Automatic Identification of Driver Inattentiveness Using Convolutional Neural Networks

The safe driving and the safety of drivers as well as passengers and properties mostly depend on driver behaviour, such as the attention status of the driver. Driver inattentiveness such as drowsiness is the most extensive aspect of road accidents nowadays. For solving this, our research works have been focused on deep learning-based model testing and designing for classification purposes of driver drowsiness. Some multi-variant dataset was obtained together and self-labelled and divided into two classes (fatigue state and active state) with 9120 RGB frames, 50% was fatigue-related, and 50% was the active posture of the human. The training, validation checking, and testing ratio of data were 65%, 17%, 18% respectively. Here we have tested the LeNet-5 CNN architecture, and then designed a another convolutional neural network, and finally we have classified those images into fatigue and active state. Our proposed model got the highest accuracy of 93.57% and F1_score of 94% over the LeNet-5 architecture.