This paper presents a drowsiness detection method for drivers based on visual features in a video sequence. Image intensities are traditionally visual features. However, it is known that they are directly influenced by lighting conditions. We propose a human eye detection method using the normalized cross-correlation between displacement vectors and gradient vectors. Gradient vectors are dependent on lighting conditions but the normalization step makes them independent of illuminations. In this way, the proposed method can detect human eyes regardless of various lighting conditions. We have also found that normalized cross-correlation can be useful, not only for detecting eyes, but also for recognizing open and closed eye states. To overcome poor lighting conditions, we used infrared auxiliary illumination in order to make the proposed method work every moment. The computation speed of the proposed method is fast enough to perform at video rates.
