is the direction of the normals. After 
scanning the cave with the Cyrax scanner (see 3.1) and 
importing the point data to 3Dipsos via an ASCII file, the 
information of the scanner viewpoint got lost and the normals of 
the triangles pointed to random directions. This leads to 
wrongly orientated triangles, which cannot be seen in the 
model. An intelligent function is needed to correct this. 
During the whole process of working with scan data and 
creating meshes, it is imperative to check the quality of the data. 
Extensive quality control functions must be provided. It should 
be possible to verify the deviations between the created model 
and the original point cloud. Records must be created and 
statistical values must be calculated for every processing step 
like registration of the scans (geometry within each point cloud 
remains unchanged), reduction of noise, detection of outliers 
and so on. 
Finally, the results of modelling have to be visualised. Tools 
must be provided to put suitable textures onto the models. This 
can be artificial textures from a catalogue or natural textures