The VICCTOR algorithm for 3D Particle Tracking Velocimetry: an efficient ghostbuster-type approach

Abstract A new algorithm for two-pulse particle tracking velocimetry (PTV) is introduced, which makes use of recent advances in 3D particle reconstruction, matching and position optimization. Contrary to most current state-of-the-art approaches, e.g. Godbersen et al. (2024), which progressively retrieve the true particles while maintaining a small proportion of outliers, we choose a symmetric path, operating as much as possible at a high rate of found true particles, and focusing most of our algorithmic choices on progressively eliminating outliers. First guesses of particles positions at the two instants are obtained introducing a modified, large-scale version of the Particle Volume Reconstruction (PVR) of Champagnat et al. (2014), which discretizes the image formation process on a fine voxel grid. Positions and intensities are then globally optimized, using an L-BFGS optimizer. Upon discarding particles with a low reconstructed intensity, clouds that recover a large fraction of the correct particles are obtained, although with a large amount of noise (i.e. ghost particles). The F-VFC algorithm introduced in Le Bris et al. (2025) is then used to match the particles from both instants while also removing a large amount of ghost particles from the particle clouds. This full process is repeated on residual images until convergence. For synthetic images, the obtained approach, named VICCTOR (Voxel-grid Initialized reConstruction with Consensus Tracking and Outlier Removal) nearly recovers all particles up to a seeding density of 0.2 ppp and with very little noise, while maintaining a very low error on particle positions. The approach also shows its excellent performance on experimental data from a Giant Von Kármán set-up, as evidenced by comparisons of its output with respect to results provided by the shake-the-box algorithm applied on full time-resolved sequences.