Consensus-based tracking for 3D PTV at high seeding densities

Abstract A robust pairing algorithm with outlier removal is introduced in the context of two-pulse 3D particle tracking velocimetry at high seeding densities, with high concentrations of ghost particles. Integrating the vector field consensus approach from Ma et al. (IEEE Trans Image Process 23:1706–1721, 2014), the algorithm, its underlying hypotheses, and its relevant input parameters are investigated in the context of turbulent flow measurements. 2D synthetic tests are first carried out to quantify the algorithm’s performance and derive simple guidelines for optimal parameter tuning strategies based on experimental quantities. It is found that 2D vector fields with up to 90% outliers can be handled by our algorithm. 3D synthetic tests are then implemented to test the tracking strategy robustness to increasing image densities and ghost particle concentrations. We show that our algorithm can be used for particle pairing in particle clouds with up to 50% of ghost particles. Results submitted on the two-pulse dataset of the first LPT challenge, using the associated data portal with automatic evaluation, also showcase the overall excellent performances of the method. Finally, the method is used successfully on experimental data from our Giant Von Kármán setup (characterized by up to 65% of ghost particles), as evidenced by comparisons of its output with respect to results provided by the Shake-The-Box algorithm and with results provided by a pairing approach using a 3D cross-correlation predictor.