1 Introduction
Over the past two decades, structure-from-motion (SfM) has developed into a mature discipline of visual information processing. The classic case of a camera moving through a static scene is essentially solved in a coherent theory [4], [8], and several practical systems exist which are able to robustly recover the structure and motion from unconstrained real-world sequences. One of the main limitations of real-world applications of structure-from-motion at present is the requirement for a static scene.