A Real-time Driver Fatigue Monitoring System Based on Lightweight Convolutional Neural Network

Traffic accidents caused by fatigue driving happen frequently. How to alarm in advance and reduce the accident rate is the focus on research. The traditional fatigue detection method occupies high computing resources, and has insufficient real-time performance. In the context of intelligent traffic safety, this paper collects facial feature data, and proposes a driver monitoring system based on MobileNetV3 and Jetson TX2 platform. The system can monitor the driver status by extracting the driver's facial features. For the night and wearing sunglasses, a infrared camera is selected, and corresponding data is collected for model training. After that, tests are conducted on public datasets YawDD, CEW and self-acquired dataset. The results show that the model achieves a real-time speed of 22 fps on the Jetson TX2 platform, and the average accuracy of the algorithm can reach 94%.