Analysis And Comparison on Image-Based Fatigue State Detection Methods

Abstract

Nowadays, fatigue driving has become one of the main causes of road traffic accidents worldwide. To ensure people's safety, the academic community has continuously proposed and applied new fatigue detection methods, but there are still many limitations in current research. Manual extraction of fatigue features cannot obtain more detailed information; individual differences among different subjects may lead to reduced accuracy of fatigue detection; the detection delay is high, and the real-time performance needs to be improved...In response, this paper presents a comparative review of image-based fatigue detection methods. It systematically analyzes recent advances by comparing their core processes, including datasets, feature extraction techniques (both handcrafted and deep learning-based), and classification models. This research finds that Convolutional Neural Network(CNN) has gradually matured in the feature extracting domain. Many studies have built upon this foundation, introducing improvements and innovations. These studies extract features through machine learning and deep learning, and perform fatigue state classification using various combined models. The innovation of personalized thresholds allow flexible adjustment of judgment threshold according to each person's characteristics to increase accuracy, and the introduction of OpenCV, Dlib libraries and YOLO models has significantly enhanced real-time performance... The systematic summary provided in this study offers support for researchers to further optimize fatigue detection models, and also provides fatigue detection references for car manufacturers and high-risk operation industries. These contributions are conducive to ensuring the safety of relevant personnel.