Driver Drowsiness Detection Using Deep Learning Models Based On Different Camera Positions

Driver drowsiness is among the most critical causes of vehicular incidents. Consequently, a solution to mitigate this problem is exponentially increasing where an intelligent vehicle system is being implemented in vehicle development. In these studies, a solution is developed by implementing Convolutional Neural Network (CNN), which is tasked to identify the driver’s drowsiness behaviour. For the datasets that are needed for training and testing of CNN models, datasets are collected at VITAL Laboratories located at UiTM Shah Alam that is recorded from 3 DVR Camera that is placed in the left (CL), middle (CM), and right (CR) positions. The models proposed in these studies are MobileNet-V2, ResNet50, and InceptionV3 architecture, where the performance of each model is evaluated by classifying the behaviour based on three different camera positions. The architecture that is being chosen is InceptionV3 from camera 5, which is placed in the middle positions that successfully achieved an overall accuracy of 85.5% in classifying drowsiness behaviour.