Kernelized Lagrangian particle tracking

In this article, we present a novel Lagrangian particle tracking method derived from the perspective of tracking-by-detection paradigm that has been adopted by many vision tasks. Under paradigm, problem consists first learning function (the tracker) maps target particle’s image projection backwardly to its possible position inferred precedent information. The actual is then detected applying learned images captured cameras. We also propose solve using kernel methods. proposed therefore named Kernelized (KLPT). current state-of-art LPT approach Shake-The-Box (STB), despite equipping highly efficient matching and shaking-based optimization procedure, tends be trapped local minimum when dealing with challenging cases featuring sparse temporal data, data extracted complex flows noisy data. KLPT can overcome these difficulties since it features robust procedure combined an linear technique. assessed our against various STB implementations both on synthetic real experimental datasets. For dataset depicting turbulent cylinder wake-flow at Re=3900, focused studying effects density, time separation, noise. outperformed in all more particles producing accurate fields. This performance gain, compared STB, prominent for larger seeding impinging jet flow water tank, capture longer tracks provides detailed reconstruction regions than STB. Overall, comparison shows significant improvements accuracy (lower positional error) robustness (more tracks). finally show results 1st challenge Our algorithm achieved state-of-the-art up 0.08 ppp (particles per pixel).