In modern AAA games with open environment and lot of static content (eg : crysis), how is occlusion culling performed?

Question

Two ideas i have in mind :

1) Scene is rendered to a invisible buffer, using low resolution and low polygon count models (or even using only bounding volumes like cubes or spheres). The buffer is then checked to know what is visible or not. Before rendering the low resolution scene, some frustum culling could be applied, to already remove as much objects as possible.

2) A tool is run on the static map and will perform complex (and thus slow) ray tracing to know for some 3d positions on the map what is visible and what is not. All that information is then stored in a efficient way that can be user later at runtime (eg: an octree) This solution would only work for static meshes (eg : building) not moving objects.

Answer 1

Jovan's answer is great, but just wanted to add that there is sometimes hardware support for asking the question "Did this geometry actually render any pixels?" (your idea #1) These are called occlusion queries - these is a good GPU Gems article about them here

This is pretty similar to your idea #1, except you don't need an invisible buffer, and can be more efficient or simpler than checking the buffer. The linked article does a great job going in-depth with how hardware occlusion queries work.

The Unreal Engine leverages hardware occlusion queries on platforms that support it, in addition to other techniques like view frustum culling and pre-calculating what static geometry is occluded.

Answer 2

1) Scene is rendered to a invisible buffer, using low resolution and low polygon count models (or even using only bounding volumes like cubes or spheres). The buffer is then checked to know what is visible or not. Before rendering the low resolution scene, some frustum culling could be applied, to already remove as much objects as possible.

This technique is used by Killzone, and is detailed in their presentation Practical Occlusion Culling on PS3. They also use portals for indoor areas.

One thing to keep in mind about large outdoor maps is that mesh simplification is as important as occlusion culling, so some variation of height-fields with geometry clipmapping is often used. Here is a good paper that covers culling as well. GPU Gems 2 has an implementation of it.

Answer 3

At least the first two Quake games used strategy 2; I don't know if idTech engines still use it. The reachable world is divided into convex spaces organized in a tree data structure ("binary space partition"), and for each node in the tree there's a bit vector representing which other nodes are visible from that node ("potentially visible set").