BTW, the most important part of this 'matchmoving' scheme was to determine, at
the beginning, what the proper POV-Ray camera ANGLE should be, to best match my
Canon camera's lens angle (at its default 'wide angle' setting, which is how I
shoot most videos.) At first, I thought this would be a simple and
straightforward
process: Take a video of some kind of subject matter that had clearly-seen
'vanishing-point lines', pick a representative still frame from it, then place
some CGI objects in the scene that 'matched' those, along with the proper
perspective (POV-Ray's Screen.inc include file was ideal for doing this.) The
process required quite a few test renders to 'zero in' on the proper matching
lens angles.

But a problem arose when doing the animated matchmoving: POV-Ray's default
perspective camera is a 'perfect pin-hole' camera. A 'real' camera is not; its
lens has spherical aberration, plainly seen at the outer edges of the image.
This discrepancy caused some visual 'sliding' of CG objects against the
background images, at the edges of the frame, as the video camera panned, tilted
and rolled. So I had to find a compromise between the two lens angles, to try
and eliminate this sliding (or at least minimize it)-- essentially by making the
POV-Ray camera lens a bit more of a 'telephoto' lens than it should be. (Meaning
that the 'perspective' of both lenses no longer matches.) But it seems that this
perspective mis-match is far less visually distracting than the (very obvious)
sliding of CG elements.

Here's something interesting: When professional CGI is done for Hollywood
movies, the original 'real' imagery-- with its spherical aberration distortion--
is first UN-distorted in software; then the CG elements are matched to that
(with a 'perfect pinhole-camera'); then the final composite imagery is
RE-distorted to add the spherical aberration back in, to all elements.

Post a reply to this message