3D particle field reconstruction method based on convolutional neural network for SAPIV.

Abstract

Synthetic aperture particle image velocimetry (SAPIV) provides a non-invasive means of revealing the physics of complex flows using a compact camera array to resolve the 3D flow field with high temporal and spatial resolution. Intensity-threshold-based methods of reconstructing the flow field are unsatisfactory in nonuniform illuminated fluid flows. This article investigates the characteristics of the focused particles in re-projected image stacks, and presents a convolutional neural network (CNN)-based particle field reconstruction method. The CNN architecture determines the likelihood of each area containing focused particles in the re-projected 3D image stacks. The structural similarity between the images projected by the reconstructed particle field and the images captured from the cameras is then computed, allowing in-focus particles to be extracted. The feasibility of our method is investigated through synthetic simulations and experiments. The results show that the proposed technique achieves remarkable performance, paving the way for non-uniformly illuminated particle field applications in 3D velocity measurements.