of the electro- 
ld this model 
iplete, leading 
ns guessed in 
electro-valves 
lans. A model 
ith Pyramide. 
present in the 
ley needed to 
> obstacles for 
   
p and front 
sumptions for 
model because 
ent. This was 
nder and filters 
  
Figure 10. Retention tank 
These objects stand apart from each other in the cell. No 
viewpoint could be found to reliably connect them to nearer 
elements. The images captured on the filters and the waste tanks 
were as well quite dark. Pyramide showed however that the 
dimensions of the retention tank in the initial model needed to 
be corrected. The proportions used in this model were wrong. 
  
  
8 
Figure 11. Waste tank, heat exchanger, filters 
7.5 Cell walls and floor 
Cell walls were assumed to be orthogonal. We supposed as well 
that the elements in the cells (tanks, pipes, etc.) were following 
the main direction of the room. 
Pyramide showed that the floor had two levels. This level 
variation of the floor is not obvious to see straight from 
monocular images. However, Pyramide indicated that modelling 
could not be coherent if we did not make this assumption. 
Figure 12. Low walls separating tanks 
For retention walls we had to use hypothesis on the dimensions 
to place them in the cell. The model of these objects is however 
not very reliable since the viewpoints are not adequate. 
7.6 Modelling accuracy 
The tank inspection task to be realised requires a few centimetre 
tolerance on the model accuracy. A theoretical estimation based 
on trigonometric calculations taking into account the modelling 
methods integrated into Pyramide showed that this requirement 
is largely fulfilled. Of course, this cannot be confirmed by on- 
site measurements. However, a second hole exists in the ceiling 
of the waste room. This hole is obstructed by safety equipment. 
This second hole was used to check our estimation of the model 
accuracy. It was modelled with Pyramide from images in the 
room. The distance found between the two holes in the model 
built was 1352 mm. The distance measured in the control room 
over the waste room is 1350 mm. 
  
Figure 13. Second hole with fire hose nozzle 
7.7 Conclusion on the modelling process 
The final model was realised from 90 images captured during 
three acquisition campaigns. The global time spent for 
modelling is about 5 days. This is longer than what was 
expected. The main reasons are the multiple modifications of 
the modelling strategy. Especially, it took quite a long time to 
prove the unexpected contact between the carrier and the 
environment and afterwards to select valid viewpoints. 
Although the modelling was less interactive than we expected, 
no additional software was developed, and only the existing 
functions of Pyramide were used to build the model. 
  
Figure 14. Model of the room 
—179—