Progress in Driver Fatigue Detection Technology Based on Image and State Fusion

Abstract

Driver fatigue state detection represents one of the key technologies for enhancing road traffic safety. This paper systematically reviews the research progress, technical challenges, and development trends of driver fatigue detection methods based on the fusion of visual images and vehicle state information. First, it provides a detailed analysis of the theoretical foundations, implementation mechanisms, and performance characteristics of three categories of methods: those based on visual physiological features, driving behavior features, and multi-modal information fusion, tracing the technological evolution from traditional methods to deep learning models. Second, it delves into three core challenges currently faced by these systems in practical applications: environmental robustness, real-time constraints, and individual generalization ability. On this basis, the paper critically reviews the effectiveness and limitations of cutting-edge technical solutions such as lightweight network architectures, selfsupervised learning, and personalized federated learning. Finally, it outlines future research directions including multi-modal large models, vehicle-road cooperative perception, and neuromorphic computing, while also discussing opportunities and challenges related to technical standardization and system integration. This review aims to provide researchers and engineering practitioners with a systematic technical reference and insights for further development.