Jornal article

Šimandl M., Straka O.

Signal Processing

88

2008

11

2784-2789

0165-1684

Ing. Ondřej Straka, Ph.D. (FAV Západočeská univerzita, katedra kybernetiky)

Estimation; Nonlinear filtering; Particle filters

Sampling density design for particle filters is treated. A new functional approach is proposed and discussed. The approach follows the idea of the auxiliary particle filter which introduces the concept of primary weights into sampling density design. In contrast to the auxiliary particle filter approach, which utilizes the primary weights given by a measurement probability density and a predictive point estimate, the functional approach is based on comparison of the measurement and predictive probability density functions through a suitable metric. Different choices of the metric and weight function are discussed and used for the comparison. The particle filter with the sampling density given by the functional approach provides estimates that are closer to exact filtering probability density function in terms of point estimates.