| description abstract | Traffic accidents caused by drivers’ actions are worrying since they result in significant losses of life and property. Due to the advancement of deep learning and computer vision technology, driver monitoring and assistance systems (DMASs) have gained widespread use to improve driving safety. Although numerous modern cars come equipped with advanced driver assistance systems (ADAS), most lack such integrated systems. This article introduces a mobile, image-focused DMAS designed for the immediate identification of driver distractions. The developed system uses the YOLOv8 algorithm due to its high accuracy and speed, which has undergone some improvements to increase the accuracy of real-time detection and renamed here GM-YOLOv8n. The system was trained and evaluated on data sets of seven categories, with approximately 6,008 images, by analyzing the face, eye, and mouth areas. In evaluation metrics, the model achieved an average accuracy of 98.90%, and the F1 score reached 98.00%, while the precision reached 97.90% and the recall reached 98.10% to evaluate the model’s performance during validation. The system was then implemented on high-performance embedded platforms and standard computers to measure its effectiveness in real-time. Test runs on the Nvidia Jetson Xavier Nx revealed that the system could analyze visual information at a velocity of 43.17 frames per second, confirming its potential for instantaneous detection. This system aims to reduce the number of accidents and protect human life during transportation by detecting the driver’s various states and issuing an audio alert. The detection results show the success of the proposed system, which classifies this work with high accuracy. | |
