Recovering structures and motions from mutual projection of cameras

In this paper, we propose a new method for recovering camera motions and structures of the scene accurately and reliably. For recovering motions and structures, it is very important to compute the epipolar geometry accurately. Although some linear methods have been proposed for computing the epipolar geometry, they are still less accurate than nonlinear methods. The non-linear methods, on the other hand, require a lot of computational power, and sometimes fall into local minima. In this paper we show that by rising the actual projection of cameras, the epipolar geometry can be computed much more reliably from less image correspondences by a linear method. We also show that by using the epipolar geometry derived from the mutual projection of cameras, the structures of the scene can be recovered Much more accurately and reliably than the existing methods.