A Labeled Multi-Bernoulli Filter Based on Maximum Likelihood Recursive Updating

Abstract

A labeled multi-Bernoulli filter is used to obtain estimates of the identities and states of targets in complex environments. However, when tracking multiple targets in dense clutters, the computational complexity of the traditional labeled multi-Bernoulli filter will increase exponentially. A labeled multi-Bernoulli tracking algorithm based on maximum likelihood recursive update is proposed, which can reduce the computational scale while maintaining tracking accuracy. Specifically, when performing posterior estimation, a maximum likelihood recursive update method is proposed to replace the complete enumeration, truncated enumeration, or sampling enumeration methods used in many traditional methods. Furthermore, combined with the Gaussian mixture technique, a maximum likelihood recursive updating labeled multi-Bernoulli tracking algorithm is constructed. Simulation results demonstrated that the proposed filter obtained a good balance between the tracking accuracy and computational efficiency.