I think a modern shadowing system should be done with shadow mapping, for many reasons. For one thing, when rendering shadow buffers you can render any geometry into it, including stuff with alpha testing, for things like trees and foliage-- you also can have sprites cast shadows. Then of course you dont need to bother with shadow volume construction, extrusions and so on.

Also it's much easier to control level of detail by switching between different resolution render targets...plus its easier to blur the shadow image for softer shadows.

In my engine i use full blown omni-directional depth buffer shadow mapping for indoors scenes with point lights, and for outdoors i use a single render target, with no depth buffer, and just render everything as black objects, then project it in my terrain and object shaders with no depth checking. This is very fast, and has the benefit of having none of those depth errors you mention.

As a side benefit, you can apply these kind of shadow projections to anything, like grass or leaf sprites, etc.

To conclude, shadow/buffer/maps are the only way to go these days, most especially for outdoor environments.

Quote:Original post by Matt Aufderheide

and for outdoors i use a single render target, with no depth buffer, and just render everything as black objects, then project it in my terrain and object shaders with no depth checking. This is very fast, and has the benefit of having none of those depth errors you mention.

Do you have screenshots? How can that be correct? If I render myself from the point of view of the light and then project that black texture onto all shadow receiving objects without doing a depth compare, I will cast a shadow on myself and the terrain, when really I should be only casting the shadow on the terrain.

Well in this case of course it only works for casting the shadows onto things which dont themselves cast shadows. in many cases this is ok, because the main thing is to shadow the terrain, or things like grass, plants, etc, which may not need to cast their own shadows.

If you want everything to cast its own shadow AND recieve shadows, yes you need to do a depth comparison.

In my outdoor renderer, i want things like trees, buildings, and characters to cast shadows onto the terrain and the grass models and plants.

My way is just simpler and faster, and looks good in many cases. Its a question of quality vs speed as always :) (but doing a simple depth buffer render for one source light like a sunlight and then the depth comp is no big deal really, so it wouldnt be a lot slower..its up to you ..)

Interesting approach. Sounds basically like the old shadow polygon method hoisted up into image space to cope with non-planar terrain. And, of course, it makes simple blurry shadows simple.

Surfaces which typically show a lot of artifacts with shadow mapping are those which are perpendicular to the light direction. This makes terrain->terrain shadowing (i.e. terrain shadowing on itself) with shadow mapping quite tricky. It is entirely possible, but usually with biasing tweaks in the lighting calc that either push or pull the lit-unlit boundary around to hide the artifacts. Object->object and object->terrain work well though.

I've found the trickiest part in terrain rendering is making the terrain shadow dynamic objects, such as characters in a valley.

T