摘要
提出一种基于面积比的人脸姿态估计方法,先分析人脸姿态发生变化时特征点之间形成的三角形的面积变化,再应用BP神经网络对位置参数和人脸姿态参数的关系进行学习,从而对人脸姿态进行估计,最后将该方法应用于虚拟环境的漫游中。实验结果表明,采用该方法对人脸转动进行估计,采用的特征点比较少,具有较高的识别率和稳定性。
A new method is proposed for face pose estimation based on the area ratio in the paper. First, the change of triangle area ratio which consist of feature points is analyzed when a face turns. Then, the BP neural network is applied to train the relationship between position parameters of area ratio of face and the parameters of pose for face pose estimation. Finally, the method is applied to roam in the virtual environment. The experimental results demonstrate that this method can estimate accurately pose using only few characteristic points, and also does well in recognition rate and stability.
出处
《中国图象图形学报》
CSCD
北大核心
2010年第8期1145-1150,共6页
Journal of Image and Graphics
基金
国家自然科学基金项目(60673186)
关键词
姿态估计
神经网络
人机交互
pose estimation, artificial neural network, human-computer interface