Random finite set–based multi-target bearings-only tracking under nonstationary measurement conditions

To address tracking divergence in underwater multi-target bearings-only tracking induced by nonstationary measurement-noise statistics, a multi-target bearings-only tracking algorithm, termed VB-GMCPHD, is proposed by integrating random finite set (RFS) theory with variational Bayesian (VB) inference. Within the RFS framework, set-based recursion is performed for multi-target state estimation, the measurement-noise covariance is modeled as a latent random variable and assigned a conjugate prior, and the VB inference is employed to adaptively estimate the measurement-noise statistics online. Consequently, the measurement likelihood is adaptively corrected during the update step and likelihood mismatch is mitigated. Simulation studies and experiments demonstrate that stable recursion is maintained under nonstationary measurement noise, enabling robust multi-target bearings-only tracking in underwater environments.