Robust Eye Center Localization Based on an Improved SVR Method

Abstract

Eye center localization is an important technique in gaze estimation, human computer interaction, virtual reality, etc., which attracts a lot of attention. Although a great deal of progress has been achieved over the past few years, the accuracy declines dramatically due to the low input image resolution, poor lighting conditions, side face, and eyes status such as closed or covered. To handle this issue, this paper proposes an improved support vector regression (SVR) method to detect the eye center based on the facial feature localization. Several image processing techniques were tried to improve the accuracy, and results showed that the SVR combining a Gaussian filter could get a better accuracy.