Takeshi Enya

Facial expression measurement for detecting driver drowsiness

This paper presents the method of detecting drivers drowsiness level from facial expressions. Our method is executed according to the following flow: taking a drivers facial image, tracing the facial features by image processing, and classifying the drivers drowsiness level by pattern classification. We found that facial expression had the highest linear correlation with brain waves as the general index of drowsiness during monotonous driving. After analyzing the facial muscle activities, we determined 17 feature points on face for detecting driver drowsiness. A camera set on a dashboard recorded the drivers facial image. We applied Active Appearance Model (AAM) for measuring the 3-dimensional coordinates of the feature points on the facial image. In order to classify drowsiness into 6 levels, we applied k-Nearest-Neighbor method. As a result, the average Root Mean Square Errors (RMSE) among 13 participants was less than 1.0 level. Our method also detected the drivers smile.