Conics-Based Homography Estimation from Invariant Points and Pole-Polar Relationships

In this paper we derive a novel set of projective invariants and we address the problem of homography estimation between two uncalibrated views from two unmodeled coplanar conics. Exploring in each view the eigenvectors of a matrix composed from the conics matrices and pole-polar relationships, we show that homography can be recovered purely geometrical from invariant points and simple intersections avoiding high-order polynomials and nonlinear equations. Our method has advantages compared with other approaches, is simple and computational efficient. The estimation process is basically linear and there are no ambiguities on the solution. Furthermore it requires neither that the physical models of the conics are known nor that the cameras are calibrated. Experimental results in both simulate data and real images show that our approach is very robust and can be efficiently used for a large number of applications.