Path: utzoo!attcan!uunet!mcvax!ukc!warwick!jfid
From: jfid@warwick.UUCP (James Fidell)
Newsgroups: comp.graphics
Subject: Ray Tracing :-)
Keywords: Ray tracing,
Message-ID: <639@ubu.warwick.UUCP>
Date: 13 May 88 12:44:19 GMT
Sender: news@warwick.UUCP
Reply-To: jfid@warwick.UUCP (James Fidell)
Organization: Computer Science, Warwick University, UK
Lines: 35


For my third year project, I have implemented a ray tracer in C on a
Sun 3/60. It does constructive solid geometry on a few simple primitive
solids (spheres, cubes, pyramids, cones etc. ), objects in the scene being
specified as combinations of rotations, enlargements and translations of
these.

All arithmetic is done using doubles.  When I wrote the cone primitive I
had problems because the code which found the intersection points between a
ray (specified in vector form r = x + lambda d) and the curved surface of the
cone, would sometimes fail, or generate erroneous intersections, because
it gave solution values which were only very close to zero ( ie 10^-15 )
when, in fact, they should be exactly zero (I am sure my code is correct :-)
I know that this might happen because of rounding or some other inherent
representational error, but my knowledge of what could happen in this area
is somewhat thin.  Does anybody know of a solution to this ?  My only
work-around was to fudge the code so that values close to zero would be
assumed zero.

Another related problem - how does one guarantee that a given intersection
point is really *on* a surface, and not just a little way above or below it -
as might happen with the above case ?

Finally, does anyone know of a standard/commonly used notation for defining
scenes ?  I am thinking of a format which could perhaps be passed through a
filter to produce a suitable input format for a polygon modeller, or a
different filter to produce input for a CSG ray tracer etc.

Thanx,
James
--
  "I walk down the street,		     | uk:  jfid@uk.ac.warwick.uu
   there's no-one there,		     | usa: ...seismo!mcvax...
   though the pavement is one huge crowd"    | eur: ..mcvax!ukc!warwick!jfid
		Eric Clapton / Cream	     |      James Fidell