A low-computational approach on gaze estimation with eye touch system.

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Computer Vision

Background:

• Light-reflection based eye tracking systems offer low complexity but moderate accuracy.
• Existing systems often have limitations in distinguishing arbitrary gaze directions.
• Previous iterations of the Eye Touch system could only identify limited predefined gaze regions.

Purpose of the Study:

• To propose a low-computational approach for gaze estimation using the Eye Touch system.
• To enhance the Eye Touch system's capability to estimate arbitrary gaze directions.
• To develop an accurate wink-based pattern classification for user interaction (clicks).

Main Methods:

• Utilized sensor measurements from the Eye Touch system with low-computational least-squares algorithms for gaze estimation.
• Implemented a pattern classification algorithm for distinguishing left, right, and double eye winks.
• Developed a custom microcontroller-based data acquisition system for robust and sensitive hardware biasing.

Main Results:

• The proposed system accurately classifies eye winks with 98% accuracy.
• Gaze direction estimation achieved an average angular error of approximately 0.93 degrees.
• The system demonstrated reduced physical size, lower cost, and improved power efficiency.

Conclusions:

• The enhanced Eye Touch system provides accurate gaze estimation and reliable wink-based control.
• Its low-computational requirements and lightweight structure make it suitable for stationary and mobile applications.
• The system presents a competitive alternative to video-based eye tracking for human-computer interfaces.