Facial Features Enhanced Multi-branch Graph Network for Driver Drowsiness Detection

Driver drowsiness detection from driver facial images presents a notable challenge. During driver drowsiness detection, drivers frequently wear masks or sunglasses, which obscure critical facial features like the eyes and mouth. This occlusion significantly reduces detection accuracy. To address this issue, we propose a multi-branch graph network (MBGN) approach designed to enhance facial feature representation. Firstly, the model locates the facial region in the image and extracts facial landmarks. These landmarks are subsequently transformed into graph node structures. Then, three branches of the graph network structure are used to extract the features of the nodes in three regions. Finally, an adaptive learning weight network (ALW-Net) is introduced to learn the weight parameters for each branch’s features, reducing the impact of occluded features and improving detection accuracy. We conducted experiments on three public datasets and a synthetic dataset. The results indicate that the proposed method improves accuracy by at least 25.9%, 28.2%, 25%, and 22.1%, respectively, with an average detection time of only 0.05 s. This approach significantly enhances facial feature representation, enabling more accurate real-time drowsiness detection even under occlusion.