A Rapid, Empirical Method for Detection and Estimation of Outlier Frames in Particle Imaging Velocimetry Data using Proper Orthogonal Decomposition

This paper develops a method for detection and removal of outlier images from digital Particle Image Velocimetry data using Proper Orthogonal De-composition (POD). The outlier is isolated in the leading POD modes, removed and a replacement value re-estimated. The method is used to estimate and replace whole images within the sequence. This is particularly useful, if a single PIV image is suddenly heavily contaminated with background noise, or to estimate a dropped frame within a sequence. The technique is tested on a synthetic dataset that permits the effective acquisition frequency to be varied systematically, before application to flow field frames obtained from a large-eddy simulation. As expected, outlier re-estimation becomes more difficult when the integral time scale for the flow is long relative to the sampling period. However, the method provides a systematic improvement in predicting frames compared to interpolating from neighbouring(1) frames.