Multi-Attention Fusion Drowsy Driving Detection Model

Drowsy driving represents a major contributor to traffic accidents, and the implementation of driver drowsy driving detection systems has been proven to significantly reduce the occurrence of such accidents. Despite the development of numerous drowsy driving detection algorithms, many of them impose specific prerequisites such as the availability of complete facial images, optimal lighting conditions, and the use of RGB images. In our study, we introduce a novel approach called the Multi-Attention Fusion Drowsy Driving Detection Model (MAF). MAF is aimed at significantly enhancing classification performance, especially in scenarios involving partial facial occlusion and low lighting conditions. It accomplishes this by capitalizing on the local feature extraction capabilities provided by multi-attention fusion, thereby enhancing the algorithm's overall robustness. To enhance our dataset, we collected real-world data that includes both occluded and unoccluded faces captured under nighttime and daytime lighting conditions. We conducted a comprehensive series of experiments using both publicly available datasets and our self-built data. The results of these experiments demonstrate that our proposed model achieves an impressive driver drowsiness detection accuracy of 96.8%.