中文摘要：

为了对复杂场景中的多视角旋转人脸进行精确跟踪,提出了一种基于子空间特征模型的多视角人脸跟踪算法.该算法根据不同的人脸姿态建立多个离线人脸模型并自动进行在线学习,同时,针对人脸跟踪提出了新的自适应粒子滤波框架,确定人脸状态.实验结果表明,该算法能够准确跟踪多视角变尺度人脸,并实时分辨人脸姿态,对人脸的旋转、尺度变化以及环境影响不敏感,具有很强的鲁棒性和精确性.

英文摘要：

This paper proposed a new algorithm based on sub-space model for robust multiview face tracking under complex environment.It combines the offline face models and online self-learning models.The paper also presented new self-adaptive particle filter algorithm to track face.The experiments demonstrate that this algorithm can handle multi-view and multi-scale face tracking steadily in complex environment.