基于低成本眼动记录系统的视线估计研究  

Research on eye gaze estimation based on low-cost eye movement recorder

作　　者：孟春宁[1] 白晋军[2] 张太宁[1] 常胜江[1] 

机构地区：[1]南开大学现代光学研究所,天津300071 [2]天津工业大学电子与信息工程学院,天津300387

出　　处：《光电子．激光》2013年第8期1600-1605,共6页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(61171027);教育部博士点基金(20090031110033)资助项目

摘　　要：高精度眼动记录技术现已非常成熟,但昂贵的价格限制了当前商用眼动仪的应用范围,发展灵活、低成本的眼动记录系统将是眼动跟踪技术的重要研究方向。本文设计了一种低成本、低侵入性的耳麦式眼动记录系统。系统由两个普通CMOS摄像机及耳麦式支架构成,两个摄像机分别记录注视场景图像及人眼图像,成本仅为当前商用眼动仪的几百分之一。针对系统采集的数据,提出一种分段加权环形Hough变换算法提取虹膜轮廓及中心作为特征,采用支持向量回归(SVR)算法估计注视点与特征点位置的映射关系。设计了一种简便的标记注视点坐标的方法,实现注视估计。实验结果表明,对场景图像中随机位置的注视方向估计平均误差为1.57°,表明低成本眼动仪可以满足日常交互需求。Recording the eye movement precisely has developed into a mature technology. However, the application field of the commercial eye trackers is very narrow, which is attributed to its high cost. Future gaze tracking systems should be low-cost and flexible. Therefore,a low cost and simple eye movement recording system is designed. The system consists of two common CMOS cameras and a headset-style holder. One camera is used to capture eye images, and the other one is used to capture the scene images. The cost of the system is only hundredths of the existing commercial eye trackers. A segmented weigh- ring-annular Hough transform is proposed to detect the irises in the eye images captured by our system, and the centers of the irises are used to estimate the gazing mapping by support vector regression （SVR）. Finally,a convenient calibration procedure is designed. The experiment results show that the average error of the random positions in scene images is 1.57°, which demonstrates that the low cost eye tracker is able to be used for everyday human-computer interaction.

关 键 词：眼动记录 视线估计 眼动仪 HOUGH变换 支持向量回归(SVR) 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]