A comparison of GLR- and multiple model filters for a target tracking problem

In target tracking problems for homing missile guidance one is confronted with the problem of estimating large and rapidly changing target accelerations. Usually this problem is tackled by modelling target accelerations as a first order Gauss-Markov process with large driving noise and application of a standard Extended or Linearized Kalman filter. In contrast to this procedure it was tried to use more sophisticated target acceleration models and incorporate these models into suitable GLR- or multiple model filter schemes. This approach was used for a simple two dimesional target tracking problem, where the target´s lateral acceleration was the main source of uncertainty. The results show small improvements in reaction speed for estimation of target acceleration and a clear potential for reliable detection of jumps in target acceleration levels. As a by-product one multiple model filter displays a-posteriori target acceleration probabilities, showing a mildly bimodal distribution near target acceleration jumps.