Curvature variation of projected cross-sections from straight uniform generalized cylinders

In this paper, we present a new approach for reconstructing Straight Uniform Generalized Cylinder (SUGC) with a scene mapped on its surface. In monocular vision, by using a priori knowledge about the support surface of pictures and projected cross-sections we reconstruct the surface in order to backproject the image on the surface. To reconstruct the surface we have to analyze the curvature of several projected cross-sections. An overview on this domain is found in the papers [Eggert 93, Huang 95, Chen 96]. The curvature variation analyses of surfaces [Kàsa 96] and projected curves [Faugeras 95, Bruce 96] show the complexity of this problem.