Sequential computation of Geometric Flows for estimating persistent motions

Abstract

Viscosity of a fluid is one of its defining parameters, measuring it has industrial, medical and pharmaceutical applications. It is possible to estimate fluid's viscosity from its apparent motion, yet existing algorithms fail to reliably determine a fluid's velocity field from video data because of their inability to incorporate noisy observations into a spatially consistent global model. In order to robustly evaluate the velocity fields from video data the algorithm described by Lin et al. is implemented and its performance is evaluated. The algorithm is improved by removing its main limitation - the memory consumption, and by incorporating the SIFT distance measure and Gaussian smoothing. The resulting algorithm is able to robustly estimate persistent velocity fields from noisy, sparse and heterogeneous observations, yet it has to be improved in order to be able to estimate the change of the velocity field with time in order to measure the fluids viscosity.