L0正则化增量正交投影非负矩阵分解的目标跟踪算法  

Object tracking via incremental orthogonal projective non-negative matrix factorization with L0 regularization

作　　者：王海军[1,2] 葛红娟[1] 

机构地区：[1]南京航空航天大学民航学院,江苏南京211106 [2]滨州学院山东省高校航空信息技术重点实验室,山东滨州256603

出　　处：《系统工程与电子技术》2016年第10期2428-2434,共7页Systems Engineering and Electronics

基　　金：山东省自然科学基金高校;科研单位联合专项计划(ZR2015FL009);滨州市科技发展计划(2013ZC0103);滨州学院科研基金(BZXYG1524)资助课题

摘　　要：针对传统跟踪算法不能在复杂场景下进行有效跟踪的问题,提出一种基于L0正则化增量正交投影非负矩阵分解(incremental orthogonal projective non-negative matrix factorization,IOPNMF)的目标跟踪算法。在粒子滤波框架下采用IOPNMF算法在线获得跟踪目标基于部分的表示以构建模板矩阵,然后将每帧中的候选样本建立基于模板矩阵的线性表示,对表示系数进行L0正则化约束,并提出快速数值解法,同时引入粒子筛选机制,加快跟踪速度。实验结果表明,新算法能够解决跟踪过程中出现的遮挡、光照变化、运动模糊等影响跟踪性能的因素,具有较高的平均覆盖率和较低的平均中心点误差。In order to solve the problem of tracking failure in complex scenes by traditional object tracking algorithms, a new object tracking algorithm based on incremental orthogonal projective non-negative matrix factorization （IOPNMF） with L0 regularization is presented. In the framework of the particle filter, template matrix is obtained on the part-based representation of the tracked object by the IOPNMF algorithm. The candidates in each frame are linearly representated by the template matrix. The representation coefficients are constrained by the L0 regularization while a fast numerical solution is proposed. At the same time, the particle selection mechanism is introduced to speed up the tracking speed. Experimental results show that the proposed algorithm can effectively overcome the influence of occlusion, illumination change, and motion blur, with higher average overlap rate and lower average center point error.

关 键 词：目标跟踪 L0正则化 粒子筛选 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]