I am creating an environment using procedural content. To optimise performance, I am batching small meshes ahead of time. For efficiency's sake, I don't want to batch objects that make only an insignificant contribution to the final frame.
My question is, how should I determine this? Is there a commonly used or provably 'correct' algorithm?
I was thinking of something like: calculate the size of the bounding box on the final frame based on the camera parameters and see if its below a certain threshold.
Is there a better one?
(PS. This is not premature optimisation - currently my batches go well over the 65k VBO limit - not batching small meshes may save a number of draw calls. Also to be clear, I am not asking about occlusion, I have a number of small highly detailed objects in a scene so large than some will simply not be noticeable if they are far enough away and can be safely culled.)