Creating a Full Stadium of People, Ready as a Game Asset (ie. not too CPU intensive)

Question

I am trying to make a boxing scene in my game. I decided to attempt to make the large stadium first (It's a boxing management game, so the scene can be 'pre-rendered' perhaps, the fight events (eg. Big hits, blocks, dodges) are going to be calculated based upon the boxers skill values and some luck etc.

I have this scene in Maya:

My first idea was to 'scatter' assorted people into the crowd. But on further advice it seems this might be too resource-heavy.

I am very keen to learn such things as LOD, or effects to make it look like the crowd is full of people. Also smoke etc effect to cover empty areas (or allow for much reduced detail) would be amazing.

A little bit broad question I realise, but I'm looking for some starting points of what to learn to achieve this stuff.

Many thanks

Answer 1

In answer to one of your comments:

I actually have the same question, but regarding the seats in the stadium. Would it be better to make them in maya (all of them) or is it better to import just one chair mesh to Unity and place them procedurally?

There is extra performance overhead for multiple GameObjects; for example, it's faster to render one mesh containing 100 chairs than to render a single chair mesh 100 times. You can reduce overhead by doing manual instanced rendering using Graphics.DrawMeshInstanced() but that is done with code and can be an extra headache to set up. I'd suggest grouping multiple chairs into one mesh.

However, you should not put all of the chairs in the entire arena into a single mesh, as this prevents frustum culling from working. Frustum culling refers to skipping render of objects that are not inside the camera's boundaries. If all of the chairs are in a single mesh that wraps all the way around the arena, that means that at least some of the chairs will be visible to the camera no matter what direction the camera is pointed, preventing frustum culling. I once worked on a racing game for mobile where the artist had made the barrier around the entire track as a single 200k poly mesh. This ended up causing major performance issues because the engine could never cull any part of the barrier, so we eventually had to break the barrier into segments.

I would suggest grouping chairs into sections, and making each section a single mesh. This way you have less overhead than if each chair was its own object, but frustum culling will still work for sections of chairs that are not in front of the camera.

EDIT: As DMGregory noted in a comment, in the likely event that the chairs are completely stationary, you can also use static batching to have Unity automatically combine multiple instances of a mesh into a single mesh. So for example, if you imported a mesh containing 1 chair and added 500 copies of that chair to the scene, using static batching you could let Unity combine multiple chairs into a single mesh. I do not know offhand how this would work with frustum culling.