改进的GMPF在动态跟踪系统中的应用

针对动态跟踪系统的非线性问题,提出一种非线性非高斯性能较优的状态估计器:考虑最新观测值的影响,使用基于平方根二阶差分(SRDD2)的高斯混合(GM)模型给出粒子滤波的建议分布.重要性采样和再采样阶段分别采用基于蒙特卡罗的重要性采样和进化的再采样方法,以减轻粒子滤波(PF)的样本退化问题,增强样本的多样性.实验结果表明,与平方根二阶差分Kalman滤波、PF、GM粒子滤波相比,该状态估计器提高了动态跟踪系统状态估计器的综合估计性能.

Application of improved GMPF on dynamic tracking system

Due to the nonlinear problem of the dynamic tracking system,a non-linear and non-Gaussian state estimator has better performance is presented.Considering the latest observations,the Gaussian mixture(GM) model based the square root of second-order divided difference(SRDD2) is introduced to produce proposal distribution.In addition,the importance sampling based Monte Carlo and the evolution re-sampling is employed in the importance sampling and re-sampling steps respectively that not reduce sample degradation of the traditional particle filter(PF) but enhance the diversity of the samples.The simulation result shows that this state estimator has better comprehensive estimation performance for dynamic tracking system than SRDD2 Kalman filter,PF and GMPF.