Deep velocimetry: Extracting full velocity distributions from projected images of flowing media
Abstract Particle image velocimetry (PIV) is a powerful image correlation method for measuring bulk velocity fields of flowing media. It typically uses optical images, representing quasi-two-dimensional experimental slices, to measure a single velocity value at each in-plane position. However, projection-based imaging methods, such as x-ray radiography or shadowgraph imaging, encode additional out-of-plane information that regular PIV is unable to capture. Here, we introduce a new image analysis method, named deep velocimetry, that goes beyond established PIV methods and is capable of extracting full velocity distributions from projected images. The method involves solving a deconvolution inverse problem to recover the distribution at each in-plane position, and is validated using artificial data as well as controlled laboratory x-ray experiments. The additional velocity information delivered by deep velocimetry could provide new insight into a range of fluid and granular flows where out-of-plane variation is significant. Graphic abstract