Possibilistic Kalman filtering for radar 2D tracking

Standard Kalman filter (SKF) introduced by Kalman in the 60s has gained a n on-estimated importance in control as well as in robotics community. Its im portance arises from the obtained optimal result in the sense of variance m inimization under stochastic, Gaussian and unbiased perturbations, and when the state model as well as the measurement model are precisely known. Howe ver, when the last requirement is relaxed such that one or more parameters governing the models are ill-defined and rather given in terms of interval evaluations, Chen et al. (IEEE Trans. Aerospace Electr. Syst. 33 (1) (1997) 251-259) have proposed Interval Kalman Filter (IKF) by extending the arith metic operations to interval calculus. In this paper, we rather assume that the uncertainty pervading some parameters of the models are given in terms of possibility distributions [21]. This leads to a formulation of possibil istic Kalman Filtering (PKF), which agrees with IKF. The same example of 2D -radar tracking is tackled. Comparisons with IKF are investigated as well t he influence of the modelling process on the performance of the filter. Bes ides, the proposal permits to capture certainty qualified information, whic h cannot be obtained from IKF. (C) 2000 Elsevier Science Inc. All rights re served.