Many computer vision problems such as optical flow, stereo reconstruction and registration require to establish corres-pondences between images. In this talk a systematic framework will be presented that models these problems in terms of continuous optimisation problems. It will be demonstrated that these models can yield highly accurate results that are robust under noise and illumation changes. By means of sophisticated multigrid algorithms real time applications are possible.
Algorithmic adaptations to modern parallel hardware will be discussed as well, and a number of applications will be presented.