Here’s a proof of concept idea I’ve been tinkering with.
Instead of using cards, flat surfaces or custom 3D geometry as a target for camera projections, I’ve been testing procedurally generated “blocks” to represent various buildings.
The idea is that from far above, buildings are just a clusters of cubes anyway, so it might enough in some situations to represent the contours of a city by using just such simple cubes.
A test video can be seen here:
http://www.youtube.com/watch?v=EDz35mmMASo
Only one image (2k resolution) was projected on to the scene.
Of course, there’s simply not enough data there in a single image if you go too extreme with the camera, but for smaller camera moves this technique does give better results than either image cards (as they have no “roof”, they betray the effect quickly as the camera flies over them), or flat surfaces (which start to stretch really fast).
Naturally, this is not as good as building custom geometry to represent the various buildings and trees and such - and/or creating multiple projections and patches - but as the process is semi-automated, it is really fast to set up.
I used the Greeble plugin for the test in Max and a simple planes as a source.