用事件相机+深度学习追踪眼球运动，低成本提升虚拟现实体验
A deep learning approach to track eye movements based on events
- 用事件相机输入，结合CNN-LSTM模型预测眼中心位置
- 在高速眼球运动下达到约81%准确率，支持实时应用
- 方法可解释性强，适合VR/AR设备优化与注意力研究
本研究针对快速眼动（最高达300°/s）时的精准眼球追踪难题，提出基于事件相机的深度学习方法。传统高精度追踪依赖昂贵高速摄像设备，而本文通过卷积神经网络与长短期记忆网络结合（CNN_LSTM）模型，利用事件相机数据实现眼中心坐标（x, y）定位，取得约81%的准确率。该方法具备成本低、可解释性高的优势，适用于虚拟现实（VR）与增强现实（AR）设备中的人类注意力预测，从而提升用户舒适度与交互体验。未来工作将引入层相关性传播（LRP）进一步增强模型可解释性与预测性能。
原文摘要 · Abstract (English)
This research project addresses the challenge of accurately tracking eye movements during specific events by leveraging previous research. Given the rapid movements of human eyes, which can reach speeds of 300°/s, precise eye tracking typically requires expensive and high-speed cameras. Our primary objective is to locate the eye center position (x, y) using inputs from an event camera. Eye movement analysis has extensive applications in consumer electronics, especially in VR and AR product development. Therefore, our ultimate goal is to develop an interpretable and cost-effective algorithm using deep learning methods to predict human attention, thereby improving device comfort and enhancing overall user experience. To achieve this goal, we explored various approaches, with the CNN\_LSTM model proving most effective, achieving approximately 81\% accuracy. Additionally, we propose future work focusing on Layer-wise Relevance Propagation (LRP) to further enhance the model's interpretability and predictive performance.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。