A Novel Method for Unsafe Driving States Detection

Abstract

With the increase in vehicle consumption in recent years, traffic accidents have become the most important cause of premature death all over the world. In particular, the unsafe driving state like fatigue and distraction, which lead to around 90 percent of crashes. In this paper, an unsafe driving state detection method based on deep learning is proposed. First, the optimized MTCNN method is used to detect human face and locate facial feature points, the optimized MTCNN method accelerating the speed up to 27 times. Second, a new ERFP method is used to achieve more accurately facial feature points locating. Meanwhile, a simple and efficient method FOD is proposed to detect the facial orientation by analyzing the change of the relative position between the nose tip and facial position. Third, this paper also built the new unsafe driving state dataset UDS for training the classification model of the status of eyes and mouth. Finally, the different standards are designed to detect the different unsafe driving states in experiments, compared with previous methods, the method proposed in this paper got a better result in drivers' unsafe state detection.