CIPA 2003 XIX 11 ' International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
Note that these operations are common for 3D modeling 
packages, but were previously not supported by multireso 
lution techniques. Since we address transmission over low 
bandwidth networks, our method is designed to efficiently 
represent the desired object information. By exploiting 
spatial and hierarchical coherence within the multiresolu 
tion data structure, the object assignment can be encoded 
with as little as 0.1 bits per triangle, which is negligible 
compared to the object geometry. 
(a) high resolution 
Q -J -J I sài 0 f~¡ 
C. ÛX4* -,Mwbw» , VfcM» , Con.-<tbOTi '- Dif.-ou.hBi Sn^cdJ» '. MdpUc» 
3D Model 2D Image 
rteg Cantera 11 Chao««! Г wueftarr* г te-.lue J— 
3^'ü ü'iv ' -¿Ы.дГ| 
(b) low resolution 
Figure 8: Querying meta-data about objects in the scene. 
Objects can be selected and their photographs displayed 
independent of the resolution of the rendered mesh res 
olution, as illustrated by the two screenshots with higher 
resolution (top) and lower lower resolution (bottom). 
The object identification subsystem communicates with the 
Web browser (embedding the WebCAME plugin) by re 
questing URLs to be loaded into new browser windows or 
frames. This is a simple, but very general communication 
method. It enables the access to arbitrary database content 
independent of the momentary object resolution, because 
the object ID is propagated with every triangle through the 
multiresolution hierarchy. Figure 8 demonstrates a pos 
sible application of this technique: an object is selected 
by clicking into the multiresolution mesh and the corre 
sponding photograph is loaded and displayed. The com 
munication framework also directly allows the execution 
of JavaScript programs, which gives it great flexibility and 
facilitates the design of more complex interaction modes. 
Figure 9: Northwest Heroon of Sagalassos with frieze of 
dancing girls, one of the objects is extracted from its origi 
nal location and viewed at a close distance (at high detail), 
while the non-transformed objects in the background are 
rendered at low detail. 
We will give an example how the ability to distinguish ob 
jects in the multiresolution framework also supports inter 
action: parts of the mesh can be extracted from their orig 
inal location for close inspection by the user. The level- 
of-detail selection procedure takes the transformation into 
account, so that the examined object, which is closer to the 
viewer than in its original position, is presented at a higher 
resolution than it would appear in its original, more distant 
location. In Figure 9, one of the pieces of the ’dancing girls 
frieze’ has been brought to the foreground and rotated by 
the user to study it in detail. 
We have also examined the possibility to include attribute 
data (such as normal vectors) into the CAME framework. 
While an efficient and conceptually simple encoding could 
be found (Grabner, 2003a), it turned out that the imple 
mentation of the encoder/decoder stages is a sophisticated 
task due to the different orders of he involved hierarchical 
data structures. Decoding normal vectors for the rendering 
system is thus left for future work. 
5 CONCLUSION 
We have presented a work-flow for automatic 3D model 
ing and fast visualization of artefacts recorded on archaeo 
logical excavation sites with a digital camera. A set of ro 
bust multi-view-modeling techniques developed in the past 
decade in the field of digital photogrammetry and com 
puter vision are used to reconstruct a 3D model with mini 
mal user interaction, and the CAME data structure, a newly 
developed technique for efficient transmission and visual 
ization of large models, is used for real-time viewing and 
interaction. 
The purpose of the presented work was not to develop new 
technology, but to integrate the different state-of-the-art