A Bayesian mixed-model approach to active tracking problems

In the past, a standard Kalman filter with some form of manoeuvre detector has been used for tracking underwater targets. This approach works well in situations characterised by a high signal-to-noise ratio, good observability and a valid state trajectory. Modern targets are now capable of highly versatile manoeuvres, and because of lower cross-section are difficult to detect. Performance of the traditional filters in such situations is not usually satisfactory. In this paper the authors develop an alternative method which is based on ideas pioneered by Harrison and Stevens (1976). The method uses a Bayesian approach to forecasting which introduces multi-process models to cater for the uncertainties as to the underlying model itself. An example is given to illustrate the technique. An example of a stationary active sonar making a measurement of a passing target ship is given