A comparison of the particle and shifted Rayleigh filters in their application to a multisensor bearings-only problem

The multisensor bearings-only problem of target motion analysis - the determination of the position and velocity of a target from noisy bearings made from several moving sensors - presents some interesting challenges to the designer of a tracking algorithm. Apart from the obvious nonlinearities in the measurements, the conditional covariance matrices of the target state are often highly singular in the initial stages. In this paper it is shown that the shifted Rayleigh filter, a 'moment matching' method of modest computational cost, can compete with the particle filter when they are applied to a realistic version of this problem in which a target moves through a mobile multisensor network