Eye localization method based on contour detection and D–S evidence theory

Abstract

Accurate eye localization is an important technological basis for driver fatigue detection, and an eye localization method based on contour detection and D–S evidence theory is put forward in this paper. Histogram equalization, face detection, median filtering, binarization and filling background are all integrated in preprocessing to obtain the effective face image. Through the horizontal integral projection, candidate vertical areas of eye are identified, and certain candidate contours can be detected. Then, the first round of screening of the candidate contour set is realized by some statistical parameters obtained by a large amount of eye area contour analysis. Further, the left and right eye areas are established to judge whether the contours are for human eye, as the second round. Finally, three feature parameters of horizontal center line position, vertical centerline position and their acreage are obtained to describe contour information, and the probability for each contour is calculated by the D–S evidence theory to obtain accurate eye contour. Experimental results show that the study can accurately locate eye under the assumed conditions, and its accuracy can be up to 98.7%. In a word, it is a simple and effective method, though its effectiveness may be reduced in some complex conditions.