[Algorithms] Occlusion maps for local lights via SH?

[Aras P. <ne...@in...>]

Hi,

I was thinking on SH, ambient occlusion maps, local (eg. point) dynamic
lights, shadows and similar stuff. I have idea like this:

1. At many points (eg. vertices) of a static model, we have a spherical
visibility-distance function. It's not plain visibility function (0 or 1),
but rather a "visible distance" function.
2. Now, we approximate the function using spherical harmonics.
3. for each point light:
    3.a. for each point (eg. vertex) of a model:
        3.a.a. evaluate SH in direction "from point to light" to obtain
visible distance
        3.a.b. compare visible distance and distance to light to get
shadow/nonshadow
        3.a.c. do old normal shading (diffuse, etc.), modulate by shadow

I don't know if low-order (eg. 25 coeffs) SH approximation of
visibility-distance function will be too crude; or if all this will
outperform conventional methods (eg. shadow maps) =]

So far, I've done only very basic tests, using only 9 coefficients for SH.
It's somewhat too low - I get shadows when "deep in shadow", but also lots
of false small shadows/nonshadows. The reason for 9 coefficients was that I
could only find Cartesian expressions of SHs up to that order... Anyone has
code/expressions for higher order SHs? :)


Aras Pranckevicius aka NeARAZ
http://www.gim.ktu.lt/nesnausk/nearaz/