基于时序特性的自适应增量主成分分析的视觉跟踪  

Adaptive Incremental Principal Component Analysis Visual Tracking Method Based on Temporal Characteristics

作　　者：蔡自兴[1] 彭梦[1] 余伶俐[1] 

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《电子与信息学报》2015年第11期2571-2577,共7页Journal of Electronics & Information Technology

基　　金：国家自然科学基金重大研究计划(90820302);国家自然科学基金(61175064;61403426;61403423)~~

摘　　要：当前基于增量主成分分析(PCA)学习的跟踪方法存在两个问题,首先,观测模型没有考虑目标外观变化的连续性;其次,当目标外观的低维流行分布为非线性结构时,基于固定频率更新模型的增量PCA学习不能适应子空间模型的变化。为此,该文首先基于目标外观变化的连续性,在子空间模型中提出更合理的目标先验概率分布假设。然后,根据当前跟踪结果与子空间模型之间的匹配程度,自适应调整遗忘比例因子,使得子空间模型更能适应目标外观变化。实验结果验证了所提方法能有效提高跟踪的鲁棒性和精度。Existing visual tracking methods based on incremental Principal Component Analysis （PCA） learning have two problems. First, the measurement model does not consider the continuation characteristics of the object appearance changes. Second, when the manifold distribution of target appearance is non-linear structure, the incremental principal component analysis learning based on fixed update frequency can not adapt to changes of subspace model. Therefore, the more reasonablea priori probability distribution of targets is proposed based on the continuity of the object appearance changes in the subspace model. Then, according to the matching degree between the current tracking results and the subspace model, the proposed method adaptively adjusts forgetting factor, in order to make the subspace model more adaptable to the object appearance change. Experimental results show that the proposed method can improve the tracking accuracy and robustness.

关 键 词：视觉跟踪 主成分分析 增量子空间学习 遗忘因子 自适应增量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]