基于改进粒子滤波的视觉目标跟踪  

Visual object tracking based on improved particle filter

作　　者：赵康[1] 王正勇[1] 何小海[1] 郑新波 田刚[3] ZHAO Kang;WANG Zheng-Yong;HE Xiao-Hai;ZHENG Xin-Bo;TIAN Gang(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China;Dongguan Institute of Advanced Technology,Dongguan 523000,China;Xinjiang Oilfield Company,Karamay 834000,China)

机构地区：[1]四川大学电子信息学院,成都610065 [2]东莞前沿技术研究院,东莞523000 [3]新疆油田公司,克拉玛依834000

出　　处：《四川大学学报（自然科学版）》2019年第5期875-882,共8页Journal of Sichuan University(Natural Science Edition)

基　　金：成都市产业集群协同创新项目(2016-XT00-00015-GX);国家自然科学基金(11176018);东莞市社会科技发展项目(2017507102428)

摘　　要：针对目标跟踪算法在精度和鲁棒性上的要求,提出一种基于改进粒子滤波的视觉目 标跟踪算法.首先,建立多种特征来描述目标外观模型,并对各特征分量的加权系数进行自适 应调节;然后,利用分类重采样方法解决原始重采样方法中的粒子退化和匮乏问题;最后,提出 一种新的模板更新机制,自适应选取运动模板或原始模板.实验结果表明,改进后的算法在具 有挑战的跟踪视频序列上实验,具有良好的跟踪精度和鲁棒性,能够应对视频图像分辨率不 高、目标转动变化、部分遮挡等复杂条件.Aiming at the accuracy and robustness of target tracking algorithm,we propose a visual targettracking algorithm based on improved particle filter.Firstly,a variety of features are established to describe the target appearance modeland the weighted coefficients of each feature component are adaptively adjusted.Then,we exploit the classification resampling method to solve the problems of particle degradation and scarcity in the original resampling method.Finally,a new template updating mechanism is introduced,whichcan adaptively select movingor original templates.The experimental resultdemonstrate that the improved algorithm has good tracking accuracy and robustness on the challenging tracking video sequences,and it can cope with some complex conditions such as low resolution of video images,rotation change of target,partial occlusion and so on.

关 键 词：模板更新 部分遮挡 粒子退化 重采样方法 运动模板 图像分辨率 外观模型 视觉目标跟踪 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]