ACCELERATED PARTICLEFILTERINGUSINGTHE VARIATIONALBAYES APPROXIMATION

Václav Šmídl, Anthony Quinn

Abstract

InBayesian filtering, themodel mayallow analytical marginalization overasubset, 01,t, oftheparameters. TheMarginalized (Rao-Blackwellized) Particle Filter (MPF)exploits this, byrequiring stochastic sampling onlyintheremaining parameters, 02,t, withthepotential formajor computational and convergence speed-ups. Themarginalized filtering distribution in01,t isexpressed asamixture ofnanalytical components, eachconditioned ononeofthenparticle trajectories in02,t; i.e. sufficient statistics mustbestored andupdated for eachparticle trajectory. Inthis paper, theVariational Bayes (VB)approximation isusedasaone-step approximation to extract necessary momentsfromthenparticles inaprincipled manner, yielding asingle-component marginalized filtering distribution. Thisformalizes andextends arecently reported certainty equivalence approach toaccelerating MPFs.The comparative performance ofthefull andaccelerated MPFs isexplored viaascalar nonlinear filtering example.
BibTeX
@inproceedings{smidl2007accelerated,
  author = {Václav Šmídl and Anthony Quinn},
  title = {{ACCELERATED PARTICLEFILTERINGUSINGTHE VARIATIONALBAYES APPROXIMATION}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2007},
  doi = {10.1109/icassp.2007.367051},
}

Loading related content...

Figure