机动目标跟踪支持向量回归学习新方法

发布时间：2018-03-31 00:03

本文选题：机动目标跟踪　切入点：支持向量回归　出处：《南京理工大学学报》2017年02期

【摘要】：针对强机动性车辆目标的运动建模、控制输入建模和噪声建模的不精确导致的汽车雷达目标跟踪滤波精度低的问题,该文提出了基于支持向量回归(SVR)的机动目标跟踪滤波新方法。在常加速度(CA)模型的基础上,对理论新息协方差与实际新息协方差残差的Frobenius范数在线学习,获得过程噪声协方差的自适应调节因子,实时调整运动模型。对汽车雷达目标跟踪系统的仿真实验表明,该文算法降低了汽车雷达目标跟踪滤波对车辆运动模型和噪声模型的依赖程度,在强机动目标跟踪滤波性能上优于CA模型,比Singer模型具有更强的机动适应性和更高的精度。
[Abstract]:Aiming at the problem of low filtering accuracy caused by imprecision of control input modeling and noise modeling, which is caused by the motion modeling of vehicle targets with strong maneuverability, In this paper, a new method of maneuvering target tracking filtering based on support vector regression (SVR) is proposed. On the basis of constant acceleration model, the Frobenius norm of theoretical innovation covariance and real innovation covariance residuals is studied online. The adaptive adjustment factor of process noise covariance is obtained, and the motion model is adjusted in real time. The simulation results of vehicle radar target tracking system show that, The algorithm reduces the dependence of vehicle radar target tracking filtering on vehicle motion model and noise model, and is superior to CA model in strong maneuvering target tracking performance, and has stronger maneuverability and higher precision than Singer model.
【作者单位】：南京理工大学计算机科学与工程学院;
【基金】：国家“863”高技术研究计划资助项目(2015AA8106043) 国家自然科学基金(61402237;61302156)
【分类号】：TP181;U463.6;U495

【相似文献】