不确定重尾量测噪声干扰下的鲁棒目标跟踪算法

在大多数目标跟踪方法中，通常假设量测噪声服从参数已知的高斯分布或对称重尾分布，但其非常受限并且在实际过程中常常无法得到满足。因此，针对存在不确定重尾量测噪声干扰下的目标跟踪问题，提出基于变分推理的鲁棒容积卡尔曼滤波算法。该算法利用Skew-T分布对不确定重尾量测噪声进行建模，在基于容积规则的数值积分过程中，结合变分推理将Skew-T分布量测噪声参数与系统状态变量进行联合递归计算，通过对近似后验概率密度函数进行变分迭代，获得系统模型和不对称重尾量测噪声参数。仿真结果表明，该算法相比变分贝叶斯扩展卡尔曼滤波算法具有较高的滤波精度。

A Robust Target Tracking Algorithm under Condition of Uncertain Heavy-Tailed Measurement Noise

In majority target tracking methods, the measurement noise is generally assumed to be known Gaussian distributed or asymmetric Heavy-tail distributed. However, this assumption is very limited and often does not satisfy the needs of the work in practical application. A variable inference robust cubature Kalman filter is proposed for the nonlinear target tracking with unknown time-varying asymmetric Heavy-tailed noise. The asymmetric Heavy-tailed noise is modeled by Skew-T distribution. In the process of the numerical calculation of cubature Kalman filter, the system state and measurement noise parameter are jointly estimated recursively by the variable inference. The system model and unknown asymmetric heavy tailed measurement noise parameters are obtained by variable iteration of an approximate posterior probability density function. The simulation results show that the proposed algorithm is higher than the variable Bayesian extended Kalman filter algorithm in filtering accuracy.