Real-time driver drowsiness detection Using YOLO algorithm-based Hybrid Features
TL;DR
A method to identify driver inattention using real-time video processing to quickly and accurately categorize driver information in response to the growing need for driver monitoring technology while driving.
Abstract
One of the leading causes of mortality worldwide is traffic accidents brought on by sleepy drivers, as the driver gradually loses focus without direct awareness. And although there are traditional techniques for monitoring the driver, most of them suffer from limited accuracy or slow response. We created a method to identify driver inattention using real-time video processing to quickly and accurately categorize driver information in response to the growing need for driver monitoring technology while driving. The detection system relies on head tilt and eye movement indicators as key determinants of drowsiness thru real-time face detection using the YOLO algorithm. The high performance and ease of use of the YOLO algorithm helped identify signs of drowsiness with low response time and high precision. In this paper, we adopted the tenth version of the YOLO algorithm. YOLOv10 detects the driver's face. then facial landmarks and vital areas are extracted by MediaPipe FaceMesh. After that, the head rotation angles (Yaw, Pitch, Roll) and the eye closure ratio are calculated separately using computer vision algorithms to determine the driver's drowsiness. The results showed that the rapid face detection of the YOLO algorithm reduces errors and increases the system's reliability. Additionally, combining head movement indicators with eye indicators enhances the accuracy and speed of drowsiness detection compared to relying on a single indicator. The model was tested on the (DDD-NTHU) database from OpenCV. The system achieved an accuracy of 96% in identifying two states of the driver: alert or drowsy.