ABSTRACT
We present a comparison methodology for optical flow methods and Particle Imaging Velocimetry (PIV) in the context of fluid mechanics applications. Optical flow (OF) algorithms Lucas-Kanade, Horn-Schunck and Farnebäck are considered, since they are well known established optical flow methods. A synthetic image generator provides the input data to compare different velocimetry methods. The image generator can be parameterized to generate tracer images corresponding to different types of flows, namely: uniform flow; parabolic flow; flow with stagnation point; Rankine vortex and Rankine vortex with superimposed uniform flow. Particle size, particle concentration, white gaussian image noise (WGIN), particle velocity, image bit depth (8-bit, 10-bit, 12-bit), particle out of plane movement (originated from turbulence), as well as laser sheet thickness are control parameters. The true values of the flow field for each generated image pair are also exported for validation, along with statistics of in-plane particle and out-of-plane particle losses per each image sub-area, or, interrogation area (IA) in PIV terminology. A total of 7200 image pairs were generated and processed by each of the algorithms referenced above. Results are presented and compared in terms of accuracy, regarding the algorithm type, flow type, IA relative velocity, WGIN, pixel particle size and particle concentration in IA. All the images generated for this study have 8-bit pixel depth. Since this method and procedure is generic it can be reused for other type of flows and optical flow algorithms.
