A Webcam-Based Real-Time Fatigue Detection Model Using Facial Landmarks

Abstract

In this project, an automated real-time drowsiness and micro sleep detection system is proposed, and it is developed with the help of the advanced computer-vision techniques. The system uses Mediapipe Face Mesh and OpenCV to identify 468 accurate facial landmarks and also calculates the important physiological measurements like the Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR) and gaze deviation. These measurements can help the algorithm to determine the level of alertness in the user and can group the individual as Awake, Drowsy, Sleeping or as one that is undergoing Micro-Sleep. The unique feature of the work is that it is applied to the Google Colab environment where webcam frames are captured with the help of the browser with the use of JavaScript integration periodically. This removes the special purpose hardware requirements and makes the system light, available and platform independent. Also, the model keeps a statistical record of the frequency of each state during the session, which allows conducting a more in-depth behavioral analysis and performance assessment. The system has shown an effective and viable solution to the early detection of fatigue by integrating geometric analysis of faces with automated state identification. Some of the areas where it can be used are driver monitoring systems, solutions to occupational safety, remote-work monitoring, and human computer interaction where alertness is of paramount importance. Generally, this project shows how a combination of facial features landmarks and basic computation ratios can be used to create efficient fatigue-detection systems that can be operated in real time in cloud based setup.