Motion Estimation in Real-Time with Optimisation Methods

The estimation of motion information from image sequences is one of the key problems in computer vision. In this context, global optimisation methods play an important role, since they allow for both a precise and dense estimation of the displacement field between consecutive frames. This thesis contributes in two ways to the research in motion analysis: (I) On one hand, novel models for optimisation techniques are presented that yield the most accurate results in the entire literature. (II) On the other hand, specifically adapted multigrid algorithms are developed, that enable the computation of the results in real-time. Thus, for the first time, it becomes possible to apply such high accuracy techniques also in practice.