Ghost hunting: an assessment of ghost particle detection and removal methods for tomographic-PIV

This paper discusses and compares several methods, which aim to remove spurious peaks, i.e. ghost particles, from the volume intensity reconstruction in tomographic-PIV. The assessment is based on numerical simulations of time-resolved tomographic-PIV experiments in linear shear flows. Within the reconstructed volumes, intensity peaks are detected and tracked over time. These peaks are associated with particles (either ghosts or actual particles) and are characterized by their peak intensity, size and track length. Peak intensity and track length are found to be effective in discriminating between most ghosts and the actual particles, although not all ghosts can be detected using only a single threshold. The size of the reconstructed particles does not reveal an important difference between ghosts and actual particles. The joint distribution of peak intensity and track length however does, under certain conditions, allow a complete separation of ghosts and actual particles. The ghosts can have either a high intensity or a long track length, but not both combined, like all the actual particles. Removing the detected ghosts from the reconstructed volume and performing additional MART iterations can decrease the particle position error at low to moderate seeding densities, but increases the position error, velocity error and tracking errors at higher densities. The observed trends in the joint distribution of peak intensity and track length are confirmed by results from a real experiment in laminar Taylor–Couette flow. This diagnostic plot allows an estimate of the number of ghosts that are indistinguishable from the actual particles.