Tracking Multiple Acoustic Sources in Reverberant Environments using Regularized Particle Filter
Fabio Antonacci, Matteo Matteucci, Davide Migliore, Daniele Riva, Augusto Sarti, Marco Tagliasacchi, Stefano Tubaro · 2007
This paper concerns the problem of tracking acoustic sources in reverberant environments by using a particle filter. The localization problem is transformed into the retrieval of the unobservable state of a dynamical model through noisy measures. Though effective, two problems are related to particle filter: the degeneracy phenomenon (all particles but one are not significative) and the loss of diversity (all particles collapse on the same point). By using Regularized par ticle filter (RPF) and Expectation Maximization (EM) we propose a solution to both problems. Experimental results validate the pro posed solution: Regularized Particle Filter enables to obtain a RMS error lower than 0.2m with a reverberation time of 0.6s.