Sequential Monte Carlo Implementation of Cardinality Balanced Multi-Target Multi-Bernoulli Filter for Extended Target Tracking
Dongdong Ma,Feng Lian,Jing Liu
DOI: https://doi.org/10.1049/iet-rsn.2015.0081
2015-01-01
IET Radar Sonar & Navigation
Abstract:It has been shown that the cardinality balanced multi-target multi-Bernoulli (CBMeMBer) filter can extract multi-target states efficiently and reliably. However, when faced with extended targets, it will lead to overestimation. Therefore, a CBMeMBer filter for extended targets tracking is proposed in this study. Three main contributions are made in this study: first, analytical recursion equations of the proposed CBMeMBer filter are derived and relevant proofs are presented. Second, a novel sequential Monte Carlo (SMC) implementation is proposed, which approximates the sample point using several sigma points rather than state transition function. Third, to reduce the particle weight degeneracy problem, a new resample method is introduced. Finally, comparisons between the CBMeMBer and probability hypothesis density (PHD) filters for extended targets tracking are studied through a non-linear example. Numerical study results show that the expanded CBMeMBer filter in the proposed SMC implementation can dramatically improve estimation accuracy in comparison with general SMC implementation. Meanwhile, the authors also find that the proposed CBMeMBer filter gives a more accurate estimation than original CBMeMBer filter while performs more time-efficiently than PHD filter for extended target tracking.