Comparison of Gaussian particle center estimators and the achievable measurement density for particle tracking velocimetry

Abstract

A series of numerical simulations were conducted to investigate the performance of two particle center estimation algorithms for Particle Tracking Velocimetry: a simple three-point Gaussian estimator and a least-square Gaussian. The smallest position error for images with reasonable noise levels was found to be approximately 0.03 pixels for both estimators using particles with diameters of 4 pixels. As both estimators performed equally well, use of the simple three-point Gaussian algorithm is recommended because it executes 100 times faster than the least-square algorithm. The maximum achievable measurement density and accuracy for the three-point Gaussian estimator were determined with a numerical simulation of an Oseen vortex. Uncertainty measures have been introduced to filter out unreliable displacement measurements. It was found that 4 to 5 velocity vectors could be obtained within a 32 × 32 pixel area with an average displacement error of 0.1 pixels. This doubles the spatial resolution of conventional cross-correlation based Particle Image Velocimetry at comparable accuracy.