Gaze Estimation Based on Eyeball-Head Dynamics
Ikuhisa Mitsugami, Yamato Okinaka, Yasushi Yagi · 2017
Human's gaze direction is a useful cue to understand his/her attention and interest. There are many kinds of eye tracking devices, but they are usually unavailable for people observed in surveillance views because they are located too far to observe their eyeballs. If we could know the gaze direction from such surveillance views, it should be very effective for many applications. Our research objective is thus to estimate the gaze direction without any eye-trackers but by observing their behaviors. This paper proposes a new gaze estimation method based on a dynamical model emulating dynamic relation between eyeball and head. Experimental results using head-mounted and body-mounted camera images confirmed its effectiveness.