MULTISTAGE ADAPTIVE 3D-CNN MODEL FOR DETECTING THE DROWSINESS AMONG THE DRIVERS

Abstract

Increased driver drowsiness-related accidents underscore the need for better detection technologies to address this global issue. Researchers have tried numerous machine learning and deep learning methods, including CNN variations, but computational complexity, low evaluation accuracy, and doubtful reliability make real-time deployment difficult. A multistage adaptive 3D-CNN model for Driver Drowsiness Detection that addresses system complexity and performance. Our models classify driver fatigue symptoms. Face attributes for sub-models are collected using different 3D CNN architectures. The head condition model finds head dozing, eyebrow elating, and glancing towards the side. The eye alignmentprototype finds eye closings, soreness, and eyes that are only partially closed. Mouth condition model identifies open and covered yawning. The specs and usual circumstances model classify states as day, sunglasses, night, and normal. All models are essential to the system, detecting distinct driver sleepiness factors with specialized algorithms. Extensively examined using KEC-DDD and NTHU-DDD datasets, representing their toughness and adaptability. Dataset validation findings confirm the sub-models' capacity to accurately capture driver behaviours in various contexts.