Recovering Generalized Cylinders by Monocular Vision

Abstract

In this paper, we propose a theoretical study of the Generalized Cylinders (GC) with constant circular cross-section viewed under perspective projection. We show that a single view of the object make it possible to infer its 3D shape. Two algorithms are described. The first method is based on the matching between two contour points from a same cross-section and each cross-section is separately modeled. The second algorithm uses a global model, parameterized with a BSpline curve. The model shape is twisted from an initial position until the limb projection fits the image content. With such an approach the matching problem between image points from the same cross-section is avoided and the global shape is estimated through a minimization process. Experiments on real data are provided, which prove the validity of the two approaches.