Differential-Algebraic Multiview Constraints

In this paper we present a novel type of multiview constraints, called differential-algebraic multiview constraints. These constraints are based on both corresponding feature points and the motion of these feature points, thus combining the discrete multiview constraints with the continuous multiview constraints. The main usage of these constraints is as a theoretical basis for filtering approaches to structure and motion recovery, enabling an update of a current motion estimate when a new image becomes available. One important feature is that the update formula becomes linear in the motion parameters in the calibrated case, which is a major improvement compared to the standard discrete approach. Another advantage is that fewer points are needed in the update formula then when calculating the motion without using a prior estimate