Camera Mapping is when you map an image onto geometry as seen through the same camera as which you are rendering from. So you basically are only texturing/mapping what will be seen by the camera.

 

 

You'll clone your render camera multiple times at various stages of it's animation path. Say at frame 0, 25, 50, 75 and at 100 (in reality it's a bit more arbitrary than this). Those clones are frozen at their respective frames so that they no longer move along the animation path. You can now use those cameras as virtual slide projectors if you will. You do a render from one, paint it up as pretty as you can and then map the end result back through that same camera onto the geometry. Do that for each of your cloned camera's and you've got your entire shot mapped.

 

Each camera sees only a slice of your scene. All together they should cover 100% of what gets seen by your render cam throughout the shot.. and nothing more. So it's very efficient in terms of texture maps. So you don't end up using an insanely huge texturemap because you need to zoom in on only a tiny bit of it.

 

This technique is often refered to as 2.5D matte painting and is extremely popular in film and more and more also in TV.