“FOLKI-SPIV: a novel approach for ultra-fast Stereo PIV”
by Dr. Benjamin Leclaire
(ONERA/Département d’Aérodynamique Fondamentale et Expérimentale, Meudon, France)
Monday November 29, 2010, at 11:00 – Room ME B 10
Abstract :
This seminar aims at presenting a new Particle Image Velocimetry (PIV) code developed at ONERA, implemented on Graphics Processing Units (GPU), which yields a dense 3-component (3C) vector field from a traditional stereo PIV image set in a typical time of 0.2 s (1000×1000 pixel images, on an NVIDIA Tesla C2050 GPU). The algorithm, called FOLKI-SPIV, relies on a new stereo PIV paradigm. The 3C displacement corresponding to a given interrogation window is calculated directly in one step, in contrast to all existing methods, which first calculate 2-component (2C) vectors corresponding to the images from each camera, and then recombine them in a stereo reconstruction step. To do so, the 3C displacement search is cast as the minimization of a nonlinear criterion combining all camera images, i.e. left and right cameras at both time instants. The dense character of the result, i.e. the fact that 1 vector per pixel is obtained, is a direct by-product of the algorithm, which comes without any time overload while offering the user a convenient degree of freedom for flow characterization. It is in fact oversampling, since the spatial resolution is linked to the interrogation window size, as in traditional PIV algorithms.
During the seminar, I will describe in detail the core principle of FOLKI-SPIV, and show why it is particularly suited to GPU implementation. The performance of the algorithm, both in terms of accuracy and computational time, will also be assessed, and a demonstration will illustrate how GPU programming may enable ultra-fast PIV vector field computation and flow diagnosis.