基于GRNN的拟蒙特卡洛粒子滤波目标跟踪算法  

Quasi-Monte Carlo Particle Filter Algorithm for Target Tracking Based on GRNN

作　　者：陈志敏[1] 薄煜明[1] 吴盘龙[1] 徐文康[1] 刘正凡[1] 

机构地区：[1]南京理工大学自动化学院,江苏南京210094

出　　处：《信息与控制》2012年第6期760-766,773,共8页Information and Control

基　　金：国家自然科学基金资助项目(61104196);高等学校博士学科点专项科研基金资助项目(20113219110027)

摘　　要：针对拟蒙特卡洛粒子滤波(QMC-PF)算法计算量大,难以满足雷达目标跟踪实时性需要的问题,提出了一种适用于雷达机动目标跟踪的新型拟蒙特卡洛粒子滤波算法(NQMC-PF).该算法利用QMC方法生成权重较大粒子的低差异性的子代粒子来替换低权重粒子,保证了样本的质量和多样性,同时利用广义回归神经网络(GRNN)计算子代粒子的权重,提高了滤波的精度和速度.实验结果表明,该算法的计算精度高于标准拟蒙特卡洛粒子滤波算法,同时运算时间短,实时性好,能够应用于对雷达目标的跟踪上.Quasi-Monte-Carlo particle filter （QMC-PF） can not meet the requirement of target tracking because of the high computational complexity. A novel Quasi-Monte-Carlo particle filter （NQMC-PF） algorithm for maneuvering radar target tracking is proposed. The algorithm applies QMC algorithm to generating the low-discrepancy offsprings of the the particles with heavy weight to replace the particles with low weight, which guarantees the quality and diversity of samples. Generalized regression neural network （GRNN） is used to calculate the weights of the offsprings, which improves the precision and the speed of the filter. The simulation results show that the calculation precision of the algorithm is higher than standard QMC- PF, and it possesses the advantages of short computation time and real-time standard. It can be applied to the radar target tracking.

关 键 词：粒子滤波 拟蒙特卡洛方法 广义回归神经网络(GRNN) 目标跟踪 闪烁噪声 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]