AI-Based Driver Fatigue Detection and Alert System: A Comprehensive Review

Abstract: Road accidents caused by driver drowsiness and fatigue represent one of the most critical and preventable safety challenges in modern transportation. The World Health Organization estimates that drowsy driving contributes to over 20% of fatal road crashes globally, imposing enormous human and economic costs. Fatigue impairs cognitive functions such as reaction time, hazard perception, and decision-making in ways that are physiologically comparable to alcohol intoxication, yet far harder for the driver to self-detect. This paper presents a comprehensive review of AI-Based Driver Fatigue Detection and Alert Systems — systems that use computer vision, machine learning, and deep learning to monitor driver facial behaviour in real time and issue timely warnings before cognitive impairment leads to catastrophic outcomes. The proposed system architecture employs a standard webcam to capture live video, applies facial landmark detection using Dlib's 68-point predictor, and computes the Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) to quantify eyelid closure and yawning frequency. When EAR remains below a predefined threshold for a sustained duration, or when yawn frequency exceeds a critical count within a rolling time window, the system triggers an auditory alert via a buzzer and displays an on-screen warning message. The implementation stack — Python, OpenCV, Dlib, imutils, and pygame — is entirely open-source, cost-effective, and capable of running at real-time frame rates on commodity CPU hardware without requiring specialised GPU acceleration or proprietary embedded systems. This survey further proposes a structured four-tier taxonomy classifying existing fatigue detection architectures by functional sophistication, conducts a curated review of fifteen representative peer-reviewed studies from 2016 to 2024, presents cross-paper comparative analysis across six performance dimensions, and identifies seven persistent research gaps that limit real-world deployment. Future enhancement pathways including IoT module integration, GPS-based geo-fencing, cloud fleet monitoring, EEG sensor fusion, and transformer-based attention architectures are discussed to guide the evolution of the field toward deployment-grade smart transportation systems.

Keywords: Driver Fatigue Detection; Eye Aspect Ratio (EAR); Mouth Aspect Ratio (MAR); Facial Landmark Detection; Computer Vision; OpenCV; Dlib; Drowsiness Monitoring; Real-Time Alert System; Convolutional Neural Network; LSTM; Smart Vehicles; Road Safety; Microsleep Detection; IoT Integration.