一种基于记忆的目标跟踪算法及实验验证  

A Memory-based Algorithm for Robust Object Tracking and Simulated Experiment

作　　者：齐玉娟[1] 牛潇然 王延江[1] QI Yujuan;NIU Xiaoran;WANG Yanjiang(College of Information and Control Engineering,China University of Petroleum,Qingdao 266580,Shandong,China)

机构地区：[1]中国石油大学(华东)信息与控制工程学院,山东青岛266580

出　　处：《实验室研究与探索》2018年第11期32-36,共5页Research and Exploration In Laboratory

基　　金：山东省自然科学基金面上项目(ZR2017MF069);山东省研究生优质教育课程建设项目(SDYKC17025);中央高校基本科研业务费专项资金资助(17CX02027A)

摘　　要：粒子滤波跟踪算法通常采用颜色直方图等全局特征作为观测模板和目标模板。然而,当目标的外观发生变化或者与背景相似时,全局特征的准确性会急剧下降。为了解决上述问题,引入人类三阶段记忆机制用于构建模板空间,模板空间结合目标的局部特征和全局特征,并记忆目标的时序变化。在目标跟踪过程中,模板空间对粒子的传播加以控制,以避免粒子的退化;同时,粒子集合对特征点的匹配范围进行约束,以减少伪匹配的发生,增强目标模板的可靠性。实验结果表明,当目标外观与背景相似,或由尺度变化、光照突变、旋转等引起目标外观发生变化时,该算法仍能鲁棒地跟踪目标。In general, particle filter （PF） tracking algorithm takes global feature as observation template and target template. However, PF fails to track the target when its appearance changes or it is similar to the background. In order to solve such problems, the three-stage memory mechanism, which considers both local feature and global feature, is introduced into the construction of template space. During tracking procedure, template space controls the propagation of particles in order to avoid the degeneracy phenomenon. In turn, particles constrain the matching area of feature points to strengthen the validity of template space. Experimental results demonstrate the tracking effect of the proposed method under the conditions such as scale variation, sudden change of illumination, rotation and so on.

关 键 词：粒子滤波 三阶段记忆模型 多特征融合 模板空间 运动目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]