Real-time human–computer interface based on eye gaze estimation from low-quality webcam images: integration of convolutional neural networks, calibration, and transfer learning.

This article focuses on developing a real-time, appearance-based eye gaze estimation system for human–computer interaction (HCI) that enables cursor control using low-quality eye images captured from conventional webcams. The system employs convolutional neural networks (CNN), calibration, and transfer learning to create a generic model adaptable to new users without requiring specialized hardware. Data were collected from multiple participants under varied lighting conditions, and the model demonstrated an average visual angle accuracy of 2.08 degrees before calibration, improving to 1.81 degrees after calibration. The approach offers a cost-effective, non-invasive alternative to traditional gaze estimation methods, with potential applications in accessibility and virtual reality, though accuracy decreases near screen corners. Future work may incorporate head movement data and eye blink detection to enhance performance and functionality.