I think it was done with 2 or more cameras, a computer, and a bunch of linear algebra. A point in an image corresponds to a visual ray in the real world. Get the same point on the rider from 2 points of view and know where your cameras are and you now have a point in the world. Do this a bunch of times and you get a 3d model of the scene which you can render from any point of view. This requires finding corresponding points on the rider in the various images. This could be done manually by clicking in the images, or automatically with some fancy computer vision techniques(SIFT and RANSAC to name a few). Sorry for the level of technicality. -Philip
|