A Dual Forgetting Factor-Based Adaptive IMM-UKF Algorithm for Vehicle Tracking

In response to the limitations of the traditional Interacting Multiple Model Unscented Kalman Filter (IMM-UKF) in high-noise environments—particularly its insufficient tracking accuracy and poor adaptability of the noise covariance matrices—this paper proposes a dual forgetting factor-based adaptive IMM-UKF algorithm (DFF-AIMM-UKF). Building upon the Improved Adaptive IMM-UKF (IAIMM-UKF), the proposed method introduces independent forgetting factor correction mechanisms for both the measurement noise covariance matrixRand the process noise covariance matrixQ. In high-noise scenarios, R and Q are dynamically adjusted using residual and innovation sequences to cope with abrupt noise changes; while in low-noise conditions, the original noise parameters are retained to maintain filter stability. Simulation results show that in different scenarios, the proposed algorithm reduces the position RMSE by approximately 90% compared with the standard UKF, and by 29%-46% compared with the IAIMM-UKF, demonstrating significant improvements in tracking accuracy and stability.