End-To-End Driver Distraction Recognition Using Novel Low Lighting Support Dataset

In this paper, seven End-To-End deep learning models, including three sequence models, were trained for driver distractions recognition. One of these models (Convolutional GRU) achieved 95.48% test-accuracy. The performance of each model was analyzed using different techniques like confusion matrix, t-SNE representation, and saliency map. In addition, the largest driver distractions dataset, to date, was presented. This dataset contains ten classes and comes with temporal information. Using the NoIR technology the dataset was captured, that makes the dataset to be able to contain samples at different lighting conditions in case of using IR LEDs. The dataset contains 70 drivers either males or females.