基于稀疏度约束与动态组结构稀疏编码的鲁棒视觉跟踪  

Sparsity-Constrained and Dynamic Group Structured Sparse Coding for Robust Visual Tracking

作　　者：袁广林[1] 薛模根[2] 

机构地区：[1]解放军陆军军官学院十一系,安徽合肥230031 [2]解放军陆军军官学院科研部,安徽合肥230031

出　　处：《电子学报》2015年第8期1499-1505,共7页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61175035;No.61379105);中国博士后科学基金(No.2014M562535)

摘　　要：目标编码系数的稀疏性使得L1跟踪成为解决遮挡目标跟踪的有效方法之一,但是现有稀疏编码算法没有利用L1跟踪中编码系数的特殊稀疏结构.本文基于目标模板系数稀疏度约束要求和小模板系数的空间连续性结构,利用块坐标优化原理提出一种两阶段稀疏编码算法用于视觉跟踪.在第一阶段,该算法利用正交匹配追踪求解具有约束稀疏度的目标模板系数,在第二阶段,该算法利用动态组稀疏编码求解具有空间连续性的小模板系数.在粒子滤波框架下,利用提出的稀疏编码算法实现了鲁棒的视觉跟踪.实验结果表明本文提出的跟踪方法比现有跟踪方法具有更强的鲁棒性和较高的跟踪精度.L1 tracker is one of the most effective methods in dealing with the occlusions for sparseness of coding coefficients of objects. However, the existing sparse coding algorithms do not use special sparse structure of coding coefficients in L1 tracker. In this paper, we propose a two-stage sparse coding algorithm for visual tracking based on constrained sparsity of target template coefficients and spatial continuity structure of trivial template coefficients with block coordinate optimization theory. At the first stage, the algorithm solves sparsity-constrained coding coefficients on target template set using orthogonal matching pursuit. At the second stage, the algorithm finds sparse coding coefficients with spatial continuity on trivial template set via dynamic group sparse coding. Robust visual tracking is achieved using the proposed sparse coding algorithm in particle filter framework. The experimental results demonstrate that the proposed tracking method has better robustness and higher precision than the state-of- the-art trackers.

关 键 词：L1跟踪 稀疏编码 约束稀疏度 空间连续性结构 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]