一种机动目标的跟踪算法研究

Research on Maneuvering Target Tracking Algorithm Based on PF-RBF-Neural-Networks

目前在机动目标跟踪领域中讨论比较多的算法包括扩展卡尔曼滤波算法、强跟踪算法、UKF算法和粒子滤波算法;扩展卡尔曼滤波算法对非线性方程进行一阶线性阶处理,这种近似所带来的误差会随着非线性化程度的严重而越来越显著,最终造成滤波器的发散;而粒子滤波作为一种基于蒙特卡洛方法的贝叶斯滤波算法,虽然不需要对非线性方程进行一阶近似,但是其计算负担过于繁重,很难满足实时性的要求,提出了一种基于粒子滤波(PF)的径向基(RBF)神经网络(PF-RBF-Neural-Networks)机动目标跟踪算法,该算法能够获得和粒子滤波几乎相同的跟踪精度,同时又克服了粒子滤波计算量大的缺陷,仿真结果验证了该算法的有效性和可行性。

At present, a lot of non-linear filter algorithms were introduced into solving maneuvering target tracking issue, which ineluded extended kalman filter （EKF）, strong tracking filter （STF）, unscented kalman filter （UKF） and particle filter （PF）. EKF algorithm dealt non--linear equation with the first step of Taylor series, such approximate error would be more and more remarkable as non-- linear degree serious, the filter would take on dispersed state finally. PF algorithm is a kind of Bayes filter based on Monte Carlo method, which is no need of approximate disposal with non-linear equation, but the heavy burden of calculating is more serious that is difficult to meet the request for real time character. A maneuvering target tracking algorithm is introduced based on PF--RBF--Neural--Networks, which can get the tracking accuracy nearly the same as PF algorithm, and overcome the defect of PF algorithm at the same time. The simulation experiment shows its validity and feasibility.