Multiview geometry in traditional vision and omnidirectional vision under the l∞-norm

This study presents a review of multiview geometry problems in traditional vision and omnidirectional vision under the L_∞-norm. The main advantage of this approach is a theoretical guarantee of global optimality. First, three core problems in multiview geometry in traditional vision are formulated as second-order cone programming feasibility problems. The extension of L_∞-norm approach for multiview geometry from traditional vision to omnidirectional vision is shown by three models, a mirror model, a sphere model and a cylinder model. Finally, the authors assess their potential for future deployment and present directions for future research.