基于两阶段稀疏表示的稳健快速视觉跟踪  

Robust Fast Visual Tracking Based on Two-Stage Sparse Representation

作　　者：刘文琢 袁广林[2] 薛模根[1] 

机构地区：[1]陆军军官学院偏振光成像探测技术安徽省重点实验室,安徽合肥230031 [2]陆军军官学院十一系,安徽合肥230031

出　　处：《光学学报》2016年第12期176-182,共7页Acta Optica Sinica

基　　金：国家自然科学基金(61175035;61379105)

摘　　要：L1跟踪对局部遮挡具有较好的稳健性,但存在对模板中的离群信息比较敏感和计算速度慢的问题。针对这两个问题,提出了两阶段稀疏表示模型,并基于块坐标优化原理设计了相应的快速求解算法。在第一阶段,该算法利用局部约束线性编码,求解目标模板表示系数,在第二阶段,该算法利用软阈值操作,求解小模板表示系数。以粒子滤波为跟踪方法,结合提出的模型和算法实现了稳健快速的视觉跟踪。利用标准图像序列对提出的方法进行了验证,实验结果表明,提出的跟踪方法在稳健性和跟踪速度方面均优于现有跟踪方法。The L1 tracker has good robustness towards partial occlusion,but the L1 tracker is sensitive to the outliers from the target templates and has slow computation speed.Aiming at these two problems,we propose a two-stage sparse representation model and design a relevant fast solution algorithm based on the block coordinate optimization theory.At the first stage,the algorithm uses the locality-constrained linear coding to solve the coefficients of the target templates.At the second stage,the algorithm uses the soft shrinkage operator to solve the coefficients of the trivial templates.Based on particle filtering method,the representation model and the algorithm are combined to achieve the robust fast visual tracking.The standard image sequences are used to verify the proposed method,and the results of the experiment show that the proposed tracking method outperforms the state-of-the-art trackers in terms of the robustness and the tracking speed.

关 键 词：机器视觉 目标跟踪 两阶段稀疏表示 粒子滤波 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]