基于无先验探测概率的改进PHD多目标跟踪算法  

Improved PHD multi-target tracking algorithm based on non-prior detection probability

作　　者：余哲翔 陈思汉 白傑 YU Zhe-xiang;CHEN Si-han;BAI Jie(School of Automotive Studies,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学汽车学院,上海201804

出　　处：《传感器与微系统》2018年第8期120-123,共4页Transducer and Microsystem Technologies

基　　金：国家重点研发计划资助项目(2016YFB0100901)

摘　　要：针对传统高斯混合—概率假设密度(GM-PHD)滤波器使用无先验检测概率和无效量测造成的性能下降问题,提出一种基于时变卡尔曼滤波(TV-KF)算法的多目标PHD滤波器。通过使用椭球门限对目标集合和量测集合进行预关联,将未知探测概率转化为目标的丢失与接收事件;将目标分为匹配存活目标,匹配新生目标和未匹配目标3类,对匹配目标,仅使用目标门限内的量测更新目标状态,对未匹配目标的权重进行衰减。仿真实验表明:所提出算法有效可行,综合性能优于传统GM-PHD算法。Aiming at problem that performance degradation problem of Gaussian mixture probability hypothesis density （GM-PHD） filter caused by non-prior detection probability and invalid measurement, a multi-target PHD filter based on the time-varying Kalman filtering （TV-KF） algorithm is proposed. Elliptical gate is used to pre- associate target sets and measurement sets, unknown detection probability is converted to and target loss and receiving event. Targets are divided into matched-survival targets, matched-birth targets and non-matched targets. For matched targets,only the measurements in threshold are used to update target state; for non-matched targets, their weights are decayed. The simulation results show that the proposed algorithm is feasible, effective and outperforms the traditional GM-PHD algorithm.

关 键 词：多目标跟踪 概率假设密度滤波器 椭球门限 量测划分 时变卡尔曼滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]