Non-Contact Eye-Gazing Point Detection Technique via Two Dimensional Multi Resolution Analysis

Abstract

The purpose of this study is to detect eye-gazing points during driving a wheelchair through the image sequence analysis of video frames. The performances can be measured without restrictions of specific devices such as any accesories and sensors, which were put on a body. Images obtained with three video cameras allow the algorithm to extract four feature points on a face and irises. Their three-dimensional positions were estimated through direct linear transformation (DLT) technique with magnification of a facial image through multi resolution analysis. The direction of a face was estimated via three-dimensional positions of the former, and it defined the facial coordinate. On the other hand, the eye-gazing point was evaluated through three-dimensional positions of the latter. Consequently, Some preliminary results were presented to illustrate the difference in variations with the eye-gazing point between driving a wheelchair and walking.