MACHINE PERCEPTION AND ROBOTICS GROUP

機械知覚&ロボティクスグループ
中部大学

First Person Vision, which attempts to understand a person’s behavioral intention, requires information on the person’s state and on what the person is looking at. We propose an inside-out camera that simultaneously obtains an image of one of the person’s eyeballs and an image of that person’s visual field, and propose a method for estimating the person’s gaze point based on the configuration of the camera. The relationship between the gaze vector obtained from the eyeball video and the gaze point obtained from the scene video is expressed by a conversion equation. In an evaluation experiment, we took scene video for both eyes when gazing at a certain landmark as evaluation data, and calculated the error between the position of the landmark in the scene video as the true value and the gaze point estimated from the gaze vector.