A particle filter with optimal discrete density for hybrid state estimation

This paper proposes a particle filter for estimating the hybrid latent states of dynamic systems in an online manner. The hybrid states generally consist of both continuous and discrete valued elements and naturally appear in a variety of tracking applications. For the hybrid state estimation, particle filters have been widely used because of its nonlinearity and non-Gaussianity. The contribution of this paper is to introduce an optimal probability density for discrete elements, and to combine the density and Monte Carlo approximated density for continuous elements. This combination can the estimation performance. Experimental results show the effectiveness of the proposed method.