Abstract:

In standard particle filter, proposal distribution is chosen without considering the most recent observation, so the variance of important weight increases with time, which results in weight degeneracy. To overcome this shortcoming, a new algorithm named the iterative sigma point filter is proposed. In the prediction stage, weighted particles are drawn from the probability density function by the sigmapoint filter. Then according to the observation data the weighted particles are updated. In the modification stage, iterative Kalman filter is adopted to optimize the mean and variance of the state distribution which are obtained in the prediction stage. Simulation of target tracking on reentry shows that this new algorithm can ensure the filter convergence. It also improves the estimation accuracy and robustness in comparison with the standard particle filter.

Key words: automatic control technology, particle filter, sigma point particle filter, iterative Kalman filter, target tracking