Real-Time Detection of Distracted Driving Behaviors Using Naturalistic Driving Study Data

Distracted driving seriously affects traffic safety and is one of the main causes of traffic accidents. In this study, a convolutional neural network (CNN) real-time detection method is proposed to detect distracted driving behaviors based on naturalistic driving experimental data. The distracted driving behaviors were categorized into using cell phone, makeup, diet and water (i.e., eating and drinking); chatting with passengers; and the normal (undistracted) driving behavior as a reference. The results show that (1) the correct detection rate for the categories of using cell phone (84.0%), makeup (77.0%), diet and water (82.0%) was about 80%; and (2) the detection accuracies of normal driving behavior (36.0%) and chatting with passengers (55.0%) were relatively low due to the similarity of these two types of behaviors being high. The findings of this paper suggest that the AlexNet model could be specifically appropriate and effective for real-time distraction detection.