Multi-Sensor Bearing-Only Multi-Target Tracking Based on Cardinality Balanced Multi-Bernoulli Filter

‍ ‍The bearing-only multi-target tracking method employs passive sensors， which do not actively emit signals to typically acquire only the angular measurement information of targets for multi-target tracking and cannot obtain radial distance information. This method faces three main challenges in practical applications： first， because passive sensors only obtain angular measurements of targets， the measurement information is incomplete； second， the measurement equations exhibit high nonlinearity； and third， the sensors might have received clutter and other non-target-generated measurements， leading to measurement source uncertainty. This study addresses these issues by proposing a multi-sensor greedy pseudo-linear particle cardinality balanced multi-Bernoulli filter. Firstly， the Rao-Blackwell theory was employed to decompose the mixed target state vector， treating position components related to measurement values as nonlinear components and velocity components unrelated to measurement values as linear components， and processing them separately using particle filter （PF） and Kalman filter （KF） to effectively reduce the dimensionality of particle filter sampling. Secondly， a novel particle sampling method was designed based on the pseudo-linear Kalman filter （PLKF） to address the problem of traditional particle sampling heavily depending on models， i.e.， using PLKF and the latest measurement information to construct the importance density function and sample nonlinear components. In the update phase， a greedy measurement partitioning strategy was employed to select the optimal measurement set， and the measurement information in the optimal measurement set was used to achieve centralized fusion estimation of multi-target states. Finally， the simulation results were verified， demonstrating that the proposed filter in this study effectively and stably tracks targets using only angular measurements in cluttered environments. Compared to the benchmark methods， the proposed filter more accurately estimates the number and state of targets.