%0 Journal Article
%T 基于视线追踪的增强现实人机交互方法<br> Eye Tracking Based Augmented Reality Human-Computer Interaction Method
%A 陈卫兴
%A 纪欣伯
%A 崔笑宇
%J Computer Science and Application
%P 1020-1028
%@ 2161-881X
%D 2019
%I Hans Publishing
%R 10.12677/CSA.2019.95115
%X <div align=\"justify\">
	<span style=\"font-size:14px;\">本文结合现有光学透视头戴显示设备，提出了基于视线进行交互的方法。受到Resnet网络思想的启发，提出了一种异构嵌套的卷积神经网络(Heterogeneous nested neural, HNN)，通过12层卷积和3层全连接层得到一个眼睛的55个特征点，在此基础上分别得出视线向量和瞳孔中心，进而得到了用户视线。在近眼部位采用两个非红外光源相机分别追踪双眼的运动，角膜中心误差降到了0.66 mm，角度误差降到了0.89？。最后本文在HNN中做了不同通道数的对比以及HNN和Resnet网络的对比。实验结果表明，相比于传统方法和普通的卷积神经网络，本文提出的HNN网络对眼睛追踪效果有明显提升。</span><span style=\"font-size:14px;\"><br />
In order to improve the
human-computer interaction (HCI) of augmented reality, we combined with the
existing optical see-through head-mounted display (OST-HMD). Inspired by the
Resnet network idea, we proposed a Heterogeneous nested neural network, which
outputs 55 feature points of an eye. On this basis, we respectively get the
line of sight vector and the center of the pupil. Two Non-IR cameras are used
to track the movement of both eyes in the near eye area. The corneal center
error was reduced to 0.66 mm and the angular error was reduced to 0.89？. The
method has achieved good results, which lays a foundation for the eye movement
interaction of the augmented reality system.</span> 
</div>
%K 视线追踪，深度学习，增强现实<br> Eye Tracking
%K Deep Learning
%K Augmented Reality
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=30567