Sorry fvdbergh but this is the strangest thing I've heard in a long time. Surely the camera should rotate around the sensor plane which is where the camera mount point is centered. Well this is what almost every panorama hint/help/tutorial has said. Or am I just misunderstanding you?
I would not trust those tutorials. Here is a top-view diagram to illustrate:
Point "B" is the camera sensor, and point "A" is the location of the entrance pupil. Think back to high-school physics, when we covered simple lenses. Light rays pass through the entrance pupil, indicated by the red dot, to form an inverted image behind the lens --- this image happens to be on our sensor (thick black line). In high-school physics parlance, the lens was a "thin lens", which means the entrance pupil was the centre of the lens (in all axes).
The top diagram illustrates how we want to rotate the camera when taking panorama shots (solid green lines represent rays of light illustrating the field of view of the camera. dotted green lines are the position of the rays after rotating the camera). You will notice that the rightmost blue block will remain hidden behind the left blue block, regardless of how we rotate around the red dot. This is crucial for panoramas, since this implies that a single point in the scene (say, the near corner of the first blue block) will always appear to be at the same position in our larger panorama coordinate system.
The bottom diagram illustrates what happens when you rotate around the sensor, which is of course what happens when you rotate the camera mounted directly on a tripod. Note that the entrance pupil now "orbits" around the sensor, which means that the two blue blocks now appear to move relative to one another. My diagram does not quite push the concept far enough, but you can see that the rightmost blue block is very nearly visible from the camera orientation indicated by the dotted orange lines. This is bad news for a panorama, because the rightmost blue block will now be visible in some of the panorama shots, but obscured by the front block in others, making it impossible to generate a unique view from the position of the sensor.
Like I said in my original post, this is only a problem if you have objects close enough to the camera for the parallax shift to be visible. I found this out the hard way trying to do indoor panorama shots.
The solution is actually simple. Put two tall objects on a table (say, two AA batteries), and align your camera so that you can see them lining up in a straight line like the blue blocks in my diagram. It might help if your camera is slightly elevated so you can see the top of the rear battery at all times. If you now rotate your camera around the tripod mount point, you will see the two batteries moving relative to one another. Now move your index finger along the lens barrel on the underside of the lens. Rotate your camera round the tip of your finger --- you should see less relative movement between the batteries. Continue to move your finger towards the front of the lens barrel, checking for movement between the two batteries as you go along. You should be able to find a point at which relative movement becomes imperceptible --- this is the location of the entry pupil of your lens.
If you want to take multi-row panoramas, you have to centre your rotation in two axes, which typically requires a panorama head, but they are ridiculously overpriced, even the manual versions.
I have build one from aluminium square tubing, which kind of works with a light lens and body.
Anyhow, let me know if my explanation needs some more work.
