A probabilistic particle tracking framework for high particle densities

Abstract

A framework for particle tracking velocimetry at high particle densities (HD-PTV) based on a Gaussian Mixture Model (GMM) is presented. This new approach is validated by tracking synthetic particles generated for a generalized turbulent pipe flow defining the ground truth. For a step size per time step of δS = 14 px and a particles per pixel (ppp) density of 0:09 the framework tracks about 90% of the ground truth particles (percentage of matched particles, pmp) already after 9 time steps without generating any ghost particles. For a lower step size of δS = 7 px, corresponding to a higher temporal resolution of the flow, and the lowest investigated particle density ppp = 0:02 a constant pmp close to 100% is reported. A decrease on pmp to 80% is found for the highest ppp = 0; 11 - corresponding to about 45000 particles in total. Increasing the step size per time step to δS = 14 px results in a similar sloping curve and pmp that are generally 5% lower compared to the lower step size. The approach is further successfully applied to a well-known experimental tracking problem, i.e. particle tracking in turbulent Rayleigh-Bénard convection, for which the motion of about 28500 particles is tracked. With track lengths up to 250 times steps the occuring structures and velocities are investigated and agree well with previous studies based on tomographic particle image velocimetry using the same data. Thus, it is concluded that the presented HD-PTV framework is an appropriate tool for the flow analysis even at high particle densities