Advanced Processing Methods for Image-Based Displacement Field Measurement

We present recent developments in data processing for velocity field estimation and visualization originating from computer vision. We review the current paradigm of PIV data processing, based on window correlation, and the regularization or variational approach which is dominant in optical flow estimation. We propose a novel unifying framework via the optimization of a compound regularized criterion written in terms of a dense displacement (or velocity) field. The paper then focuses on algorithmic issues. A fast iterative window correlation method leading to a highly parallel lgorithm termed FOLKI is described. Thanks to a GPU (Graphical Processing Unit) implementation, FOLKI reaches video rate for typical PIV data. Then we present more sophisticated solvers able to deal with the regularization term of the criterion, notably multigrid methods. In our view, these two components form the foundation of a video rate velocity field visualization and interpretation toolbox which, together with recent advances in experimental apparatus and numerical simulation, opens the way to a major development in experimental fluid science.