A non-intrusive drowsiness detection model for driver safety using facial features and machine learning.

Driver drowsiness and fatigue are leading causes of road accidents, particularly due to drivers having to operate vehicles for long hours under challenging physical and adverse weather conditions. Despite extensive research on subjective, physiological, and vehicle-based detection methods, recent reports still highlight drowsiness and fatigue as significant contributors to accidents and injuries on highways. However, existing drowsiness detection models are often intrusive or expensive. To address these limitations, we propose a low-cost, non-intrusive drowsiness detection model based on facial features, leveraging principles of image processing and machine learning. Our research involves a multi-step approach. Initially, image frames are extracted and subjected to image processing techniques. The images are subsequently passed through a Dlib classifier to detect and localize 68 facial landmarks, including the mouth and eyes. From the localized eye landmarks, salient features such as eye aspect ratio (EAR), pupil circularity (PUC), and eyelid closure over the eye pupil are extracted. To classify the driver's state, these collective features are inputted into various machine learning classifiers. We conducted experiments using the National Tsing Hua University Computer Vision Lab dataset, implementing the proposed approach in the Python OpenCV environment. Notably, the KNN algorithm demonstrated the highest accuracy when the test data was evaluated, yielding a paramount accuracy of 78% on the NTHU DDD dataset. In conclusion, the non-intrusive facial features-based driver drowsiness detection model offers a cost-effective solution to address the pervasive problem of drowsiness-related accidents. The promising results obtained through the utilization of machine learning techniques underscore the potential for practical implementation. [ABSTRACT FROM AUTHOR]