Adaptive Forgetting Factor Probabilistic Data Association Filtering for Radar Target Tracking in Cluttered Environments With Unknown and Varying Measurement Noise Covariance

Abstract

An adaptive forgetting-factor probabilistic data association filter, termed AR-$\lambda$PDAF, is proposed for radar target tracking in cluttered environments with unknown and varying measurement noise covariance. In conventional PDAF, inaccurate measurement noise covariance may lead to mismatched validation gating, association probability calculation and state correction. To address this problem, an exponentially weighted moving average estimator is embedded into the PDAF recursion to estimate the measurement noise covariance online from the combined innovation sequence. Moreover, the forgetting factor is adaptively regulated according to the Mahalanobis distance of the combined innovation. When a significant statistical deviation is detected, a smaller forgetting factor is assigned to improve the response to current measurements; otherwise, a larger forgetting factor is used to retain historical information and suppress clutter-induced fluctuations. The design criterion of the adaptive forgetting factor is established, and the approximate steady-state unbiasedness of the proposed estimator is analysed. Monte Carlo simulations under constant velocity and coordinated turn models show that, in the present experiments, AR-$\lambda$PDAF generally achieves better tracking accuracy than AR-PDAF and fixed-forgetting-factor EWMA-PDAF, with more pronounced advantages under abrupt covariance changes.