一种基于标准化人眼图像的视线估计算法  

A Gaze Estimation Method Based on Normalized Eye Image

作　　者：任冬淳[1] 张鹏 王敏[1] 古楠楠[3] 

机构地区：[1]中国科学院自动化研究所复杂系统管理与控制国家重点实验室,北京100190 [2]民政部国家减灾中心,北京100124 [3]首都经济贸易大学统计学院,北京100070

出　　处：《电子学报》2015年第3期536-543,共8页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61033011;No.61210009;No.61101221;No.61303174;No.11426159);国家科技支撑计划项目(No.2012BAI34B02)

摘　　要：本文提出了一种基于标准化人眼图像的视线估计算法,能有效解决头部运动对视线估计的影响.本算法在一个红外摄像机和三个红外光源的条件下,利用三个光斑(光源在角膜表面的反射点)的图像信息,将人眼图像变换到标准化人眼图像.通过引入标准化人眼图像,使屏幕平面到标准化人眼图像平面的射影变换为定值,从而使标准化人眼图像的瞳孔中心和屏幕上的视点的变换关系为定值,由此解决了头部运动对视线估计的影响.实验结果验证了该算法在头部运动情况下的有效性.In order to improve the effects of head movements on gaze estimation, this paper proposes a gaze estimation method based on normalized eye image. The proposed method is built on the installation of one infrared camera and three infrared light sources. We utilize the image of three light spots, which are the reflections of lights on cornea surface, to transform original eye image into the normalized one.By iraroducing normalized eye image,the projective transformation from screen plane to normalized eye image plane is constant. Consequently, the transformation from the pupil center in a normalized eye image to the viewpoint on the screen is fixed. Therefore,the effects caused by head movements can be eliminated.Experimental results demonstrated the effec- tiveness of the proposed method under various conditions of head movements.

关 键 词：视线估计 视线追踪 标准化人眼图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]