Analysis of single Gaussian approximation of Gaussian mixtures in Bayesian filtering applied to mixed multiple-model estimation


Orguner U., Demirekler M.

INTERNATIONAL JOURNAL OF CONTROL, vol.80, no.6, pp.952-967, 2007 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 80 Issue: 6
  • Publication Date: 2007
  • Doi Number: 10.1080/00207170701261952
  • Journal Name: INTERNATIONAL JOURNAL OF CONTROL
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.952-967
  • Middle East Technical University Affiliated: Yes

Abstract

This paper examines the effect of the moment-matched single Gaussian approximation, which is made in various multiple-model filtering applications to approximate a Gaussian mixture, on the Bayesian filter performance. The estimation error caused by the approximation is analysed for both the prediction and the measurement updates of a Bayesian filter. An approximate formula is found for the covariance of the error caused by the approximation for a general Gaussian mixture with arbitrary components. The calculated error covariance is used for obtaining a mixed multiple-model estimation algorithm which has a performance near that of GPB2 with less computations.