Operation of the system will be as follows: 
(i) As soon as the submersible reaches the work site the cameras will 
pan around to photograph the calibration frame i.e. photos A and 
B (see Figure 5a). 
(ii) As the submersible moves around the site, photographs are taken 
of any objects of interest in the customary manner i.e. photos 
C to X (see Figure 5b). 
(iii) Upon completion of the work, the cameras will again pan around to 
re-photograph the calibration frame before the submersible leaves 
the site i.e. photos Y and Z. 
Suitable photographic processing and computing facilities already exist on 
the support ship. A suitable photograph coordinate measuring instrument, 
such as a stereocomparator, equipped with digitising system and interfaced 
to a suitable recording device will also be made available on board. 
Initially it is planned to use a Zeiss Jena Steko 1818 stereocomparator 
linked either to a digital data cassette unit or an on-board HP9845B 
computer via an MDR 2 microprocessor based display and counting unit. 
The analysis procedure is shown diagrammatically in Figure 6. Firstly, 
the calibration frame photographs (A,B,Y,Z) are measured and processed by 
the DLT method. Essentially the DLT approach is to use a projective 
transformation with an added polynomial to cater for systematic distortion. 
The polynomial allows for up to five additional parameters to be included 
in the solution. Work is in progress to establish the optimum distortion 
polynomial to cope with the underwater environment. By establishing such 
a transformation for each of the calibration frame photographs, the 
various inner orientation parameters (principal distance, principal point 
coordinates and distortion function) can be established and control point 
coordinates computed by intersection. Further, the relative orientation 
elements can be abstracted from the projective transformation parameters, 
as can the exposure station coordinates. 
The stereopairs of the desired objects of interest (photographs C to X) 
are now measured and object space points are computed directly using an 
analytical relative orientation procedure. This uses the inner 
orientation parameters and the distortion function computed by the DLT 
method and is scaled using the exposure station coordinates. 
The problems posed by incomplete stereopairs when carrying out relative 
orientation can be partially overcome by computing object space points for 
photographs C to X directly using the transformation parameters established 
for photographs A,B,Y,Z. Such photographs, where 40% of the stereo image 
may consist of water, are not uncommon. Typically the objects of interest 
are small and localised. In the event of multiple models being required to 
cover a single damaged area, then some form of external control is required 
and the problems of incomplete models resurface once more as the 
intersection method is unlikely to overcome the poor conditioning of the 
incomplete model when joining models together. 
A high degree of accuracy level determination is possible with this 
46