Comparative Evaluation of Multiple-Model Kalman Filters for Highly Maneuvering UAV Tracking

Tracking highly maneuvering, non-cooperative UAVs poses significant challenges due to rapid and unpredictable changes in target dynamics. Under such conditions, traditional single-model filters often fail to maintain reliable state estimates, resulting in degraded tracking performance. Multiple-Model Kalman Filter (MMKF) approaches, including the Generalized Pseudo Bayesian (GPB1) and Interacting Multiple-Model (IMM) algorithms, improve robustness by simultaneously considering multiple candidate motion models and weighting them according to the observed target behavior. Adaptive strategies, such as chi2-test-based or t-test-based methods, further enhance performance by dynamically responding to changes in maneuvering patterns. This paper presents a multi-criteria comparative assessment of four MMKF formulations-GPB1, IMM, chi2-test-based, and t-test-based filters- under a consistent modeling and simulation framework. Particular emphasis is placed on systematically analyzing the role of the transition probability matrix (TPM), investigating how fixed, adaptive, and TPM-free strategies affect estimation accuracy, robustness to noise, and mode-identification performance. Beyond conventional Root Mean Square Error (RMSE) metrics, the filters' comparison is carried out through confusion matrices and dwell time analysis to highlight performance nuances and trade-offs. This allows to establish which filter formulation is preferable in different operational conditions.