改进粒子滤波算法的比较

Comparison of Improved Particle Filtering Algorithms

重要性密度函数的选择对粒子滤波至关重要,围绕重要性密度函数的选择,已提出许多改进粒子滤波算法,典型的有扩展卡尔曼粒子滤波(EPF),不敏卡尔曼粒子滤波(UPF)、辅助粒子滤波(APF)及正则化粒子滤波(RPF)。详细讨论了4种改进粒子滤波算法的基本思想、性能特点及主要步骤。通过对一典型标量非线性系统的滤波实验,对4种改进算法的性能进行了仿真比较,实验结果表明,4种改进算法都从不同程度上改善了粒子滤波器的性能,其中,UPF的性能最优。最后,分析了各算法的改进原因。

The choice of importance density function is very important for the particle filtering. Concerning the choice of importance density function, many improved particle filtering algorithms have been proposed, such as： Extended Particle Filter（EPF）, Unscented Particle Filter （UPF）, Auxiliary Particle Filter（APF） and Regularized Particle Filter（RPF）. The basic thought, characteristics of performance and main steps of the four improved algorithms are discussed in detail. Through a filter experimentation on a typical scalar quantity non-linear system, we made a comparison of the performance on these algorithms, and the results showed that these improved algorithm improve the performance of particle filter in different degree, and the UPF has the best performance. Finally, reasons of performance improving of each algorithm are analyzed.