Lucas-Kanade Fluid Trajectories for time-resolved PIV.
Résumé
We introduce a new method for estimating fluid trajectories in time-resolved PIV. It relies on a Lucas-Kanade paradigm and consists in a simple and direct extension of a two-frame estimation proposed in that context. The so-called Lucas-Kanade Fluid Trajectories (LKFT) are assumed to be polynomial in time, and are found as the minimizer of a global functional, in which displacements are sought so as to match the intensities of a series of image pairs in the sequence, in the least-squares sense. All pairs involve the central image, similar to other recent time-resolved approaches (FTC and FTEE). As switching from a two-frame to a time-resolved objective simply amounts to adding terms in a functional, no significant additional algorithmic element is required, and similar to FOLKI-SPIV, the method has an important potential for GPU acceleration. Tests on synthetic data with translating and rotating motions show that in the current implementation, and using a cubic B-Spline interpolator for image deformation, LKFT has a total error comparable to that of FTEE, while improvements can still be brought to the algorithm. Besides, results on case B of the third PIV challenge confirm its ability to drastically reduce the random error in situations with low signal-to- noise ratio.