Abstract
The driver drowsiness detection used in human security systems aims to decrease the number of accidents. We describe in this paper an approach developed to detect the driver drowsiness state from a video-based system. Our approach uses a noninvasive method which excludes any human related elements. The latter calculates two geometric features to calculate a non-linearly and non-stationary signal. We analyze the signal extracted from the previous step by combining the two methods EMD (Empirical Mode Decomposition) and BP (Band Power) for filtering. This analysis is confirmed by the SVM (Support Vector Machine) to classify the driver alertness state.