Machine Learning Integrated Driver Drowsiness Detection System Using Facial Features

Driver drowsiness causes road accidents as the drowsiness or sleepiness affects the ability of the driver to focus. Drowsiness may happen due to variety of reasons such as lack of sleep, increased driving hours or even because of medical conditions. For the timely detection of sleepiness the recognition of early signs of drowsiness, such as frequent yawning, eye blinking variations, closure of eyes is essential. The proposed non-wearable machine learning integrated driver drowsiness detection system evaluates the early signs of drowsiness. The real time video of the driver is captured by the camera. The Histogram of Oriented Gradient(HOG) based face detection followed by extracting the 68 specific landmark points that defines the key facial features. The points in the Region of Interest(ROI) for drowsiness detection including the eyes and the mouth is extracted for the calculation of Eye Aspect Ratio(E.A.R) and Mouth Aspect Ratio (M.A.R) respectively. EAR is used to for eye open/close detection and MAR used to detect yawning. The proposed method is implemented on Raspberry Pi 4 model B using Python and is tested on various standard datasets. By conducting tests on various machine learning models the effectiveness of our proposed approach is determined. For the Ensemble Learning model which is based on bagged trees, consisting of 30 decision trees trained using the Bagging, our proposed approach obtained an AUC of 0.9439 which shows the reliability and overall effectiveness of the proposed model.