Analysis of Facial and Acoustic Features for Driver Drowsiness Detection

An innovative framework for identifying driver drowsiness is presented based on two approaches - analysis of facial landmarks of eye closure and mouth opening along with analysis of variations in the speech signal. For facial analysis the system uses a continuous face-tracking algorithm, without equipping the driver with any external bodily gadgets. The system independently tracks the eye and mouth regions using facial landmarks. The parameters extracted from eye region are eye aspect ratio and blink rate while computation of mouth opening indicates yawning as the sign of fatigue of the driver. The experimentation also involves a more practical approach for detecting variations in the speech parameters arising due to fatigue. The vocal parameters explored are pitch, formants and timbral features associated with speech quality. The comparison of vocal parameters between normal and drowsy speech signal has revealed significant variations in the vocal features, especially the formants (F3 and F4) and timbral features. These large and concurrent shifts of the vocal parameters are interpreted as indicative of fatigue by the system. The system accurately determined the driver fatigue state and is found to be independent of variations in the age group, gender, and eye capture status (with or without wearing eyeglass). The proposed model is fast, reliable and the alert system helps to restore the driver’s attention avoiding consequences due to fatigue of the driver.