Smart Driver Assistance: Real-Time Drowsiness Detection Leveraging Facial Cues with MediaPipe and OpenCV

Abstract The primary aim of this research is to identify driver drowsiness to prevent car accidents and improve road safety. This study explores and proposes potential solutions to mitigate drowsiness-related accidents and enhance overall road security by monitoring the driver's eye, mouth, and head movements, which are key indicators of drowsiness. An algorithm has been developed to track these movements. The study involves analyzing CNN models and computer vision models for eye detection, yawn detection, and head movement, with the objective of identifying the most effective approach among the models. For building CNN models, we use the MRL Eye Dataset and YawDD datasets for eye and mouth tracking. The performance of the models developed using these datasets will be assessed based on various metrics such as accuracy and loss. We achieved an accuracy of 84.53\% using the CNN approach for the MRL Eye Dataset, while for the YawDD Dataset, we obtained an accuracy of 96.42\%. Despite these impressive results, CNN techniques encounter several challenges, including poor performance in low light conditions, varying outcomes for different ethnicities, and unreliability in dynamic environments. To address these challenges, we utilize Dlib and MediaPipe for tracking facial landmarks. By leveraging these techniques, we can simultaneously monitor head movement, eye movements, and detect yawns. Between Dlib and MediaPipe, we prefer MediaPipe due to its capability to detect the 3D coordinates of key points.