摘要
头部姿势估计在人机交互中有着重要的意义,为此提出了一种在单目图像序列中估计三维头部姿势的方法.运用块匹配方法跟踪人脸特征.结合人脸的几何知识提出了一个匹配评价标准.并提出了一种基于特定人脸几何信息和透视投影模型的算法来估计头部姿势,不需使用三维模型和任何面部标记.实验结果表明,该方法对于头部姿势估计是有效的.
For head pose estimation plays an important role in human machine interface, the authors propose an approach to estimate 3D head pose from a monocular sequence. A block matching criterion is designed and the effect of block matching for feature tracking is improved. An algorithm based on facial geometry information and projective model is presented to estimate the head pose. The experimental results show that our method is effective for head pose estimation.
出处
《北京工业大学学报》
CAS
CSCD
北大核心
2005年第2期220-224,共5页
Journal of Beijing University of Technology
基金
Projeet Supported by Beijing Foundation of Natural Science(D070601-01).
关键词
面部跟踪
块匹配
姿势估计
几何透视模型
feature tracking
block-matching
pose estimation
geometry projective model
