A comparison of particle filters for recursive track-before-detect

Abstract

Track-before-detect is a powerful technique for detection and tracking of targets with low signal-to-noise ratio. This paper presents a performance comparison of two particle filters recently proposed for this application using several different particle proposal densities designed for track initiation. The first particle filter is a standard SIR particle filter, which treats the track-before-detect problem as a hybrid estimation problem by incorporating a discrete random variable, "target existence," into the state vector. The second particle filter formulates the probability of existence calculation in a different way, avoiding the need for hybrid estimation. Three different particle proposal densities are considered, which are designed to compare performance when the data is used to aid in particle proposal.