Spatial-temporal improvements of a two-frame particle-tracking algorithm

A novel algorithm for particle-tracking velocimetry is proposed and tested with both synthetic and real images. It uses nearest-neighbour cluster matching which performs better than fixed area approaches in terms of spatial adaptivity. The algorithm includes several temporal multi-frame improvements, i.e. extrapolation of the expected particle positions in subsequent frames and the frame-gap technique. To further improve the tracking algorithm performances, the particle identification procedure was modified with respect to the traditional background subtraction, local thresholding and grey level weighted averaging by using the optical flow equation. The local maximum of grey levels around each feature extracted is identified and the barycentres of the particle associated with it are calculated by using Gaussian fitting. The novel algorithm works well with several seeding densities, both homogeneously and inhomogeneously distributed. The multi-frame approach substantially improves the average trajectory length and the number of long trajectories in images with and without noise. The number of barycentres correctly identified by employing the feature extraction is significantly larger than when traditional techniques are used, which in turn increases the number of velocity vectors, allowing a better characterization of the flow field under investigation.