It turns out that the camera distortion as a function of altitude/insrot (at least for the Canon MCS camera) is significantly simpler in pixels than in mm, permitting a good 5-parameter fit (dx, dy, theta, scale, scale2) whereas the transform directly to mm is more complicated, presumably due to the complexity of the pixel -> mm transform.

We should therefore try first using the fiducials to fit the distorted camera image, then transform to mm.  Please make changes in the PfiTransform code to permit this.