DPIR: Dense Particle Identification and Reconstruction for Dual Frame and Time-Resolved Volumetric PTV

DPIR is a novel technique for improving the accuracy and reconstruction yield for both dual-frame and time-resolved volumetric PTV data. The technique utilizes peaks, projections, and paths within a recursive process to increase the accuracy of 2D particle identification. In high seeding densities, overlap ratios >50% are common, so DPIR fits multiple particle images within a support set using information from up to three sources: 1) image intensity peaks, 2) projections of previously triangulated particles, and 3) paths of 2D particle image trajectories in the image space. 

The technique is described and thoroughly assessed using synthetic particle images from both uniform and turbulent flows in low, medium, and high seeding densities. Notably, for the case of dual-frame data and ppp = 0.05, 95.3% of the real particles were reconstructed with 8.0% ghost particles in 15 iterations.  The reconstruction error was 0.16 pixels.  For time-resolved data, 2D paths were fit using a 2nd order polynomial for the case of turbulent flow with a high particle displacement of about 15 pixels. For this worst-case test, the paths returned image coordinate predictions with a maximum error of approximately 0.18 pixels at the ppp = 0.1, accurately fitting 93% of the particle images.