Disentangling High-Paced Alternating I/O in Gaze-Based Interaction

Gaze-based input to machines utilizes the ability of eye-gaze to serve as a user’s “output”. However, gaze should also support information flow in the opposite direction, namely, “input” to the user’s visual system from a machine’s output. The two functions can be easily separated in some tasks, like eye typing, but more complex scenarios typically require users to perform additional actions to avoid misinterpreting their intent. In this study, we modeled a free-behavior interaction with rapid transitions between visual search, decision-making, and gaze-based input operations through an engaging game called EyeLines. When playing the game, 15 volunteers selected screen objects using a 500 ms dwell time without additional actions for intention confirmation. By applying machine learning algorithms to gaze features and action context information, we achieved a threefold reduction in false positives, improved the quality of in-game decisions, and increased participant satisfaction with system ergonomics. To our knowledge, this is the first study that demonstrates the effectiveness of machine learning applied to gaze features in enhancing gaze-based interaction within visually challenging environments.