Homography estimation in omnidirectional vision under the L<inf>∞</inf>-norm

Abstract

Solving the vision problem using convex optimization theory is now a focus in computer vision and robot communities. Second Order Cone Programming (SOCP) is especially effective in these methods. This paper discusses homography estimation in omnidirectional vision under the L <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∞</sub> -norm, which provides a theoretical guarantee of global optimality and a wide field of view. We give three different kinds of frameworks in this paper. This approach provides a theoretical guarantee of global optimality. A robot with this algorithm, which provides global optimality and a wide field of view demonstrated by good performance in experiments for synthetic and real data, has a more exact location and 3D reconstruction ability, which cannot be provided by traditional homography estimate method under traditional vision system.