When running the code offline to test the fibre identification, there appears to be an instability in the coordinate transformations, which causes issue in the calculation of cobra patrol regions.  Below are examples from the sequence starting with frame 7209200, showing the transformed position of a non-moving cobra over a sequence of frames. The transformation is stable for most frames, but shows scatter in some frames, at up to a 0.6 mm scale. In addition, there is an offset of a few tens of mm between the transformed position and the true position of the fiducial fibres. 

In addition, the production version of the code (6.1.3) produces different results than the version in the master branch on github, with the former being better.