nbul 2004 International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
TI 
ee Le 
  
um 
ate (result 
he goal of 
cal in that 
ings. This 
je to the 
limestone 
nks at the 
out 28 m. 
There are six projecting sections with five recesses. Inside these structure-from-motion approaches, which tend to use video 
recesses, the podium is only about 1 m wide, whereas the two 
outside parts project appr. 4 m, while the width at the four 
central parts is appr. 2 m. During the course of its history 
the building has been partly dismantled, repaired, and modified, 
probably because of an earlier earthquake. 
As part of these changes, the socle mouldings in the niches were 
partly removed to make room for sculptures and pedestals 
inside, which the original podium width of ca. 0.4 m would not 
have allowed. The central curved niche, crowned by a fluted 
concha, had one projecting console in its upper part from which 
the water fell in a cascade to fill a basin of 81 m^ capacity. 
Excess water flowed over the top of the basin, where it could be 
collected in jars. The larger lateral aediculae each held four 
columns in two rows, the other aediculae held only a single row 
of columns. The monolithic columns, about 3.35 m high, had 
Attic-Ionic bases and Corinthian capitals. They carried an 
entablature of architrave and frieze blocks, and a cornice. These 
six aediculae supported richly decorated pediments. The 
pediments of the lateral aediculae had a double S-shaped outline 
decorated with opposed volutes, while those on either side of 
the central niche were curved and decorated with Gorgoneia 
(Medusa heads). The second and fifth pediment were triangular 
with a similar decoration. Inside, the aediculae were roofed with 
large horizontal slabs, with coffers decorated with theatre 
masks, heads of mythological figures, or floral motifs. The 
height of the nymphaeum was ca. 7 m in the niches and 7.8 m 
in the aediculae. A total of 17 different materials (different 
types of breccia, limestone, and marble) were combined into a 
splendour of natural colors. 
The virtual reconstruction was based on 
I. expertise of the archaeological team concerning the 
architectural style of the period (the determination of 
which is supported by archaeological evidence) and of 
the region 
2. photographs of the remains of the building 
3. architectural drawings of the building, produced as part 
of the documentation of the excavations and of the 
preparation of the building's anastylosis. These were 
made with the help of a total station. 
4. 3D reconstructions of parts, using both passive and 
active techniques 
5. textures, synthesised on the basis of sample images of 
intact material 
6. comparison with contemporary monuments elsewhere 
The reconstruction intends to reflect the state of the building 
around the middle of the seventh century AD. Excavations on 
the nymphaeum started in 1994. More than 90% of the building 
elements have been found, albeit more often than not in a 
seriously damaged state. The finds hàve been documented 
through a large set of photographs and drawings. 
2.2 Shape-from-stills 
This already allowed us to build some reconstructions directly 
from the photographic material based on our ‘shape-from-stills’ 
pipeline (Van Gool, 2002). It needs as its input a series of 
sufficiently overlapping photographs, from which it 
automatically extracts the camera parameters and positions, as 
well as the 3D structure of the photographed object. It can be 
considered to be a generalisation of the self-calibrating 
data. (Pollefeys, 2002) Not only is the acquisition of stills 
typically easier, the images can have far higher resolution, 
which then translates into higher 3D model quality. The price 
that one has to pay is that wide-baseline correspondences need 
to be found, which is a much harder problem than the usual 
tracking of features between consecutive video frames. We also 
found it necessary to systematically take radial distortions into 
account. 
2.3 1-shot, structured light 
Apart from this ‘passive’ 3D extraction technique, we also used 
Eyetronics' ShapeCam (Eyetronics, 2004). This structured light 
device was used for some of the highly decorated elements, 
where the geometry could be quite intricate (e.g. for part of the 
statuary). This system is shown in fig. 4. 
  
Figure 4: Top: The ShapeCam system consists of a flash 
projecting a grid and a camera. The camera takes an image 
from a direction that is slightly different from the direction of 
projection. Bottom: A regular square pattern is projected on the 
scene, as seen in this detailed view. 3D co-ordinates are 
calculated for all the line intersections, resulting in the 
simultaneous measurement for thousands of points. 
An early version of this system had originally been developed in 
our computer vision lab in Leuven. A grid is projected onto the 
object by the use of a flash, simultaneously an image is taken, 
and from the grid's deformation in the image, a complete 
surface patch is reconstructed in 3D. In both cases — shape- 
from-stills and the ShapeCam — the apparatus is easy to carry 
around. As a matter of fact, the camera that is part of the 
ShapeCam can also be used to take the input for the shape- 
from-stills technique. From the point of view of the 
archaeologists, 3D shape extraction only requires taking images 
with a normal camera.