R. Shallom, R. Hagege, and J.M. Francos (Israel)

Computer vision, Image registration, Parameter estimation, Nonlinear estimation, Multidimensional signal processing

We consider the general problem of jointly estimating the orientation and deformations of a 3-dimensional object based on a finite set of known templates, and a single obser vation. The observation is obtained by projecting the object onto a plane (the observation plane) whose location relative to the object is a-priori unknown. We propose a method that formulates the original problem as an equivalent prob lem of analyzing a set of polynomials in a low dimensional space. In this setting, the procedure for estimating the ori entation may be formulated as an iterative algorithm for deriving a maximal ideal in the minimal polynomial ring containing the above mentioned polynomial set. Once the orientation of the observation plane relative to the object has been estimated, the deformation parameters are recov ered by solving a system of linear equations.

Important Links:

Go Back