Driver Distraction Detection Using Deep Neural Network

Abstract

Driver distraction, drunk driving and speed are three main causes of fatal car crashes. Interacting with intricated infotainment system of modern cars increases the driver’s cognitive load notably and consequently, it increases the chance of car accident. Analyzing driver behavior using machine learning methods is one of the suggested solutions to detect driver distraction and cognitive load. A variety of machine learning methods and data types have been used to detect driver status or observe the environment to detect dangerous situations. In many applications with a huge dataset, deep learning methods outperform other machine learning approaches since they can extract very intricated patterns from enormous datasets and learn them accurately. We conducted an experiment, using a car simulator, in eight contexts of driving including four distracted and four non-distracted contexts. We used a deep neural network to detect the context of driving using driving data which have collected by the simulator automatically. We analyzed the effect of the depth and width of the network on the train and test accuracy and found the most optimum depth and width of the network. We can detect driver status with 93% accuracy.