Enhanced Two-Stage Sparse Bayesian Learning Algorithm for Multivehicle Precise Detection and Localization With MIMO-FMCW Radar

Accurate identification and positioning of multiple vehicles is a critical challenge in autonomous driving, particularly over long distances. While sparse Bayesian learning (SBL) methods have shown promise in radar-based direction-of-arrival (DoA) estimation due to the superior signal reconstruction capabilities, their performance degrades in multi-vehicle scenarios, often resulting in target loss or reduced positioning accuracy. To address this, we propose an enhanced two-stage SBL algorithm for high-precision multi-vehicle localization in MIMO frequency-modulated continuous wave (FMCW) radar systems. Specifically, the proposed algorithm employs a hierarchical two-stage framework. In the first stage, a coarse search is conducted over sparsely sampled grids to efficiently identify potential target regions while significantly reducing computational complexity. In the second stage, an iterative refinement process is applied within locally constructed high-resolution grids to achieve precise joint range and angle estimation. To further enhance estimation robustness, a Sigmoid-based local maximization mechanism is incorporated to mitigate sidelobe interference resulting from dictionary coherence. Moreover, an iterative interference cancellation strategy is employed to mitigate mutual interference among targets, thereby enhancing localization reliability and parameter estimation precision for multiple vehicles. Monte Carlo simulations demonstrate that the proposed approach achieves millimeter-level range accuracy and sub-degree angular precision, outperforming existing methods in both estimation accuracy and runtime efficiency. Field tests further validate the algorithm’s effectiveness in reliable target localization in practical multi-vehicle scenarios.