Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!purdue!decwrl!shelby!lindy!news
From: MA.JOE@forsythe.stanford.edu (Joe Bayus)
Newsgroups: comp.graphics
Subject: ERRORS IN ROTATIONS
Message-ID: <6841@lindy.Stanford.EDU>
Date: 3 Jan 90 19:03:28 GMT
Sender: news@lindy.Stanford.EDU (News Service)
Lines: 51


REGARDING: peterl@ibmpcug.co.uk (Peter Leaback)
           Errors in rotations

Peter, here are some comments, from an admitted novice in the field:

1. Might try displaying at less resolution - then 500/1000
   reconstruct might not be visible, as errors are below level of
   resolution. (might be possible  to increase size of matrix
   cells from 16-bit to 24 ????).

2. Might try calculating rate of distortion & when it becomes visible
   (via tracing and sampling) (e.g., every n rotations).  Then, re-
   construct every n-1 rotations ( this is really a variation of 1.,
   above).

3. Might also take distortion into account, designing reduced-
   perspective (object farther away) occurrences to coincide with
   object movements, and reconstructing when jumps might be credible
   ("hyperspace" leaps, for example).  This might be cute, but
   doesn't really solve your problem, does it?

4. Might try maintaining a "read-only" copy of the matrix, to which
   cumulative rotation calculations are applied to produce the
   displayed image (as an oversimplified example, 5 object rotations
   are done currently as follows:
       Mat 1 -> rotation1 -> Mat 2 -> display
       Mat 2 -> rotation2 -> Mat 3 -> display
       Mat 3 -> rotation3 -> Mat 4 -> display
       Mat 4 -> rotation4 -> Mat 5 -> display
       Mat 5 -> rotation5 -> Mat 6 -> display
   Instead, you might try something like:
       "read-only" Mat -> rot1        -> display Mat -> display
       "read-only" Mat -> rot1 + rot2 -> display Mat -> display
        etc.,
   Probably you would then have to "refresh" the read-only Mat to be
   equal to the display Mat every so often, so that the number of
   "rotN + rot(N+1)" items can be "started over", to prevent
   "animation slow-down" .  This should reduce the distortion errors
   to accumulate only at the "refresh" rate.  You'd have to play with
   this, I think, to tune for optimum speed vs image-integrity.

   If the rotation formulae are functional (e.g., logarithmic,
   hyperbolic,  parabolic, etc.), you may not have to actually add
   "rotN" items, instead pulling them iteratively from a function
   table - you MIGHT(?) be able to  do this anyway, if you know all
   the moves that will be made (good luck  for a billion-move game!!)
   and you have LOTS of memory to play with and a good algorithm for
   placing entries in the table.

Hope this is useful...these are novitiate comments by all accounts !!