ıl 2004 
  
  
ble in a 
vhether 
iods of 
vith an 
wed by 
oration 
within 
se than 
ing a 3 
:hieved 
10d of 
ility of 
  
| ability 
ccuracy 
subject 
thin the 
nmetric 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
  
monitoring (Fraser 2001) provided retro-reflective targets are 
used. In this case, following tests on glue permanence, it was 
possible to use standard engineering grade retro-reflective 
targets 2mm in diameter in areas of bare wood and to use 
"natural points", with an acceptance that the precision of these 
would be lower, in painted regions. 
A multi-photo network design could then be instigated based 
upon the constraints of the conservation studio, the requirement 
to monitor both sizes of the Retable and the imaging properties 
of a Kodak DCS460 camera fitted with 28 and 24mm lenses. 
The chosen network design is shown in figure 4. The datum for 
the coordinate system was defined by selection of key retro 
targeted points on the structure, with scale defined by calibrated 
invar metrology scale bars. A self calibrating bundle adjustment 
was used to ensure rigorous data processing. 
  
Figure 4: Photogrammetric network — black points are natural 
features, red points are retro-targets. 
  
  
  
  
  
  
  
  
Number of target image observations 14,275 
Number of exposures in the network 108 
Number of targets and natural features 483 
RMS image residual 0.84um 
Mean precision of target coordinates 20 um 
Worst case target precision 90 um 
Relative precision for the network | : 202,000 
  
  
  
Table 1: Typical adjustment data 
The series of bundle adjustment outputs consisting of the XYZ 
coordinates of targets and natural features, the exterior 
orientation of each photo in the network, the interior orientation 
(calibration) of each camera in the network and their associated 
variance / covariance are described with respect to the 
coordinate system datum. This data must be processed to 
determine which (if any) points have moved significantly and 
then to present any movement in an appropriate manner to the 
art conservator. A valid test of movement requires that the 
datum of the object coordinate system must be based only on 
3D points which have not moved. The method chosen utilises 
the differences between object co-ordinates at each epoch in 
conjunction with their associated covariance matrices. Full 
details are given in (Robson 1995), but in essence the method 
proceeds according to the following steps: 
* An initial common datum is provided by the method 
of inner constraints including all target coordinates in 
the constraint matrix. 
* A global congruency test is carried out to determine if 
there is significant movement between the two epochs 
e Any unstable points are removed one by one. Each 
time a point is removed the datum will change and is 
redefined by a localisation procedure which 
transforms both the targets and their associated 
covariance matrices to the new datum 
® A local congruency test is applied after the removal of 
each point 
2.2 Results 
Monitoring whilst upright indicated that the Retable warped, to 
the order of +/-1.5mm, but did not twist. Dimensional changes 
in the plane of the Retable were found to range between -0.9 
and 0.7mm. Most movement was found to occur in the end 
panels which were less mechanically constrained (Figure 5). 
Significantly there were no identifiable disparate movements 
between points located either side of joins in the horizontal 
supporting boards. 
    
      
ARE A AY Ay HE ST i A OUEN sa i ANS N A 
    
Figure 5: Epochs 3 to 4 (Retable in upright position in 
supporting cradle). 
A horizontal support could therefore be constructed that would 
accommodate the degree of deformation that the Retable would 
encounter due to change in orientation and changes in humidity. 
The support distributed tensions so that new cracks should not 
nucleate and existing cracks should not significantly propagate. 
Comparison of several photographic surveys of the Retable, 
undertaken during its treatment, indicated that no visible 
damage has been caused by its change of orientation. 
Figure 6 demonstrates changes occurring in laying the Retable 
on its rear surface. After 1 week, settling of the order of 1.6mm 
has occurred in the centre whilst the outside edges have lifted 
by up to 1.4mm. In this case the level of detectable movement 
was 0.13mm 
  
m m DOTT M un 
         
Figure 6: Deformation between Epochs 4 and 5 (Retable rotated 
into a horizontal plane) 
2.3 Photogrammetric monitoring summary 
Multi-photo photogrammetric monitoring has proven a useful 
tool in the periodic monitoring of the Retable giving the