On Mon, 5 Oct 1998 07:13:12 +0300, "Peter Popov" <pet### [at] usanet> wrote:
>
>This rises a question I've often wanted to ask and always forgot to. How
>does Lightscape work? What is radiosity tracing anyway? If it is a method of
>a completely different nature than raytracing should we expect POV to be as
>good as Lightwave in radiosity (or vice-versa: LW be as good in glasses,
>caustics, etc.)?
>
>Peter
>

The so-called "radiosity" in POV is badly named.  POV actually uses an
algorithm more properly called _distributive ray tracing_.  The radiosity
algorithm is something entirely different than what POV does.

From thermodynamics in physics, there is a problem where you have a bunch of
flat surfaces in space that all begin with an initial heat.  The heat will
then begin to transfer to the other surfaces where it will eventually reach an
equilibrium.  If all the surfaces are flat and emitting heat in all
directions, you can solve for the equilibrium with a system of linear
equations set in a matrix.   Someone thought that the same technique could be
used for 3-dimensional computer scenes, wherein light was substituted for the
heat.  Radiosity was born....

Radiosity does not trace any rays, so it tends to be enormously faster than
other algorithms. But it requires some things that aren't true to life or
helpful for modelling. 

I. All the surfaces diffuse light homogeneously in all directions.  This is
not true for glossy surfaces or mirrors, or any refracting object for that
matter.

II. Every object in the scene must be "tessalated", i.e. divided into a patch
of small triangles.  We already know about objects that this won't work on.
Blobs, fractals, implicit surfaces...the list goes on.  This tessalation
burdens memory, because it multiplies the number of objects in the scene by
unpredictable amounts.  Some scenes are even IMPOSSIBLE to tessalate on a PC,
because they require more than say, 64Mb of storage.

III. The light emmitted by any very small, bright patch will not shade objects
or make them cast shadows.  This is extremely unrealistic.  The avoid this,
you must then do a ray-tracing pass, but then you are back to tracing rays
again!  In general, if an important bright patch begins to reach the size of a
point, rather than a large luminaire, radiosity breaks down, and ray-tracing
steps in as the victor.

The bottom line is that ray-tracing and radiosity _seem_ to go hand in hand.
To each is its own solution.  We know light moves in specific directions. But
we also know that it globally illuminates scenes.  Here are the strong points
of each rendering algorithm. So....

Radiosity     ------> Global Illumination.
Ray Tracing -----> Any phenomenon involving the direction of light.

Its not to say that ray-tracing is _doomed_ in calculating indirect/global
illumination.  There are algorithms such as monte-carlo, photon mapping, etc,
that use the tracing of rays, that are arguably more powerful than radiosity
could ever hope to be.

Any questions?
---------------------------------------
Steve

Post a reply to this message