基于Huber的高阶容积卡尔曼跟踪算法

为改善高阶容积卡尔曼滤波算法的滤波精度和鲁棒性, 提出了一种新的基于Huber的高阶容积卡尔曼滤波算法. 在采用统计线性回归模型近似非线性量测模型的基础上, 利用Huber M 估计算法实现状态的量测更新. 进一步结合高阶球面-径向容积准则的状态预测模块构成基于 Huber的高阶容积卡尔曼跟踪算法. 重点分析了Huber代价函数的调节因子对算法跟踪性能的影响. 通过对纯方位目标跟踪和再入飞行器跟踪两个实例验证了所提算法的跟踪性能优于传统高阶容积卡尔曼滤波算法.

Huber-based high-degree cubature Kalman tracking algorithm

In recent decades, nonlinear Kalman filtering based on Bayesian theory has been intensively studied to solve the problem of state estimation in nonlinear dynamical system. Under the Gaussian assumption, Bayesian filtering can provide a unified recursive solution to the estimation problem that is described as the calculation of Gaussian weighted integrals. However it is typically intractable to directly calculate these integrals. The numerical integration methods are required from a practical perspective. Therefore, nonlinear Kalman filters are generated by different numerical integrations. As a representative of nonlinear Kalman filter, cubature Kalman filter (CKF) utilizes a numerical rule based on the third-degree spherical-radial cubature rule to obtain better numerical stability, which is widely used in many fields, e.g., positioning, attitude estimation, and communication. Target tracking can be generalized as the estimations of the target position, the target velocity and other parameters. Hence, nonlinear Kalman filters can also be used to perform target tracking, effectively. Since the CKF based on the third-degree cubature rule has a limited accuracy of estimation, it is necessary to find a CKF based a cubature rule with higher accuracy in the case of target tracking system with a large uncertainty. High-degree cubature Kalman filter is therefore proposed to implement state estimation due to its higher numerical accuracy, which is preferred to solve the estimation problem existing in target tracking. To improve the filtering accuracy and robustness of high-degree cubature Kalman filter, in this paper we present a new filtering algorithm named Huber-based high-degree cubature Kalman filter (HHCKF) algorithm. After approximating nonlinear measurements by using the statistical linear regression model, the measurement update is implemented by the Huber M estimation. As a mixed estimation technique based on the minimum of l₁-norm and l₂-norm, the Huber estimator has high robustness and numerical accuracy under the assumption of Gaussian measurement noises. Therefore, the Huber-based high-degree cubature Kalman tracking algorithm is generated by combining the state prediction based on the fifth-degree spherical radial rule. In this paper, the influence of tuning parameter on the tracking performance is discussed by simulations. Simulations in the context of bearings only tracking and reentry vehicle tracking demonstrate that the new HHCKF can improve the tracking performance significantly.