Underwater Multiple Targets Tracking Using Multi-Bernoulli Filter and Improved Data Association

Traditional multi-target tracking methods using Random Finite Set (RFS) theory mainly generate trajectories by using label RFS or performing data association after RFS filtering. For the case of target trajectories intersect, the too small distance between adjacent targets (i.e. targets are in close proximity) can cause label confusion or association errors, making it challenging for these methods to generate accurate trajectories. To solve this problem, we propose a target tracking method that integrates the multi-target multi-Bernoulli (MeMBer) filter and the improved data association. The MeMBer filter is used to obtain the target state estimations before data association. To avoid association errors when targets are in close proximity, we make the following improvements based on the nearest-neighbor data association method. When multiple targets are far from each other, we directly use the existing nearest-neighbor method to generate trajectories; when multiple targets are in close proximity, we treat the trajectory predictions of the last time step as the current trajectories, therefore avoiding data association errors. The simulation results demonstrate that the proposed method effectively solves the problem of inaccurate trajectory estimation during the intersection of target trajectories.