CAJ | 학술논문

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。
당전기우증량주성분분석(PCA)학습적근종방법존재량개문제，수선，관측모형몰유고필목표외관변화적련속성；기차，당목표외관적저유류행분포위비선성결구시，기우고정빈솔경신모형적증량PCA학습불능괄응자공간모형적변화。위차，해문수선기우목표외관변화적련속성，재자공간모형중제출경합리적목표선험개솔분포가설。연후，근거당전근종결과여자공간모형지간적필배정도，자괄응조정유망비례인자，사득자공간모형경능괄응목표외관변화。실험결과험증료소제방법능유효제고근종적로봉성화정도。
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.