I liked this paper a lot. The notation was clear and the idea was well-presented. One thing I am confused about is what information about the camera matrix you can get just from the eight ground points alone, not considering trying to correspond with the other cameras. In other words, why can't you directly solve for the camera matrix P given the eight correspondences u = Px where the u are the image coordinates and the x are the known ground points? What is the ambiguity not defined?