机动目标跟踪支持向量回归学习新方法

Novel approach in maneuvering target tracking based on support vector regression

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。

Aiming at the problem that the motion modeling,the control input modeling and the noisemodeling of the intense maneuvering target is not so accurate that bringing out the low accuracy in the automotive radar target tracking task, a novel method based on the support vector regression （SVR） for the manuevering target tracking is proposed here. Through the online learning on the Frobenius norm of the residuals of theory innovation covarivance matrix and the actual innovation covariance maxtrix based on the constant accleration （ CA ） model, a scaling factor of the process noise covariance matrix is estimated to correct the previous motion model and the noise model in real time. Experiments indicate that the approach has the low dependency on the motion model and the noise model,is far superior to the constant acceleration model （ C A ）, and it has higher accuracy and adaption than the Singer model.