基于最小二乘的多特征概率数据关联EM方法  

Expectation Maximization Algorithm for Multi-Feature Aided Probabilistic Data Association Based on Least Square

作　　者：曾斯[1] 江朝抒[1] 陈祝明[1] 

机构地区：[1]电子科技大学电子工程学院,四川成都611731

出　　处：《信号处理》2011年第5期690-696,共7页Journal of Signal Processing

基　　金：自然科学基金民航联合基金重点项目(60736045);国家自然科学基金委与中国工程物理研究院资助项目(10776003)

摘　　要：提出了一种多目标多特征信息的数据关联算法。在强噪声密集杂波环境下,针对传统PDA算法对多目标跟踪时出现精度较差的问题,在跟踪过程中融入目标的特征状态信息,利用期望极大化(EM)算法对目标状态估计的最小二乘(LS)误差函数迭代求最小,将目标运动状态和特征联合的关联概率作为估计参数不断修正,从而获得对目标状态的最优估计。仿真结果表明,该算法能够增强区分目标和杂波的能力,减小相近特征量测所引起的跟踪误导,弱化对检测概率的依赖性,显著并稳定地提高对目标的跟踪精度。A multi-feature aided probabilistic data association is developed in this paper.In the environment of high noise and high-density clutter,tracking multi-target used standard PDA algorithm may occur a poor association accuracy.Aiming at this problem, we incorporate target feature information into the tracking process,and use the expectation maximization（EM）algorithm to compute the least square＇s error function of estimating target states iteratively.Via gradually modifying the estimated parameters that are target states and multi-feature aided associated probabilities,the target states could be estimated more accurately.The simulated results show that the new algorithm can strengthen the discrimination between targets and clutter,reduce the missed tracking or false tracking due to similar feature information,weaken the dependence on the probability of detection,and improve the estimated accuracy of target states significantly.

关 键 词：期望极大化(EM) 最小二乘 多特征信息 目标跟踪 

分 类 号：TN953[电子电信—信号与信息处理]