Computational imaging, optical design, particle image velocimetry, deep-learning network, Applied optics. Photonics, TA1501-1820, Optics. Light, QC350-467

A three-dimensional (3D) particle image velocimetry (PIV) system typically consists of multiple cameras. However, micro-PIV systems for measuring microscale velocity fields lack sufficient space to accommodate them. In this work we propose an alternative approach based on computational imaging, enabling monocular micro-PIV systems to perform 3D flow field measurements without additional hardware or complex structure. The microscopic objective is designed to satisfy the required parameters, and the point spread function (PSF) responses of the system to different depths of the object surface are obtained. Additionally, a particle dataset generation method based on the PSFs of the optical system is proposed, and a deep-learning network is constructed for training. To validate the feasibility, particle images are captured in experiments and inputted into the network to reconstruct depth images and build three-dimensional flow fields. Simulation and experimental results demonstrate that the measurement deviation is within 13.2%, indicating the practicality of the proposed model.