摘要
Condensation跟踪算法只能完成一些简单的目标跟踪。由于人脸容易出现被遮挡,姿态和表情也经常发生变化,导致人脸特征点极易跟踪失败。针对这些复杂变化,提出一种改进的Condensation人脸特征点跟踪算法。该算法对跟踪的特征点利用增量PCA方法实现特征基和均值的在线更新,同时,加入一个遗忘因子,使得在新样本的更新过程中,考虑了旧样本的存在,更新了均值。实验证明,该算法可以有效地克服复杂变化带来的影响,实现了人脸特征点的准确跟踪。
Condensation tracking algorithm can only complete some simple target tracking. As the faces are prone to be occluded,and gestures and facial expressions are often changing as well,these results in easily failing in facial features points tracking. For those complex changes,we propose an improved Condensation facial feature point tracking algorithm. It uses incremental PCA method on the feature points tracked to achieve online updating of eigenbasis and mean. Meanwhile,a forgetting factor is added which considers the existence of the old sample and updates the mean when a new sample is appeared in updating process. Experimental results show that the algorithm can effectively overcome the impact of complex changes and achieves accurate tracking on facial feature points.
出处
《计算机应用与软件》
CSCD
2015年第12期154-159,213,共7页
Computer Applications and Software
