Joint Probabilistic Data Association Filter Using Adaptive Gibbs Sampling

This paper proposes a novel adaptive Gibbs sampling algorithm to implement joint probabilistic data association filter for multiple targets tracking. Instead of uniformly visiting and sampling each single element in one joint association hypothesis, the proposed algorithm selects an optimal element visiting sequence that tends to keep the most probable single association hypothesis. Compared to the random Gibbs sampling, it has been demonstrated that the proposed adaptive Gibbs sampling provides faster convergence speed, thus improving the tracking accuracy when the number of samples is limited, and improved robustness against the variation of the number of burnin samples. Extensive empirical simulations are undertaken to validate the performance of the proposed approach.