A fast and accurate algorithm to distinguish between open and closed eye by efficient combining of texture and appearance features

In this paper, a fast and accurate algorithm to distinguish between open and closed eye is proposed. In the proposed approach, we use a fast and accurate preprocessing stage based on Haar features to detect the face area, color and intensity mapping to extract the eye candidate areas, and some simple geometrical constraints for final approval of the eye area. Then, for detecting the eye state with high accuracy, texture features extracted from local binary pattern (LBP) and mean local binary pattern (MLBP) histogram in eye areas are applied to two SVM classifiers. Finally, in the case of conflicting results of classifiers based on LBP and MLBP, the amount of exposed sclera is used for final decision making of eye state. The proposed algorithm uses a logical combination of texture and appearance features to increase the accuracy of distinguishing between closed and open eye, and because of limiting the search space at each step for the next one, has an acceptable computational cost. Experimental results on test images show that the proposed algorithm can correctly detect the eye state by the ratio of %99.1, which is higher than other similar algorithms. In addition, this algorithm has never wrongly detected a closed eye as open one; so, it can be used safely in applications such as driver drowsiness detection.