Motion Estimation via Dynamic Vision

Estimating the 3D motion of an object from a sequence of projections is of paramount importance in a variety of applications in control and robotics. Although "visual motion estimation" is an old problem, only recently tools from control and estimation theory have hinted at acceptable solutions. Moreover, the problem raises a number of issues of system theoretic interest, such as nonlinear estimation and identification on topological manifolds and observability in a projective geometric framework. In this paper we analyze a formulation of the visual motion estimation problem in terms of identification of nonlinear implicit systems with parameters on the so-called "essential manifold"; the estimation is performed either in the local coordinates or in the embedding space of the parameter manifold