利用N型序贯蒙特卡罗PHD滤波的多目标跟踪  

Multi target tracking using N-type sequential monte carlo PHD filter

作　　者：黄兴 HUANG Xing(College of Big Data and software,College of Mobile Telecommunications,Chongqing University of Posts and Telecom,Hechuan,Chongqing 401520,China)

机构地区：[1]重庆邮电大学移通学院大数据与软件学院,重庆合川401520

出　　处：《光学技术》2021年第5期613-621,共9页Optical Technique

基　　金：重庆市教育委员会科学技术研究项目(KJQN201902002)。

摘　　要：针对传统跟踪算法受背景杂波与不同目标类型混淆影响较大的问题,提出了一种利用N型序贯蒙特卡罗PHD(SMC-PHD)滤波的多目标跟踪。基于随机有限集理论提出一种新的扩展概率假设密度(PHD)滤波器模型,即N型PHD滤波,从概率生成函数中导出更新的PHD;利用序贯蒙特卡罗(SMC)方法产生随机样本,结合N型PHD滤波,提出N型SMC-PHD滤波算法,以减少不同目标类型的混淆检测;利用量测驱动的SMC-PHD滤波器,通过门控方法对量测的属性进行区分,从而抑制杂波对已有目标跟踪的影响。基于MATALB仿真平台,在MOT17以及VS-PETS’2003足球视频数据集中对所提算法进行实验论证,结果表明,所提方法的平均最优子模式分配(OSPA)距离显著降低,且在高杂波、目标类型较多的情形下具有较高的可靠性。Aiming at the problem that the traditional tracking algorithm is greatly affected by the confusion between background clutter and different target types,a multi-target tracking method using n-type sequential Monte Carlo PHD(SMC-PHD)filter is proposed.Firstly,based on the stochastic finite set theory,a new extended probability hypothesis density(PHD)filter model,namely n-type PhD filter,is proposed,and the updated PHD is derived from the probability generating function.Then,the sequential Monte Carlo(SMC)method is used to generate random samples.Combined with N-type PHD filter,N-type SMC-PHD filtering algorithm is proposed to reduce the confusion detection of different target types.Finally,the measurement driven SMC-PHD filter is used to distinguish the measured attributes by the gating method,so as to suppress the influence of clutter on the existing target tracking.Based on MATLAB simulation platform,the proposed algorithm is tested in MOT17 and VS-PETS’2003 soccer video data sets.The results show that the average optimal sub mode allocation(OSPA)distance of the proposed method is significantly reduced,and it has high reliability in the case of high clutter and more target types.

关 键 词：多目标跟踪 序贯蒙特卡罗 N型PHD滤波 门控方法 背景杂波 OSPA 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]