The mirror problem is relatively easy to overcome; a very crude approach
would be to trace a ray from all the corners of the mirror's bounding box.
If the sample point is invisible to all these rays, we can safely deduct
that the point is not visible to the mirror.
Multiple mirrors would still present a problem, but solving that wold get
too complex (parse time is an issue, too). Culling would have to be turned
off in that case.
I just downloaded the paper suggested by Jerry, I'll see if this provides a
more plausible solution...

I can't quite visualize a solution for the second problem. When the angle
between the camera ray and the surface normal is >90 degrees, the point is
not visible (provided that all normals point outwards). But how do I know
how far the point is from the nearest visible point?
The answer may be simple, but I don't have it. Perhaps you would describe
how you'd go about solving it?

Margus

Stephen Lavedas wrote in message <36E1DA3E.44410F86@virginia.edu>...
>I'd be happy to work with you on it.  Actually I have a few ideas.. the
>first being to add a "culling" keyword (ie you know there is a mirror
>somewhere, so you need all the hair) or perhaps to give the vector to
>the mirror and calculate off of both.  The other is to add a margin to
>the occlusion formula.  Don't occlude EXACTLY at the point where the
>point is no longer visible..occlude 5 degrees further or some such.
>
>Steve
>(of course I still need to dl the super patch)
>

Post a reply to this message