摘要
基于单视觉主动红外光源系统,提出了一种视线检测方法.在眼部特征检测阶段,采用投影法定位人脸;根据人脸对称性和五官分布的先验知识,确定瞳孔潜在区域;最后进行人眼特征的精确分割.在视线方向建模阶段,首先在头部静止的情况下采用非线性多项式建立从平面视线参数到视线落点的映射模型;然后采用广义回归神经网络对不同头部位置造成的视线偏差进行补偿,使非线性映射函数扩展到任何头部位置.实验结果及在交互式图形界面系统中的应用验证了该方法的有效性.
A gaze tracking method is proposed based on active infrared source system. In the eye feature detection stage, projection method is applied to locate face. Symmetry axis of face is detected, potential region of pupil is found based on the knowledge about facial structure, and human eyes features can be precise segmented. In the eye-gaze modeling stage, a model with nonlinear polynomial to map the gaze parameters to gaze point under the circumstances of static head is set up. Then the deviation of gaze caused by different head position is compensated by GRNN neural network, so that nonlinear mapping function can be extended to any head position. So gaze can be estimated accurately under head movement. The experiment results and the applications in interactive graphical system show the effectiveness of this method.
出处
《控制与决策》
EI
CSCD
北大核心
2009年第9期1345-1350,共6页
Control and Decision
基金
国家自然科学基金项目(60574090)
国家863计划项目(2007AA01Z160)
博士后科学基金项目(20064000400)
