DARTPIV : Dynamic Adaptive Real-Time Particle Image Velocimetry

Abstract

Particle Image Velocimetry (PIV) is a technique that allows for the detailed visualization of fluid flow. By performing computational analysis on images taken by a high-sensitivity camera that monitors the movement of laser-illuminated tracer particles over time, PIV is capable of producing a vector field describing instantaneous velocity measurements of the fluid captured in the field of view. Nearly all PIV implementations perform offline processing of the collected data, a feature that limits the scope of the applications of this technique. Recently, however, researchers have begun to explore the possibility of using FPGAs or PCs to greatly improve the efficiency of these algorithms in order to obtain real-time speeds for use in feedback loops. Such approaches are very promising and can help expand the use of PIV into previously unexplored fields, such as high performance Unmanned Aerial Vehicles (UAVs). Yet these real-time algorithms have the potential to be improved even further. This thesis outlines an approach to make real-time PIV algorithms more accurate and versatile in large part by applying principles from another emerging technique called adaptive PIV, and in doing so will also address new issues created from the conversion of traditional PIV to a real-time context. This thesis also documents the implementation of this Dynamic Adaptive Real- Time PIV (DARTPIV) algorithm on a PC with CUDA parallel computing, and its performance and results analyzed in the context of normal real-time PIV.