Vision-based drowsiness detector for a realistic driving simulator

This paper presents a nonintrusive approach for monitoring driver drowsiness, based on computer vision techniques, installed in a realistic driving simulator. An IR stereo camera is placed in from of the driver in order to obtain PERCLOS, the most confident drowsiness parameter [1], in real-time and in a robust and automatic way. Our proposal doesn't need a calibration process and includes three main stages. The first is the pre-processing stage, which includes face and eye detection based on appearance strategy using the Viola and Jones algorithm, and the equalization of the eyes using a Hat transformation. An eye tracking strategy in a sequence of image frames is then carried out. The second stage executes the pupil position extraction and its characterization using integral projection techniques and a Gaussian model. The final stage executes the PERCLOS estimation, depending on the eyes closed rate on duration of time interval and fusing information obtained for each eye in the two images of the stereo camera. For evaluation of the proposed system several experiments have been designed by psychologists and carried out. A preliminary study about the performance of the proposal, based on confusion matrixes, is presented.