基于低秩稀疏表示的加权增量目标跟踪算法  

Weighted Increment Target Tracking Algorithm Based on Low-rank Sparse Representation

作　　者：应艳丽 张家树[1] 瞿遥 

机构地区：[1]西南交通大学信息科学与技术学院,成都611756

出　　处：《计算机工程》2016年第1期248-253,共6页Computer Engineering

摘　　要：现实中目标物体所处背景往往受到遮挡、光照变化等复杂环境的影响,容易导致跟踪漂移。为提高目标跟踪的精确度,以加权增量主成分分析算法为模板更新机制,提出一种新的目标跟踪算法。通过主成分分析基向量模板和平方模板对变化的目标外观进行线性表示,把目标跟踪问题视为低秩稀疏优化问题,求解低秩稀疏解,得到候选目标重构系数,将基于重构误差后验概率最小的跟踪目标作为当前跟踪结果,并在增量主成分分析算法更新基向量模板过程中,对每个跟踪目标进行加权,从而有效抑制低质量目标样本的影响。实验结果表明,与增量视觉跟踪算法、最小软阈值跟踪算法等相比,该算法在复杂环境的目标跟踪中具有较好的鲁棒性。The objects under complex environment often are affected by occlusion illumination changes and so on which lead to tracking drift.In order to improve the accuracy of visual tracking,a novel robust visual tracking algorithm is introduced by using Low-rank Sparse Representation（LRSR） and using Weighted Incremental Principal Component Analysis（WIPCA） method as the model updating mechansim.This algorithm uses Principal Component Analysis（PCA）base vector templates and square templates to model the target appearance,and considers the object tracking problem as the low-rank sparse optimization problem.It solves the low rank sparse solutions,gets candidate target reconstruction coefficient,then it considers the tracking target which is based on the reconstruction error and the posterior probability is the smallest as the current tracking results.At last during the process of WIPCA updating base vector template,it weights each trace object,in order to effectively restrain the influence of low quality target samples.Experimental results show that compared with Incremental Learning for Robust Visual Tracking（IVT） and Least Soft-thresold Squares Tracking（LSST） /etc,this algorithm has better robustness in the target tracking of complex environment.

关 键 词：目标跟踪算法 低秩稀疏表示 平方模板 增量主成分分析 加权增量 重构系数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]