核粒子滤波目标跟踪算法  

Kernel Particle Filtering Target Tracking Algorithm

作　　者：初红霞[1,2] 谢忠玉 王科俊[2] 杜娟[1] CHU Hongxia;XIE Zhongyu;WANG Kejun;DU Juan(College of Electrical and Information Engineering, Heilongjiang Institute of Technology, Harbin 150001, China;College of Automation, Harbin Engineering University, Harbin 150001, China)

机构地区：[1]黑龙江工程学院电气与信息工程学院,哈尔滨150001 [2]哈尔滨工程大学自动化学院,哈尔滨150001

出　　处：《北京工业大学学报》2018年第6期855-861,共7页Journal of Beijing University of Technology

基　　金：哈尔滨市科技局科技创新人才研究专项资金资助项目(2017RAQXJ134);国家自然科学基金资助项目(61573114);黑龙江省自然科学基金资助项目(A201418)

摘　　要：针对粒子滤波的退化问题以及使用单一特征跟踪鲁棒性不高的缺点,提出了一种基于多特征融合的核粒子滤波目标跟踪方法.首先在核粒子滤波中提出新的权值更新方法,然后将颜色和纹理特征在核粒子滤波方法框架下进行融合实现鲁棒跟踪.对颜色和纹理特征的计算分别采用空间直方图和积分直方图的计算方法,这2种计算方法有效地克服了2种特征自身存在的缺点.该算法提高了采样效率,解决了粒子滤波的计算量大和粒子退化问题.最后应用本文算法在复杂背景和严重遮挡等情况下的目标序列上进行了测试,实验表明该算法不仅能准确地跟踪目标,而且能很好地处理目标遮挡等问题.To solve the problem of particle filter’s degradation and lower tracking robustness with single feature application,a kernel particle filter tracking method was proposed by multi-feature fusion. Firstly,the method of new weight updating in kernel particle filter was put forward. Then the robust tracking was achieved by integrating the color and texture feature under the framework of kernel particle filter method.Spatiograms and integral histogram was used respectively to calculate color and texture feature. The disadvantages of their own were effectively overcome by the two kinds of calculation methods for two characteristics. The sampling efficiency was improved by using the algorithm and the larger calculation problem of particle filter and particle degradation was solved. Finally target tracking experiment was conducted by adopting the methods for complex background and serious occlusion circumstances.Experimental results show that the proposed algorithm can track target accurately and may well deal with object occlusion.

关 键 词：核粒子滤波 多特征融合 直方图 目标跟踪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]