Gaussian particle filtering based on coarse and fine-scale sampling on target tracking algorithm

Sequence Monte Carlo estimation for dynamic target involves recursive algorithm and predictive distributions of unobserved time varying signal based on noisy observations. A new Gaussian particle filtering based on coarse and fine-scale coupled chain sampling is presented in this paper, which approximates the posterior distributions by single Gaussians. By using a coarser scale, the target state chain can run faster and better explore the posterior while a fine scale sampling can guarantee the accuracy of target tracking; the coupled chain is included updates that allow target state information to pass between the two scales. Simulation result show the improved GPF (Gaussian particle filtering) reduces the complexity and ensures the accuracy of target tracking.