Gaze-based text entry with common RGB cameras

Gaze-based text entry has seen success in both academic researches and commercial products. However, such solutions typically involve specialized hardware, limiting their scalability and feasibility. Commercial gaze tracking solutions that use common RGB webcams yield coarse gaze regions, which are not directly ready for text entry tasks. In this project, to fill the open gap of the cost-efficient gaze-based text entry task with common RGB cameras, we propose a solution that leverages the advantage of both appearance-based and model-based methodologies of gaze tracking, and tackle challenges for the text entry task. With a prototype as the proof-of-concept, we show that our system can enable a user to conduct tap typing on a virtual keyboard with a gaze point on the screen, at a typing speed ranging from 18 to 30 letters per minute with an accuracy of over 90%.