Gaze Tracking in 3D Space with a Convolution Neural Network “See What I See”

Adiba, Amalia Istiqlali; Asatani, Satoshi; Tagawa, Seiichi; Niioka, Hirohiko; Miyake, Jun

doi:10.1109/aipr.2017.8457962

Cited by 1 publication

(1 citation statement)

References 25 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Adiba 等 [8] 利用神经网络模型对人眼图像中的视觉数据进行训练, 以获得头部姿态和眼睛注视方向信息. Lai 等 [9] 结合瞳孔光斑特征和轮廓特征提取眼动信息, 并使用 2 部红外相机捕捉眼部图像, 通过计算 2 个通过眼睛光轴和摄像机平面的交线得到 3D 注视方向.…”

Section: 相关工作unclassified

Long Distance 3D Gaze Estimation Combined with Radar Target Detection

Zhang¹,

Duan²,

Zhu³

et al. 2020

J Comput-Aided Design Comput Graphics

View full text Add to dashboard Cite

The 3D gaze estimation method based on 2D mapping and two lines of sight intersection has high accuracy in angle estimation. However, there are still large errors in the estimation of gaze distance. Thus, a 3D gaze estimation method based on an auxiliary millimeter wave radar target detection is proposed. First, it uses a movable calibration object to form a virtual calibration plane to calibrate the 2D planar fixation point.Next, the 2D fixation points of the right and left eyes on the calibration plane are calculated by the two-dimensional mapping function, and the 3D gaze direction is calculated combining the position of the eyes. Finally, the candidate point which has the smallest error from the gaze angle is selected from the target points detected by radar. A virtual plane is constructed at the distance of this point, by which the left and right lines of sight are intercept, the midpoint of the two intercept points is the estimation result of 3D fixation point. An experimental device is designed to fuse millimeter wave radar, camera and mobile calibration target. The 3D gaze point is tested in the range of 4.0~10.2 m. Experiments shows that with this method, the average angle error is 0.9°, the average Euclidean distance error is 130 mm. The result is better than binocular data geometric mapping method and depth camera aided estimation method.

show abstract

Section: 相关工作unclassified