一种由粗及精的视线追踪系统平面视线参数检测方法

Planar Gaze Parameter Detection in Gaze Tracking System with Active Infrared Light Source

针对视线追踪系统视线参数检测需求和现有方法的不足,基于亮暗瞳差分方案提出了一种采用由粗及精策略的平面视线参数检测方法。在暗瞳图像应用AdaBoost方法定位人脸,根据人脸五官分布先验知识初步确定人眼区域并标记暗瞳图像对应的人眼区域。在差分图像采用投影法确定瞳孔潜在区域,通过形态和尺度分析滤波定位瞳孔。对瞳孔进行边缘检测和椭圆拟合,提取瞳孔中心参数。在暗瞳图像瞳孔对应区域附近搜索并检测普尔钦斑并提取普尔钦斑中心参数。根据瞳孔和普尔钦斑中心参数获取平面视线参数。在视线追踪系统中的应用表明该方法能够可靠、精确的检测平面视线参数。

In order to meet the requirements and improve the existing methods for parameter detection in gaze tracking system, an approach based on difference image of bright and dark pupils for planar gaze pa- rameter detection was proposed. First, an iteration algorithm called AdaBoost was used to find human face, and according to the priori knowledge of facial features, the eye region was determined and marked in the dark pupil image preliminarily. Then, in difference image, a projection method was applied to lo- cate the possible area of pupil, and pupil would be detected accurately by analyzing its shape and scale. Furthermore, edge detection and an ellipse fitting were used to extract the pupil center parameters, and in the corresponding pupil region in dark pupil image, Purkinje spot and its center parameters were detec- ted. The planar gaze parameters were then calculated according to the center parameters of pupil and Purkinje spot. Applications in gaze tracking system show that the method can detect the planar sight pa- rameters accurately.