The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
1078 
longer ranges. Image-based modelling (IBM) techniques can 
produce accurate and realistic-looking models using low cost 
portable digital cameras. But they are highly interactive which 
limit the amount of details a model can have. Fully automated 
IBM methods are still unproven in real applications and require 
large number of closely spaced images, which is impractical for 
large monuments. Also occlusions and lack of textures are 
persistent problems for 3D from imaging methods. Due to all of 
the above, we decided to use a combination of technologies in 
this project: 
1. A high accuracy mid-range laser scanner for most parts 
2. A long-range laser scanner for sections on top unreachable 
by first scanner 
3. Image-based methods to fill gaps in hard to access areas 
4. Images from a balloon to model the landscape and upper 
parts not captured by the above techniques 
Figure 3 outlines the data acquisition and 3D reconstruction 
steps designed for this project. The most time consuming 
operations for large complex site, are: 
1. Deciding on the next best view 
2. Registration of the multiple scans 
3. Registration of texture images with the geometric model 
4. Editing and filling holes to create a watertight model 
Developing procedures to facilitate or fully automate these 
operations is a necessity and remains an active research area. 
Terrestrial Images 
Range Sensors 
Geometric Modelling 
Integration and meshing 
Mesh simplification 
Editing & hole filling 
Detailed Textured Model 
Reference system 
Aerial Images 
Tops & Landscape 
vSfyfrl Rendering / Visualisation 
Interactive Animation / Movie 
Non-immersive / desktop Pre-set walkthrough 
immersive (V£) j Ughtk>B 
Figure 3. 3D imaging, modelling, and visualisation steps 
This paper deals with 3D modelling from range sensors only. 
Image based modelling is covered in Remondino et al., 2008. 
1.3 Previous Work 
There is a large body of work on using laser scanners for 
heritage applications. Here we focus on work related to the 
Acropolis and similar large-scale monuments. Models of some 
Acropolis structures have been created in the past few years. A 
computer animation “The Parthenon” virtually reunited this 
main Acropolis structure with its sculptures, which have been in 
various museums for over two centuries (Stumpfel et al., 2003, 
Debevec, 2005). The models were created using 3D laser 
scanning, structured light, Photogrammetry, and photometric 
stereo. The movie also used image-based rendering and inverse 
global illumination. A project on digitising the Parthenon with a 
time of flight (TOF) laser scanner at 12mm spatial resolution 
(Lundgren, 2004) was reported. Managing the resulting huge 
datasets, starting with about 7 billion raw 3D points, was 
attempted by using a volumetric approach that divide the data 
into voxels of different sizes. The highest-resolution model 
contained 87 million polygons. Extensive study of changes to 
the Erechtheion from the 16 th century to 2004, including an 
AutoCAD-based 4D model was carried out (Blomerus & Lesk, 
2007). The model was based on paintings, drawings, and photos 
from those periods. Also pertinent to our project, issues with 
detailed scanning of large marble statues were addressed 
(Levoy et al., 2000). Difficulties to digitally reconstruct large 
complex sites, particularly due to the considerable manual work, 
were identified (Beraldin et al., 2006). Thus, automating some 
steps such as registering multiple scans and texture mapping is 
highly desirable (Allen et al., 2005). Interactive visualisation 
with huge models remains a very active research area. Luebke 
et al., 2002 and Dietrich et al., 2007 cover many aspects. Aliaga 
et al., 1999 presented a system for rendering very complex 3D 
models at interactive rates. It selects a subset of the model as 
preferred viewpoints and partition the space into virtual cells. 
Each cell contains near geometry rendered using LOD and 
visibility culling, and far geometry rendered as a textured depth 
mesh. GigaWalk (Baxter et al., 2002) is a system for interactive 
walkthrough of huge environments. It combines occlusion 
culling and LOD and uses two graphics pipelines with one or 
more processors. Geo-morphing of LOD (GoLD) is a view- 
dependent real-time technique for multi-resolution models 
(Borgeat et al., 2007). It uses geo-morphing to smoothly 
interpolate between both geometric and texture patches 
composing a hierarchical LOD structure to maintain seamless 
continuity between adjacent patches. 
2. THE MAIN CHALLENGES 
In this project, several challenges were encountered. Data 
acquisition, processing, and visualisation, all had problems 
related to the size, complexity, and material of the monument. 
2.1 Data Acquisition: 
The size, setting, and the monument surface created several 
problems. The height made coverage from ground level difficult 
on top parts. Some problems due to obstructions and terrain 
(figure 4) caused delays and resulted in missed areas. Some 
parts shape complexity caused self-occlusions, and 
impediments from plants/trees created holes in the coverage. 
Figure 4. Examples of difficult on-site scanner setting 
Due to many restorations, the monument marbles varied in age 
and amount of dirt deposits on surface. Laser spot scattering 
from marble crystals causes increase in noise while apparent