An ensemble filter estimation scheme for Lagrangian trajectory reconstruction

International audienceParticle tracking velocimetry (PTV), also referred to as Lagrangian particle tracking (LPT) has recently gained considerable revival. The trend started with the Iterative Particle Reconstruction (IPR) method that applied a projection-matchings cheme, to reconstruct 3D particles’ positions rather than voxel-based intensity, like in Tomographic PIV. Recently, IPR has given rise to the Shake-The-Box (STB) method able to tackle densely seeded flows with considerably high accuracies and reasonable computational efforts. However, in most of 3D turbulent flows, image-based experiments can only provide sparse spatiotemporal data, for which STB is not able to track particles. If more robust estimations are possible, something use-ful may be learnt from the coupling between dynamical models and image data. In responding to these problems, we introduce a novel approach originated from the data assimilation technique comprising a sampling-based optimal estimation algo-rithm, namely a group of ensemble-based filtering variational schemes. We found that employing such an ensemble-based optimal estimation method helped tackling the problems associated with STB : the inaccurate predictor and/or the robustnessof the optimization procedure. The proposed method (ENS) was quantitatively eval-uated with synthetic particle image data built by transporting virtual particles in aturbulent cylinder wake-flow at Reynolds number equal to 3900. We examined the mean positional error of the reconstructed particles, the fraction of track lost particles as well as the required CPU time/memory. We observed that even at large ppp levels (>0.1), the mean positional error of ensemble method was considerably lower than the one given by the STB method. Besides ENS performed equally well interms of data series of relatively large time separation. These preliminary resultsindicates that the ensemble-based method was indeed effective