摘要： 在对给定的人脸图像序列准确提取正面人脸特征点后，利用改进的KLT方法跟踪非正面人脸图像的特征点。根据人脸形状特性，使用人脸的多个特征点作为人脸模型。在近似估计人脸姿态后，以改进的BFGS算法精确估计3D人脸空间姿态。实验结果证明，该方法可以获得唯一的3D人脸空间姿态，相比同类方法有更好的姿态估计精确度。

关键词: 人脸姿态估计, 多点模型, 特征点跟踪, 拟牛顿法

Abstract: Aiming at the face sequence, the face feature points are tracked by using modified Kanade-Lucas-Tomasi(KLT) after extracting face feature points accurately. According to characteristics of face shape, it uses a number of face feature points as the face model. After estimating the face pose approximately, 3D space face pose is estimated accurately via modified Broyden-Fletcher-Goldfarb-Shanno(BFGS). Experimental result shows that the method can gain only 3D face pose, and has better estimation accuracy compared with similar methods.

Key words: estimation of face pose, multi-point model, feature points tracking, Broyden-Fletcher-Goldfarb-Shanno(BFGS)

中图分类号: 

• TP391