\r田 丹^{1, 2}，张国山¹，孙申申\r²\r
\r
AuthorsHTML:\r田 丹^{1, 2}，张国山¹，孙申申\r²\r
\r
AuthorsListE:\rTian Dan^{1, 2}，Zhang Guoshan ¹，Sun Shenshen \r²\r
\r
AuthorsHTMLE:\rTian Dan^{1, 2}，Zhang Guoshan ¹，Sun Shenshen \r²\r
\r
Unit:\r\r1. 天津大学电气自动化与信息工程学院，天津 300072；\r
\r\r2. 沈阳大学信息工程学院，沈阳 110044\r
\r
\r
Unit_EngLish:\r1. School of Electrical and Informatin Engineering，Tianjin University，Tianjin 300072，China；
2. School of Information Engineering，Shenyang University，Shenyang 110044，China\r
\r
Abstract_Chinese:\r现有的低秩稀疏优化目标跟踪算法容易存在下述两方面问题：①需要求解大量l₁ 优化问题，计算复杂度高．②在目标突变运动情况下，经常出现跟踪漂移现象．为此，提出一种反向低秩稀疏约束下基于融合最小绝对值收缩和选择算子(Lasso)的目标跟踪算法．首先，建立目标表观的反向稀疏表示描述，利用候选粒子反向稀疏表示目标模板，将在线跟踪中l₁ 优化问题的数目由候选粒子数简化为1．其次，将融合Lasso 模型引入到目标跟踪建模中，约束表示系数差分的绝对值之和，保证表示系数稀疏性的同时，使其连续性差异亦稀疏．从而限制目标表观在相邻帧间具有较小差异，但允许个别帧间存在较大差异性，以适应目标的突变运动．再次，利用核范数凸近似低秩约束，限制目标表观的时域相关性，以适应目标的外观变化．实验结果表明，面向具有严重遮挡、光照和尺度变化、目标突变运动等挑战性的标准跟踪数据集，提出算法能完成复杂场景下的跟踪任务．与目前几种热点算法进行定性与定量分析比较，提出算法具有更高的跟踪精度和较快的跟踪速度，特别是在目标突变运动情况下具有更好的鲁棒性．\r
\r
Abstract_English:\rThe existing low-rank sparse optimization-based object tracking algorithms usually feature the following problems：① requirement for solutions to considerable optimization problems l₁ and high computation complexity；② usually occurring drift phenomenon when facing object abrupt motion. Therefore，a fused least-absolute shrinkage and selection operator-based object tracking algorithm with inverse low-rank and sparse constraint is proposed. First，an inverse sparse representation formulation for object appearance is built using candidate particles to represent the target template inversely，simplifying the number of optimization problem l₁ for online tracking from candidate particle number to one. Second，the fused Lasso model is introduced to the object tracking modeling to constraint the absolute value sum of the sparse coefficient difference，resulting in the sparse representation coefficient and continuity difference. This condition can restrict the object appearance with minimal difference between the consecutive frames but allow for the greater variation between the individual frames to adopt an abrupt motion. Third，the low-rank constraint based on the nuclear norm is used to restrain the temporal correlation of the objective appearance to adopt the appearance change. The experimental results show that when subjecting the benchmark tracking dataset to challenging situations under serious occlusion，illumination and scale variation，and object abrupt motion，the proposed algorithm can perform tracking in complicated scenes. Both qualitative and quantitative evaluations demonstrate that the proposed algorithm features higher precision and faster speed against other state-of-art algorithms，especially exhibiting better adaptability for object abrupt motion.\r
\r
Keyword_Chinese:目标跟踪；反向稀疏表示；低秩约束；融合Lasso；粒子滤波\r

Keywords_English:object tracking；inverse sparse representation；low-rank constraint；fused Lasso；particle filter\r


PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6377