基于时间序列自适应建模的粒子滤波动态目标跟踪  

Particle filter tracking moving target based on adaptive modeling by time series

作　　者：杨淑莹[1] 邓飞[2] 丁承君[3] 张迎[1] 吴涛[1] 

机构地区：[1]天津理工大学智能计算及软件新技术重点实验室,天津300191 [2]佛罗里达国际大学工程和计算机学院,迈阿密33174 [3]河北工业大学机械工程学院,天津300130

出　　处：《光电子．激光》2011年第11期1725-1730,共6页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金资助项目(61001174);天津市高等学校科技发展基金资助项目(20071308)

摘　　要：在目标跟踪中为达到目标的运行模型与实际轨迹相符,本文提出基于时间序列自适应建模的粒子滤波算法(TS_PF)。采用时间序列方法动态构建预测模型,并将粒子滤波算法中一系列加权粒子以该模型进行状态转移,运用粒子滤波重采样技术,使预测误差进一步减小,预测精度逼近最优估计。仿真实验表明在粒子滤波算法中采用时间序列自适应建模,能够实时反映目标的运行轨迹,克服了单一模型无法准确跟踪动态目标,以及交互多模型需要先验知识的缺陷,提高了动态目标跟踪的准确性。In order to make good match for a target moving model and its actual track, time series-basect adaptive modeling in particle filter （TS PF） is presented in this paper. The prediction model is dynami cally made by the time series analysis. The states of weighted particles in particle filter are transferred according to the prediction model. By the resample technique of particle filter, the prediction error is further reduced and the prediction accuracy approximates to the optimal estimation. The simulations show that the time series adaptive modeling in particle filter can make good match with its actual track and overcome the defects of a single model＇ s inaccuracy and IMM＇s apriorism. The accuracy of the dynamic target tracking is improved by TS_PF.

关 键 词：目标跟踪 时间序列 粒子滤波 

分 类 号：TN391[电子电信—物理电子学]