Vortex Tracking in Scale-Space

Abstract

Scale-space techniques have become popular in computer vision for their capability to access the multiscale information inherently contained in images. We show that the field of flow visualization can benefit from these techniques, too, yielding more coherent features and sorting out numerical artifacts as well as irrelevant large-scale features. We describe an implementation of scale-space computation using finite elements and show that performance is sufficient for computing a scale-space of time-dependent CFD data. Feature tracking, if available, allows to process the information provided by scale-space not just visually but also algorithmically. We present a technique for extending a class of feature extraction schemes by an additional dimension, resulting in an efficient solution of the tracking problem.