If anyone is interested in Matt's solution, here is my implementation.
answered May 07 '12 at 03:25 PM
You could get the bounding volume using Renderer.bounds, then convert those coordinates to screen coordinates. The min and max of the screen X & Y coordinates will be a bounding box.
However, this is only an approximation. Because it is axis-aligned, AND because the camera angle will distort the shape further. You might end up with a box that's bigger than necessary depending on the shape. It will work better for compact, convex objects.
answered Mar 01 '11 at 12:55 AM