A Practical Implementation of Real-Time Mental Fatigue Detection Using Webcam-Based Eye Analysis

Mental fatigue significantly increases the risk of errors and accidents, underscoring the need for reliable detection methods. This study presents a real-time, non-intrusive fatigue detection system using a standard webcam and open-source tools. Facial landmarks are tracked with MediaPipe Face Mesh, and key temporal eye features—Eye Aspect Ratio (EAR), Blink Rate, Average Blink Duration, PERCLOS, and EAR Variance—are extracted for analysis. A Random Forest Classifier trained on the UTA-RLDD dataset classifies user states as Alert or Fatigued, achieving approximately 80% classification accuracy on unseen data. The results demonstrate the feasibility of accessible, low-cost fatigue monitoring and highlight the system’s value as a deployable prototype that integrates established computer vision and machine learning techniques.