A novel gaze detection method with robustness in face rotation using facial feature tracking

Abstract

Gaze detection is to detect someone's gaze direction, and it is possible to detect accurate gaze when considering both face and eye directions at the same time. Although previous methods detected facial features of eye and lip regions each frame by an object detector based on Haar-Like and Adaboost, there was a problem that eye and lip regions were not accurately detected according to face rotation or eye and mouth shape. In order to solve this problem, this paper aimed to detect face and eye directions by extracting feature points from the face and eye regions and tracking the extracted feature points based on the face rotation. This paper also used methods to extract feature points from the eye and lip regions detected from the first frame of an image, to store the specific sized window region based on the feature points, and to track the eye and lip regions by histogram matching in the next frame. Moreover, the number of cases shown according to the face and eye directions in gaze detection was set at 81, and gaze direction was detected based on this. As a result, it was possible to solve feature points' false detection caused by the face rotation and to detect accurate gaze resistant to face rotation. © 2014 International Information Institute.