| Drowsy Driving is a major, though elusive, cause of vehicle crashes. The National Highway Traffic Safety Administration (NHTSA) conservatively estimates that 100,000 police-reported crashes are a direct result of driver fatigue each year in the United States. A detection system that can predict drowsiness and alert the driver by monitoring eye closure (PERCLOS) could reduce the number of fatigue-related crashes. Video sequence containing the driver's face can be analyzed to evaluate whether the eyes are open or closed. Ambient illumination may vary during the video sequence resulting in changes in perceived skin color. Therefore, a drowsy driver detection system demands robust, yet efficient algorithms to monitor a driver in real-time. In the present thesis, we evaluate different skin detection models to initialize the face of the driver followed by face tracking algorithms. Template-based tracking and Level-set based face tracking algorithms are implemented and optimized for real time operation. Results of the eye closure (PERCLOS) estimate are provided on a video database with multiple subjects. |
