Published Paper Details:

The paper proposes a machine learning-based and computer vision-based real-time driver drowsiness detection system to reduce 30% of road accidents caused by driver fatigue. The system uses OpenCV for video capture and Dlib for detection of facial landmarks, i.e., 68 facial points for calculation of Eye Aspect Ratio using Euclidean distance. Temporal behavior and statistic attributes obtained through EAR values are utilized by machine learning classifiers for discriminating among alert and sleepy conditions. For detection of sleepiness, real-time audio alarm output is presented. Experiment testing was conducted on various subjects subjected to varying conditions of light, as well as road conditions, and providing a measure of drowsiness detectability with a measure of accuracy at 94% and a number of zero false alarms. The system's response time averaged 0.3 seconds from detection to alert, giving drivers sufficient reaction time to prevent accidents. Our method incorporates a novel adaptive thresholding mechanism that dynamically adjusts according to the baseline EAR values of individual users, resulting in significant performance under diverse facial structures and eye shapes. The system also includes an additional temporal analysis window that is utilized to analyze EAR patterns across adjacent frames for canceling spurious alarms due to natural blinking or temporary closures. With support for Python 3.8 and MySQL databases, this non-intrusive monitor solution promises high potential to enhance road safety via early detection of drowsiness. The suggested method combines commonly employed computer vision techniques with novel feature extraction methods to offer robust detection performance under varying lighting conditions and face orientations.