基于两次多项式拟合的人眼注视方向估计

Eye gazing estimation based on twice polynomial fitting

针对单相机、无辅助光源的人眼视线估计系统固有的精度低、头部运动受限等问题,本文提出了基于两次多项式拟合的、单相机、单辅助光源的视线估计方法。在准确提取辅助光源在角膜表面产生的反射光斑和虹膜中心的基础上,以反射光斑为基准点考察虹膜中心相对于反射光斑的偏移与注视点之间的映射关系。首先通过第一次多项式拟合考察注视点不变而反射光斑位置发生变化,对虹膜中心与反射光斑相对位置的影响,以此确定相对于任意位置的反射光斑,对应测试板上15个测试点的理想虹膜中心偏移。然后通过第二次多项式拟合考察当反射光斑在图像中的位置确定后,虹膜中心偏移与注视点之间的对应关系。通过两次拟合,实现了人眼注视方向的有效估计,在允许头部自由运动的前提下,测试样本的水平方向平均注视误差为1.31°,竖直方向平均注视误差为1.00°。

In this paper, we propose an eye gazing estimation method by using single camera and single light source based on twice polynomial fitting to overcome the common drawbacks of gazing estimation system with one camera but no light source,i, e. ,low accuracy and restricted head movement. After the corneal reflection spot and the iris center have been precisely located, we study the mapping relationship between the iris center displacement relative to the corneal reflection spot and the gazing point on the test board. We use the first polynomial fitting to address how the relative displacement of iris center change when the location of the reflection point changes while the gazing point is fixed. In this way,we can calculate the ideal iris center displacement relative to any given reflection point position when the testee is gazing at the fifteen test points on the board. Then we use the second polynomial fitting to study the relationship between the relative displacement of iris center and the gazing point when the location of the reflection spot is determined. By using twice polynomial fitting, we achieve a valid estimation of the gazing point on the test board with unrestricted head movement. The horizontal average error of the test samples is 1.31° and the vertical average error is 1.00°.