Fast, large-scale, particle image velocimetry-based estimations of river surface velocity

摘要

A modified high-speed implementation of cross-correlation (CC) based, large-scale particle image velocimetry (LSPIV) was used to estimate the surface velocity of a river with video collected from a gray-scale camera. To improve the quality of results in the high-noise low-signal environment, we introduce a temporal correlation averaging (TCA) scheme that merges a small number of correlation surfaces in the time domain. The TCA scheme is combined with a multi-size macroblock (MMB) sampling method that provides correlation scores from four different macroblock sizes. The TCA scheme is also used in conjunction with a signal-level indicator computed on the macroblock. The signal-level indicator is used to reject correlation scores prior to computation and helps to keep noisy results out of the TCA. These modifications were tested by comparing LSPIV calculations to Acoustic Doppler Current Profiler measurements. The percent difference of measured velocity between LSPIV with TCA and MMB and without TCA and MMB when compared to the ADCP was reduced by as much as 30%. The low processing cost of our modifications along with an efficient multithread implementation of LSPfV facilitates high speed processing of up to a few thousand vector points at rates that exceed the capture speed of common hardware.