Testing image-velocimetry methods for turbulence diagnostics

Abstract

Two image-based velocity-inference techniques, cross-correlation time-delay estimation (CCTDE) and dynamic time warping (DTW), were tested. These techniques are conventionally used in the study of plasma dynamics, but they can be applied to any data where features propagate across the image field-of-view. Differences between the techniques were investigated, which showed that the shortcomings of each technique are complemented well by the strengths of the other. Thus, the techniques should be used in conjunction with each other for optimal velocimetry. For ease of use, an example workflow that applies the results in this paper to experimental measurements is provided for both techniques. The findings were based on a thorough analysis of the uncertainties for both techniques. Specifically, the accuracy and precision associated with inferred velocity fields were systematically tested using synthetic data. Novel findings are presented that strongly improve the performance of both techniques, some of which are as follows: CCTDE was able to operate accurately under most conditions with an inference frequency as short as 1 per 32 frames, as opposed to the typical 1 per ≥256 frames used in the literature; an underlying pattern in CCTDE accuracy depending on the magnitude of the underlying velocity was found; spurious velocities due to the barber pole illusion can now be predicted prior to CCTDE velocimetry through a simple analysis; DTW was more robust against the barber pole illusion than CCTDE; DTW performance with sheared flows was tested; DTW was able to reliably infer accurate flow fields from data with as low as 8 × 8 spatial channels; and however, if the flow direction was unknown prior to DTW analysis, DTW could not reliably infer any velocities.