Dynamical models for tracking with the variable rate particle filter

The problem of tracking moving targets is often handled by modelling using hidden Markov models. This approach is attractive because it allows standard algorithms to be used, including the Kalman filter and particle filter. However, it often ignores a significant amount of temporal structure in the path of a manoeuvring target. Variable rate models treat the target motion as deterministic when conditioned upon a sequence of changepoint times and manoeuvre parameters. In this paper, new variable rate models for tracking are presented. Previously, a 2-dimensional model has been developed which parameterises the motion with tangential and normal accelerations. This model is extended by introducing additional variables to improve resilience to model error. It is then modified for use in 3 dimensions by assuming manoeuvres are planar. Simulation tests demonstrate improved tracking performance on a benchmark aeroplane trajectory.