On variational methods for fluid flow estimation

Abstract

We present global variational approaches that are capable of extracting high-resolution velocity vector fields from image sequences of fluids. Starting points are existing variational approaches from image processing that we adapt to the requiremements of particle image sequences, paying particular attention to a multiscale representation of the image data. Additionally, we combine a discrete non-differentiable particle matching term with a continuous regularization term and thus achieve a variational particle tracking approach. As higher-order regularization can be used to preserve important flow structures, we finally sketch a motion estimation scheme based on the decomposition of motion vector fields into components of orthogonal subspaces. © Springer-Verlag Berlin Heidelberg 2007.