Abstract
Line-of-sight (LoS) detection, which is a technology for detecting what the user is looking at, is an attracting as a new type of user interfaces. The eyeball also moves very rapidly, which is called as saccade, as well as the conventional LoS tracking, and the saccade is also expected to be used as an another type of user interfaces. Although real-time tracking of saccade is difficult by conventional LoS detection systems for their long processing time or latency. The authors have been developing LoS detection system using Vision Chip, which is an integration of the image sensor and image processing circuitry on single chip, to achieve fast and low latency processing. LoS is calculated by the center position of the pupil from an eye image. However, the pupil may be covered in some cases, such as by a chipped circle by the eyelid, eyelashes, shadows, or anything around the eyeball that results in an error in the LoS calculation. In this paper, the image processing strategy for improving the LoS calculation accuracy that can be implemented on the Vision Chip's parallel architecture, is discussed.
