P-FDCN Based Eye State Analysis for Fatigue Detection

Driver fatigue endangers traffic safety increasingly. In this paper, a novel fatigue detection approach based on eye state analysis is proposed. Specifically, we first build a fatigue detection convolutional network (FDCN) based on common convolutional neural network (CNN). Then we incorporate projection cores into FDCN to construct P-FDCN that not only enhance the resistance of learned features to the scale changes but strength the learning of texture information. Experimental results demonstrate the proposed approach achieves the recognition rate of 94.9%, yielding a 1.2% promotion in the average accuracy rate compared with the results obtained using the CNN alone on the Closed Eyes in the Wild (CEW) database. In addition, our approach has 1.0% accuracy improvement on the ZJU database, as well as significantly outperforming the Faster RCNN model and the traditional projection method.