Three-Stream Convolutional Neural Network With VotingClassifier for Driver Distraction Detection

Car accident has recently been one of the most critical issues in transportation systems. It is also one of thetop ۱۰ causes of death worldwide. According to the report by the statistics center of Iran, around ۱۷۱۸۳ people diedfrom car accidents in ۲۰۱۷. It was investigated that drivers' distractions are the leading cause of most road accidents.So, it is necessary to decrease the number of accidents and people injured in accidents. One way to do that is to analyzethe action and behavior of drivers while driving. With the breakthrough of Deep Learning in recent years, differentdeep models have been proposed to distinguish drivers' actions and behaviors in different states like eating, drinking,and talking on the phone, just to mention a few. In this paper, we propose a three-stream Convolutional NeuralNetwork (CNN)-based model to recognize the distracted driver and classify the type of distraction. A shallow CNNalong with two fine-tuned pre-trained CNN models, VGG-۱۹, and ResNet۵۰, are used in the proposed model. Finally,a voting classifier is used for the final classification. Results on the State Farm Distracted Drivers dataset confirm thesuperiority of the proposed model compared to state-of-the-art models in driver distraction detection

کلیدواژه ها:

Driver-Distraction ، Deep learning ، Machine Learning ، convolutional neural network