Current Issue Cover
眼球光心标定与距离修正的3维注视点估计

张远辉,段承杰,朱俊江,何雨辰(中国计量大学)

摘 要
目的 在基于双目视线相交方法进行3维注视点估计的过程中,眼球光心3维坐标手工测量存在较大误差,且3维注视点估计结果在深度距离方向偏差较大。为此,提出了眼球光心标定与距离修正的方案对3维注视点估计模型进行改进。方法 首先,通过图像处理算法获取左、右眼的PCCR矢量信息,并使用二阶多项式映射函数得到左、右眼的2维平面注视点;其次,通过眼球光心标定方法获取眼球光心的3维坐标,避免手工测量方法引入的误差;然后,结合平面注视点得到左、右眼的视线方向,计算视线交点得到初步的3维注视点;最后,针对结果在深度距离方向抖动较大的问题,使用深度方向数据滤波与Z平面截取修正法对3维注视点结果进行修正处理。结果 选择两个不同大小的空间测试,实验结果表明该方法在30~50cm的工作距离内,角度偏差0.7°,距离偏差17.8mm,在50~130cm的工作距离内,角度偏差1.0°,距离偏差117.4mm。与其它的3维注视点估计方法相比较,在同样的测试空间条件下,该方法在角度偏差和距离偏差方面均显著减小。结论 提出的眼球光心标定方法可以方便准确地获取眼球光心的3维坐标,避免手工测量方法带来的误差,对角度偏差的减小效果显著。提出的深度方向数据滤波与Z平面截取修正法可以有效抑制数据结果的抖动,对距离偏差的减小效果显著。
关键词
3D gaze estimation using eyeball optical center calibration and distance correction

Zhang Yuanhui,Duan Chengjie,Zhu Junjiang,He Yuchen()

Abstract
Objective Gaze estimation can be divided into 2D (two-dimensional) gaze estimation and 3D (three-dimensional) gaze estimation, the 2D gaze estimation based on polynomial mapping only use single eye PCCR (pupil center cornea reflection) vector information to calculate 2D (x, y) point of regard (POG) in a plane, and the 3D gaze estimation based on binocular lines of sight intersection need to use left and right eye PCCR vector information and the 3D coordinate of left and right eyeball optical center (the point at which the eye sight emitting) to calculate 3D (x, y, z) point of regard in a three-dimensional space. In the process of 3D gaze estimation, there exist the measurement error introduced by manual measurement the three-dimensional coordinates of the eyeball optical center and the large deviation of the 3D gaze estimation results in the direction of depth. Based on the traditional binocular lines of sight intersection method for 3D gaze estimation, we propose two primary improvements, on the one hand, we use a calibration method to get the three-dimensional coordinates of the eyeball optical center to replace the manual measurement, on the other hand, we use data filtering in depth direction and Z-plane intercepting correction method to correction the 3D gaze estimation results. Method Firstly, the subject gaze the 9 marked points on a calibration plane which is at the first distance away from human eyes, and in the same time, there is an infrared camera in front of the subject to capture eye images, by the image processing algorithm, it can get the PCCR vector information of left and right eyes. According to the second order polynomial mapping principle between the PCCR vector and the plane marked points, the mapping functions of left and right eyes on the first plane can be solved; Secondly, move the calibration plane to a second distance, the subject gaze the 9 marked points again. Using the mapping functions of left and right eyes, 2D point of regard of the left and right eye at the first calibrated distance can be calculated, connect the 9 mark points at the second distance to the left and right 2D point of regard at the first calibrated distance respectively, multiple lines will intersect at two points, calculating these two equivalent intersection points can get the calibration result of the 3D coordinates of the eyeball optical center; Thirdly, 3D gaze estimation can be performed, combining left and right planar 2D point of regard with the 3D coordinates of the eyeball optical center, and establishing appropriate space coordinate system (taking the calibration plane as the X and Y plane, and taking the depth of the distance as the Z axis), lines of sight of left and right eyes can be calculated. According to the principle of human binocular vision, lines of sight of left and right eyes will intersect at one point in space, calculating the intersection point can get the rough 3D point of regard. Due to the calculation and measurement error, the binocular vision lines are generally disjoint, choose the midpoint of the common perpendicular as the intersection; Finally, for the larger jitter of the result in depth direction, using the proposed data filtering in depth direction and Z-plane intercepting correction method to correct the rough result. In this method, the data sequence of depth distance direction (Z coordinate) is firstly filtered, using the filtered distance result generates a plane which perpendicular to the Z axis, the plane intercepts the left and right lines of sight to get two points, choose the midpoint of two points as the correction results of the other two directions (X and Y), after this filtering and correction process, it can get a more accurate 3D point of regard. Result We use two different size of workspaces to test the proposed method, the experiment result shows that in the small workspace (24×18×20cm3), work distance in depth direction is 30~50 cm, the angular average error is 0.7°, the Euclidean distance average error is 17.8mm, and in the large workspace (60×36×80cm3), work distance in depth direction is 50~130 cm, the angular average error is 1.0°, the Euclidean distance average error is 117.4mm. Compared with other traditional 3D gaze estimation methods, the method we proposed is significantly reduced in Angle deviation and distance deviation under the same distance testing condition. Conclusion The proposed calibration method for the eyeball optical center can get the three-dimensional coordinates of the eyeball optical center conveniently and accurately, it can avoid the eyeball optical center measurement error introduced by manual measurement, and it can reduce angle deviation of 3D point of regard significantly. The proposed data filtering in depth direction and Z-plane intercepting correction method can reduce the jitter of the 3D point of regard result in depth direction, and can reduce distance deviation of 3D point of regard significantly. This method is of great significance for the practical application of 3D gaze.
Keywords
QQ在线


订阅号|日报