The camera matrices and 3D structure are then computed for the frames 1-2-3-4, for example by first resectioning and then bundle adjustment.