Gonz wrote:
Another way to think of it Steve, is this: (again I'm talking about
moving the whole lens/sensor combo) imaging an infinitely long parallel
assembly holding a camera perpendicular to an infinite wall with a
varying image, as you move the camera/lens along the assembly, if
magnification was a factor, eventually you would be looking at things
that were way beyond the fov of the lens, i.e. if for every 1mm you
moved the camera lens, the image moved 10mm, a magnification of 10, then
after 100meters, your lens would be looking at something 1km off the axis.
Right, I see what you're talking about now (and hence what Bob Shell was
saying earlier in this thread, I think). I agree with what you say
above, but please now think of the situation where you have *two* walls,
one some distance behind the other.
The problem with *lateral* movements is that due to parallax, despite
the fact that you're moving the camera sideways at a constant speed,
things at different distances from the lens take *different times* to
move across the field of view.
In the case of the two walls, this means that the closer one will "move"
across the sensor faster than the further one. So you can keep one of
them "stationary" by moving the sensor, but not both.
To calculate - and hence correct for - the apparent movement of one of
the two walls, you need to know:
1. How far away it is - the lens needs to know the distance to which the
lens is focussed.
2. How much its apparent motion will be magnified by the lens. A point
on any given wall will flick across the field of view much faster with a
longer lens than with a wide-angle. The faster this movement is, the
further you will have to move the sensor to correct for it. So the
stabilisation system needs to know the lens focal length.
If you want to correct for the motion without knowing these facts, you
have to move both the sensor *and* the lens.
Sheesh - far simpler just to use a tripod, I reckon ;-)
S
PS I also note that parallax implies occlusion, so if you're in a
situation where your lateral shift changes how much you can see of a
distant object, you can't correct for it by simply moving the sensor
anyway. Perhaps it's better to forget about correcting for lateral
movement entirely. Or record everything as holograms and forget about
this "lens" nonsense... ;-)