In this talk we will discuss the image geometry of both central catadioptric systems, employed in panoramic viewing, and cameras with non-linear distortions, such as fish-eye lenses and medical endoscopes. Unlike conventional perspective cameras, the projection in these systems is non linear in standard homogeneous coordinates. Such fact difficults the understanding of geometric properties, and precludes the use of closed-form algorithms for calibration and estimation of multi-view relations. We propose to overcome the problem by lifting the image plane using Veronese maps. We prove the existence of general projection matrices, plane homographies and fundamental matrices within this novel representation framework. The usefulness of the approach is illustrated through several application examples, namely in the calibration of catadioptric cameras, estimation of fundamental matrices for mixtures of sensors, and calibration of a wide-area camera network. In the final part of the talk we will briefly introduce an ongoing project in computer surgical navigation based on endoscopic imagery.
Jo?o P. Barreto is Assistant Professor in the Faculty of Science and Technology, University of Coimbra, and Senior Researcher in the Institute for Systems and Robotics. He got his degree in Electrical Engineering in 1997, and the PhD in Computer Vision in 2004, both from the University of Coimbra. He was a Post-Doc Research Fellow at the GRASP Lab., University of Pennsylvania for two years, working in camera networks for tele-immersion applications. His research interests go from active vision to hybrid camera networks, passing by 3D reconstruction, non-conventional vision sensors and surgical navigation. He is author of several articles in the most prestigious vision conferences and journals, and he is, or has been, involved in both European and National funded projects. He also worked as consultant in Product Development actions in the field of real-time operating systems.
For more information see at http://www.isr.uc.pt/~jpbar/
