基于双目立体视觉的视线跟踪系统标定  

Calibration for Gaze Tracking System Based on Binocular Stereo Vision

作　　者：温晴川[1] 达飞鹏[1] 方旭[1] 

机构地区：[1]东南大学自动化学院,江苏南京210096

出　　处：《光学学报》2012年第10期152-162,共11页Acta Optica Sinica

基　　金：国家自然科学基金(51175081;61107001);江苏省自然科学基金(BK2010058)资助课题

摘　　要：针对使用主动红外光的视线跟踪系统存在的限制,构建了在自然光条件下允许头部自然移动的视线跟踪系统,该系统通过双目立体视觉系统获取眼睛特征参数,在三维空间中重构眼睛视线向量来估计眼睛的视线。通过头戴式标定板来估计头部姿态,减少了视线跟踪过程中头部移动带来的误差影响;提出了一种视线跟踪系统中个人参数的标定方法,该方法先假定眼睛视轴与光轴重合,由此计算出眼球的中心、半径、眼睛视轴与光轴的夹角的初始估计值,用非线性优化算法对这些初始值进行优化。通过实验表明,提出的视线跟踪算法对使用者限制较少,同时具有较高的视线估计精度,能满足很多视线跟踪应用场合的需要。To solve the limitation in the gaze tracking system using active infrared light, a gaze tracking system under the condition of natural light allowing the human head moves freely is constructed. Binocular stereo vision method is employed to obtain the characteristic parameters of eyes, and then the sight line can be estimated. The head pose can be estimated by using headset calibration board and the error caused by head moving can be reduced. Then a calibration method of individual parameters in gaze tracking system is proposed. This method assumes the optical axis is coincided with the line of sight. Then the center and radius of eyeball, the angle between eye optical axis and visual axis can be calculated. The optimal parameters can be obtained by employing the nonlinear optimization algorithm to process the initial estimate values. Experiment results verify that the proposed method is almost unconstrained, and the estimation accuracy of sight line is high. So this method can satisfy the need of gaze tracking system in many cases.

关 键 词：机器视觉 视线跟踪 立体视觉 个人参数标定 摄像机标定 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]