Analýza očí řidiče ve vozidle s využitím neuronových sítí

Abstract

This bachelor thesis addresses the issue of driver fatigue and microsleep detection through eye image analysis. The theoretical part describes key physiological indicators of fatigue, conducts a review of available datasets, and introduces computer vision methods ranging from traditional geometric approaches to modern deep learning architectures. The primary objective of the work is to design and implement an experimental program that utilizes convolutional neural networks for evaluation of the driver’s state based on the PERCLOS metric and blink dynamics. In the practical part, three distinct processing pipelines are developed and implemented (utilizing the MediaPipe framework, the YOLO detector, and a custom convolutional classifier). The conclusion of the thesis presents an objective experimental comparison of these methods in terms of their overall accuracy, robustness to facial occlusion, and computational speed.