Non-linear/non-Gaussian filtering and the bootstrap filter

The bootstrap filter is a random sample (stochastic simulation) based approach to implementing general Bayesian filters. The central idea of this approach is to represent the required p.d.f. by a set of random samples, rather than as a functional form over state space. This technique is able to cope with any functional nonlinearity and system and measurement noise of any distribution. There are also significant reparameterisation and p.d.f. summarisation advantages of a sample based approach. We outline the bootstrap filter approach together with several techniques for improving the efficiency of the basic algorithm and then present a Monte Carlo analysis of a bearings-only tracking problem to illustrate performance.<>