parallel shadows

General Discussions
#1

Hi

I would like to implement shadow mapping for directional light sources (parallel light rays). I have everything already done. But I don’t know, how to compute the screen scaling, that I put into the orthographic matrix calculation.


glOrtho( -screenScale, screenScale, -screenScale / aspect, screenScale / aspect, near, far );

Does anybody know a way to realisticly calculate this screenscale value?

Thanks in advance.

Marvin

#2

hum, isn’t it your shadow map texture size ? 256256 512512 etc…

#3

Yes. But the texture size is irrelevant. It just defines the shadow resolution (quality). I am talking about the screenScale, that defines, how much of the scene is rendered into the depth texture. Please tell me, if you need more info to understand my problem.

Marvin

#4

ok maybe I have trouble to understand the problem, can you please explain a little more ? hard for now to see what could be a good answer or way to search, I mean the ortho frustrum should be over the 3D volum you want to shadow :

bigger => lower quality for a bigger area shadowed
smaller => higher quality for a smaller area shadowed

you want it to :

fit the visible area from current camera ?

or

fit the object you want to shadow ?

in the second case you can do a projection of the object vertices (or faster surrounding sphere/box) onto the ortho camera plane and pick xmin/max ymin/max on this plane to get both x/y scale.

using the light direction vector and its 3d location, you have to choose an arbitrary perpendicular axis for y and then a x vector. using those normalized vector (x/y) compute vertice projection relative to light pos on the plane using scalar product.

I guess I am going to be wrong way again :slight_smile:

#5

Ah! This really sounds like we’re moving closer to a solution. Thank you very much.

Since noting beyond the visible space is interesting for the shadowing algorithm, I guess, I want the shadow to fit the visible area from current camera. Since I have a directional light, I don’t have a light position, but just a light direction. Do you have a solution for your first case, too?

Marvin

#6

True, you dont have light pos so just use the center of the shadow caster or one of its vertices.

Current camera visible volume is a Hexahedron (thk google:) ) that can be used as an object witch 8 vertices same as second solution.

Using the camera frustrum/visible volume you can compute eight points and project those one on your ortho/light plane.

you can also before that, clip it using the max z for the rendered frame.

maybe :

  • compute the eight points of the camera frustum culling volume
  • compute your light ortho plane equation using light vector and one origin (one vertice of shadowed or shadow caster object)
  • compute x/y vector lying on the ortho plane perpendicular from each other
  • make a projection of the eight vertices from the camera culling volume to the orthos plane
  • use xmax-xmin & ymas-ymin as the scale values