Abstract
A filtering method called Grid Filtration Filter (GFF) is proposed based on Bayesian inference. First, we select the high-probability region of the current state according to the confidence parameter α, and obtain samples uniformly in this region. Second, these samples are regarded as discretized “potential states” and their posterior weights are calculated based on the Bayesian inference. Third, the grid filtration method is used to choose these “potential states” with high weights, and these selected “potential states” and their normalized weights are used to represent the posterior distribution, and thus estimate the state. we finally verify the feasibility of the GFF algorithm in a typical two-dimensional linear non-Gaussian filtering scenario. Results show that the GFF has a slightly better accuracy than the particle filter (PF) algorithm, and the calculation speed is better by a factor of approximately 45 compared with the PF with 10,000 particles. We finally validate the GFF algorithm in a collaborative target tracking scenario. Results show that the accuracy of the GFF estimation is slightly better than that of the extended Kalman filtering and that of the unscented Kalman filtering, while the advantage of the estimation accuracy of velocity and acceleration is more obvious.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have