Motion estimation plays an important role im many emerging
computer vision application ranging from robot navigation
over driver assitance systems to video compression. The
currently most accurate algorithms for estimating motion in
image sequences are based on variational approaches.
However, these methods are often considered to be too slow for
real-time applications. The goal of this talk is to introduce
a class of multigrid algorithms that allow reliable motion
computation on usual PCs at interactive and even real-time
frame rates. This opens the door to a number of interesting
applications. Joint work with Andres Bruhn.