Unscented卡尔曼滤波-卡尔曼滤波算法

Unscented Kalman filter-Kalman filter algorithm

针对条件线性高斯状态空间模型,提出unscented卡尔曼滤波-卡尔曼滤波unscented Kalman filte-ring-Kalman filtering,UKF-KF算法,该方法用UKF估计条件线性高斯状态空间模型中的非线性状态,用KF估计线性状态。为了有效地融合UKF和KF估计的后验状态分布,将蒙特卡罗方法应用于KF估计的线性状态均值和方差,获得了与UKF sigma点相同数量的后验线性状态估计分布的样本,然后将这些样本与UKF中sigma点进行合成去获得系统中非线性状态的估计。该算法应用于机动目标跟踪的仿真结果表明:与Rao-Blackwellized粒子滤波器(Rao-Blackwellized particle filter,RBPF)相比,该算法虽在估计精度上略有下降,然而计算时间明显降低,有效提高了实时性。

A new filtering method, called as the unscented Kalman filtering-Kalman filtering （UKF-KF） algorithm is presented for conditionally linear Gaussian state space models. The method uses UKF to estimate the nonlinear states of a conditionally linear Gaussian model while its linear states are estimated by the KF. In order to fuse the two posterior distributions of the estimated states obtained by the UKF and KF respectively, Monte Carlo method is applied to generating samples from Gaussian distributions with the KF estimated means and covariance. The generated samples with the same number as sigma points have are then combined together with the sigma points to estimate the nonlinear states in the model using the UKF. The simulation results of the proposed method applying to tracking the maneuver target, compared with the RBPF （Rao-Blackwellized particle filter）, show that the proposed UKF-KF only consumes 8% the computing time required by the RBPF with a little bit filtering performance decline.