Avaliação de algoritmos baseados em Filtro de Kalman aplicados em Bearing-only tracking

Abstract

This master’s thesis presents an evaluation of different Bearing-Only Tracking algorithms based on Kalman Filters, aiming to subsidize an implementation in the combat support system developed by Brazilian Navy’s Research Institute. Nine algorithms were studied, seven of which developed specifically for this problem and one of them being a form of initialization running filters in parallel that can be applied to all of them. The models were simulated in 15 scenarios, combining 5 ways of initialization problem and 3 ways of observer and target ships movement. The results were compared regarding mean squared-error metrics for range and target velocity estimation (considering Cramér-Rao Lower Bound as minimum error reference), credibility of predictions, based on the accuracy of prediction error estimation, besides route convergence and algorithm stability rates. Results showed there was not a single dominant solution. The two models based on polar coordinates presented stability problems, two other models had convergence issue in some scenarios, and a fifth model reached 100% of onvergence rate, but having higher error level. The algorithms Bias- Compensated Pseudo-Linear Kalman Filter (BC-PLKF), Ensemble Kalman Filter (EnKF) and Selective Angle Measurements Instrumental Variable-Based Kalman Filter (SAM-IVKF) stood out, reaching best overall results regarding estimation error and credibility in all scenarios and then they were recommended for a future application. In addition, initialization based on the range and velocity parameterization was indicated, which ensured superior performance in terms of error and credibility, and greater robustness of the algorithms, increasing the convergence rate of those that had a problem.