鲁棒边缘粒子滤波及在目标跟踪中应用

Robust Marginalized Particle Filter and Its Application in Target Tracking

摘要:

边缘粒子滤波是组合导航和目标跟踪中状态估计的高效方法。本文目的是研究附加量测噪声具有时变未知方差的鲁棒边缘粒子滤波的算法并对算法仿真验证。设计方法是使用Rao–Blackwellised原则实现混合模型中状态降维，然后状态与量测方差同时分别估计；量测分布模型设置为具有鲁棒性质的学生t分布，通过这种量测似然模型得到粒子权值；变分推断方法加入混合滤波方案进行量测噪声方差参数的实时递推估计。重采样阶段粒子权值与状态及噪声参数一起进行重采样，结果是给出状态与噪声参数估计的鲁棒边缘粒子滤波。通过对常速目标运动跟踪模型量测噪声方差渐变和突变两种情况的仿真设置分析，验证了所提算法在量测方差变化情况下性能优于边缘粒子滤波算法的结论。

Abstract:

Marginalized particle filter is an efficient estimation method for navigation and target tracking. The purpose of this paper is to study the Marginalized filter algorithm with time-varying unknown measurement noise variance. The design method is to achieve state dimensionality reduction and estimation of state and measurement variance respectively by using Rao–Blackwellised idea. The measurement distribution model is set as the robust student t-distribution, and the particle weights are obtained through the measurement likelihood model. In this paper, a real-time recursive estimation of the variance parameters of measurement noise is performed by combining the mixed filtering scheme with the Variational inference method. In the resampling stage, the particle weights are resampled together with the state and noise parameters, as a result, robust marginalized particle filter is presented after the state and noise parameters are estimated. Through the simulation analysis of two time-varying cases of gradual change and abrupt change of measurement noise variance of the given target motion model, the conclusion that the performance of the proposed algorithm is better than that of the marginalized particle filter in the case of time-varying measurement variance is verified.