Scholarlink

<p>Eye-tracking devices are convenient for interpreting human behaviors and intentions, enabling contactless human–computer interaction, such as in medical image interpretation using eye-gaze tracking Recent advances in wearable eye-tracking devices have further allowed wearers to move freely and use them in regular activities. However, gaze estimation from wearable devices tends to be less precise than that from standard stationary eye-tracking devices. This is due to device design constraints and a lack of interpretation of the relationship between the scene and the wearer. In this work, we propose to enhance the accuracy of gaze estimation in wearable eye-tracking devices through a framework that incorporates two neural networks, <i>CorNN</i> and <i>CalNN</i>. The <i>CorNN</i> corrects the bias induced by the distance between the observer and the gaze locations, primarily resulting from the parallax and lens distortion effects. Meanwhile, the <i>CalNN</i> focuses on improving calibration specific to each wearer. To collect precise training data for these networks, we have implemented an automated robotic data collection pipeline. The proposed framework was demonstrated on the Pupil Labs Invisible eye-tracking device and tested on 11 wearers, showing improved average gaze estimation accuracy for all wearers.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing gaze estimation accuracy in wearable eye-tracking devices using neural networks

  • 期刊论文

  • 发表日期：

    2025年6月4日

查看全文
  • Jerome Charton,
  • Tianpeng Zhang,
  • Na Li,
  • Quanzheng Li

摘要

Eye-tracking devices are convenient for interpreting human behaviors and intentions, enabling contactless human–computer interaction, such as in medical image interpretation using eye-gaze tracking Recent advances in wearable eye-tracking devices have further allowed wearers to move freely and use them in regular activities. However, gaze estimation from wearable devices tends to be less precise than that from standard stationary eye-tracking devices. This is due to device design constraints and a lack of interpretation of the relationship between the scene and the wearer. In this work, we propose to enhance the accuracy of gaze estimation in wearable eye-tracking devices through a framework that incorporates two neural networks, CorNN and CalNN. The CorNN corrects the bias induced by the distance between the observer and the gaze locations, primarily resulting from the parallax and lens distortion effects. Meanwhile, the CalNN focuses on improving calibration specific to each wearer. To collect precise training data for these networks, we have implemented an automated robotic data collection pipeline. The proposed framework was demonstrated on the Pupil Labs Invisible eye-tracking device and tested on 11 wearers, showing improved average gaze estimation accuracy for all wearers.