On multi-Bernoulli approximations to the Bayes multi-target filter

摘要

Mahler recently proposed the Multi-target Multi-Bernoulli (MeMBer) recursion as a tractable approximation to the Bayes multi-target recursion, and outlined a Gaussian mixture solution under linear Gaussian assumptions. These proposals are speculative in the sense that, to date, no implementations have been reported. In this paper, it is shown analytically that the MeMBer recursion has a significant bias in cardinality that results in a high number of false tracks. A novel approximation that alleviates the bias problem is proposed. In addition, a sequential Monte Carlo implementation (for generic models) and a Gaussian mixture implementation (for linear Gaussian models)are given. Comparisons with Mahler's original MeMBer filter via simulations show significant reduction of false tracks.