Due to the high cost of eye-tracking systems based on pupillary corneal reflections, efforts to develop a webcam-based eye-tracking system have increased to provide an affordable alternative to disabled people in recent years. However, due to the camera specification and location, ambient light changes and positional changes of the users, the gazing point of the eyes has not yet been determined precisely by such a system. Therefore, only 8 different gaze directions or up to 10 gaze regions could be detected in the previous webcam-based human–computer interaction studies. In this study, a novel gaze input system has been proposed to make the best use of the limited performance of webcam-based eye tracking and offer an economical alternative for disabled people. To reduce the impact of head movements, the webcam has been mounted to an ordinary glasses frame and positioned in front of the eye. For estimation of the gaze regions, a feature-based method (Hough transformation) was used by considering the circular shape of the iris and the contrast between the iris and sclera. The central coordinates of the iris image captured by the webcam were given to the k-nearest neighbor classifier. We performed a series of experiments with 20 subjects to determine the performance of the system and to investigate the effect of ambient light on the system’s accuracy. The 23 regions that were gazed at by subjects were determined with an average accuracy of 99.54%. When the ambient light level was reduced by half, the accuracy decreased to 94.74%. As a result, it has been found that the proposed prototype allows more accurate recognition of a larger number of regions on the screen than previous webcam-based systems. It has been observed that system performance decreases if the ambient light is reduced byhalf.

中文翻译：

---

一种基于安装了网络摄像头的眼镜的虹膜跟踪的新型注视输入系统

由于基于瞳孔角膜反射的眼动追踪系统成本高昂，近年来开发基于网络摄像头的眼动追踪系统的努力已经增加，以为残疾人提供负担得起的替代方案。然而，由于相机的规格和位置、环境光的变化和用户的位置变化，这样的系统还不能准确地确定眼睛的注视点。因此，在之前的基于网络摄像头的人机交互研究中，只能检测到 8 个不同的注视方向或多达 10 个注视区域。在这项研究中，提出了一种新颖的注视输入系统，以充分利用基于网络摄像头的眼动追踪的有限性能，并为残疾人提供一种经济的替代方案。为了减少头部运动的影响，网络摄像头已安装在普通眼镜架上并位于眼前。对于凝视区域的估计，通过考虑虹膜的圆形形状以及虹膜和巩膜之间的对比度，使用基于特征的方法（霍夫变换）。将网络摄像头捕获的虹膜图像的中心坐标提供给 k-最近邻分类器。我们对 20 名受试者进行了一系列实验，以确定系统的性能并研究环境光对系统精度的影响。受试者注视的 23 个区域被确定，平均准确率为 99.54%。当环境光水平降低一半时，准确度下降到 94.74%。因此，已经发现，与以前的基于网络摄像头的系统相比，所提出的原型可以更准确地识别屏幕上的大量区域。已经观察到，如果环境光减少一半，系统性能就会下降。