The multiple model multi-Bernoulli filter based track-before-detect using a likelihood based adaptive birth distribution

摘要

In this paper, a novel multi-Bernoulli filter based track-before-detect (MB-TBD) algorithm is proposed, to solve the tracking problem of multiple maneuvering targets. We incorporate the multiple motion models into the basic MB-TBD filtering, and then derive the closed-form recursive equations including both prediction and update steps based on optimal Bayesian filtering. Moreover, to accommodate the case of unknown prior knowledge for target births, a likelihood based adaptive birth distribution for the MB-TBD is proposed. The implementation of the proposed algorithm with adaptive birth distribution is presented using sequential Monte Carlo (SMC) technique. The performance of the proposed algorithm is demonstrated in challenging scenarios including multiple highly maneuvering objects.