Realization of high-reliability non-contact eye input device for remote nursing support system

Eye-operated input devices have a common problem, namely, false operations caused by the incorrect detection of the eye position or gaze direction. In this study, we developed an eye-position prediction algorithm with polynomial regression to eliminate incorrect and overlooked eye detection, which induce false operations in the eye input device. We also investigated the suitable number of data points (history of eye positions) and the polynomial order for eye-position prediction by the Akaike information criterion (AIC). The experimental results showed that the first order and the history of five eye positions were appropriate for artificial eye movements approximated sine or rectangular functions.