Epipolar geometry estimation for contour-based visual servoing

The problem of determining the epipolar geometry between two planar views is investigated. A novel epipolar geometry estimation algorithm from apparent contours (E.G.E.C.) is presented. EGEC searches for a minimum parameterization of epipoles position by the use of tangency lines at the apparent contours (or silhouettes) of unknown observed object. The retrieved epipoles are then used to design a visual servoing algorithm for mobile robots. Retrieving epipolar geometry from corresponding feature points is usually more involved than contours. However the estimation problem is hard to solve because the apparent contours of the projected scene objects can be of any shape. The existence and uniqueness of the estimated parameters is discussed and several simulations and experimental results are presented