A Dynamic Programming based Optical Flow technique
has been applied to the Particle Image Velocimetry (PIV) problem.
It has been used for the extraction of dense velocity fields in a
planar section of a fluid flow illuminated by a thin laser light sheet.
Two (in-plane) components of the velocity vectors can be recovered
using a single camera and all three components can be recovered
using two or three cameras.
Quantitative performance tests have been carried out on calibrated
synthetic image sequences from the PIV Standard Project of the
Visualization Society of Japan (VSJ). Results are presented for the
2D flow based sequences (STD01 to STD08 Standard Images) and
the 3D flow based sequences (STD301, STD302, STD331 and STD337
Standard Images). The RMS error is within the 2-3 % range and within
the 4-8 % range for recovery of the two-component and the
three-component velocity vectors respectively.