视线追踪系统中基于黎曼几何的落点补偿方法研究

针对非接触式视线跟踪系统中注视点估计算法鲁棒性差的问题,对单相机双光源的视线追踪系统进行了改进.在空间相似三角形注视点估计算法的基础上,提出一种基于黎曼几何的视线落点补偿方法.在眼球模型结构的基础上分析了眼球角膜曲面对视线落点偏差的影响.根据人类眼球生理特点,在黎曼几何中对眼球曲面结构进行建模,提出以最短测地线长度为边长构造平面三角形,以补偿欧式几何中直线距离造成的视线落点的误差.利用非线性多项式模型对眼球视轴和光轴之间的偏差进行拟合,得到最终的视线落点.实验证明,该方法在水平和垂直方向上最大误差均小于1 cm,对视线落点补偿具有显著效果.

Gaze point compensation method based on Riemannian geometry in eye tracking system

Aiming at the problem of poor robustness of gaze point estimation algorithm of un-contact eye tracking system,we modified the eye tracking system consist of single camera and two light sources. One novel gaze point compensation algorithm in Riemannian space was proposed in this paper based on the estimation algorithm of gaze point in space similar triangles. First we made the analysis of the influence of ocular cornea surface to the gaze point error based on eyeball structure.Then, according to the physiological characteristic of human eyes, the ocular cornea surface was modeled in the system of Riemannian geometry. The triangle constructed with the three shortest geodesics was presented to compensate the error of gaze point in Euclidean geometry eye tracking system. Finally the nonlinear polynomial model was also used to fit the error of eye visual axis and optical axis. In our experiment, the deviation of this algorithm is less than 1 centimeter both in hori-zontal and vertical directions. The result showed that the algorithm we proposed had significant effect on the compensationof gaze point.