If I understand your first question correctly, the issue is that if you take images from different camera positions, the diameter is the same but the checkerboard you use for calibration changes perspective and this alters the computed diameter of the cylinder. If this is the case, you should be able to perform a projective transformation of each subsequent image to your "gold standard" image. You could use "cpselect" to manually select the control points, "fitgeotrans" to create the transformation matrix, and "imwarp" to transform the image into the desired reference field.
Another potential consideration would be to use a cylinder of known diameter for calibration in place of the checkerboard. You may have reasons to not use a cylindrical calibration object that I haven't thought of yet, but it may be worth considering.
Regarding question #2, what do you mean when you say the cylinder is tilted? Tilted in which direction and with respect to what?
