Evaluation of Tracking Filters for Tracking of Manoeuvring Targets

This thesis evaluates different solutions to the target tracking problem with the use of airborne radar measurements. The purpose of this report is to present and compare options that can improve the tracking performance when the target is performing various manoeuvres while the radar measurements are noisy. A simulation study is done to evaluate and compare the presented solutions, where the evaluating criteria are the estimation errors and the computational complexity. The algorithms investigated are the general pseudo Bayesian of order one (GPB(1)) filter and the interacting multiple model (IMM) filter, each using three motion models, along with several single model Kalman filters. Additionally, the impact on the tracking performance by different choices of radar parameters is also examined. The results show that filters using multiple models are best suited for tracking targets performing different manoeuvres. The tracking performance is improved with both the GPB(1) and IMM algorithms compared to the filters using a single model. Looking at the estimation errors, IMM outperforms the other algorithms and achieves a better general performance for different kinds of manoeuvres. However, IMM have a much higher computational complexity than the filters with a single model. GPB(1) could therefore be more suited for applications where computational power poses a problem, since it is less computationally demanding than IMM. Furthermore, it is shown that different radar parameters have an impact on the tracking performance. The choice of pulse repetition frequency (PRF) and duty cycle used by the radar affects the accuracy of the measurements. The estimation errors of the tracking filters become larger with poor measurements, which also makes it more difficult for the multiple model algorithms to make good use of the different motion models. In most cases, IMM is however less sensitive to the choice of PRF, in relation to how the models are used in the algorithm, compared to GPB(1). Nevertheless, the study shows that there are cases where some combinations of radar parameters drastically reduces the tracking performance and no clear improvement can be seen, not even for IMM