On the Development of a Volumetric Velocimetry Technique using Multiple Plenoptic Cameras

Abstract

Plenoptic PIV was recently introduced as a viable three-dimensional, three-component velocimetry technique based on light field cameras. One of the main benefits of this technique is its single camera configuration allowing the technique to be applied in facilities with limited optical access. The main drawback of this configuration is decreased accuracy in the out-of-plane dimension. This dissertation presents a solution with the addition of a second plenoptic camera in a stereo-like configuration. A framework for reconstructing volumes with multiple plenoptic cameras including the volumetric calibration and reconstruction algorithms are presented. It is shown that the addition of a second camera doubles the reconstruction quality and removes the `cigar'-like elongation associated with the single camera system. In addition, it was found that adding a third camera provided minimal benefit for the reconstruction quality of sparse particle fields. Further metrics of the reconstruction quality are quantified in terms of particle density, number of cameras, camera separation angle, voxel size, and the effect of common image noise sources. In addition, a synthetic Gaussian ring vortex is used to compare the accuracy of the single and two camera configurations. It was determined that the addition of a second camera reduces the RMSE velocity error from 0.85 to 0.23 voxels. Finally, the technique is applied experimentally on a ring vortex and comparisons are drawn from the four presented reconstruction algorithms.

The trade-off between spatial and angular resolution is the main consideration when designing a plenoptic camera. This dissertation provides guidelines for the selection of the microlens array using theoretical analysis as well as synthetic and experimental data for validation. It was determined that the optimal selection of the microlens size depends heavily on the desired volume depth and a good rule-of-thumb is the span of the volume should be ~1.1 DoFp (single pixel, or perspective, depth-of-field). It was also determined that while this is the optimal selection, the robustness of the cross-correlation algorithm mitigates the effect of sub-optimal microlens selection allowing for a single configuration to be used in a wide variety of situations.