Distracted Driver Detection Using Deep Learning

Abstract

The increasing number of vehicle road accident is worrying with more than 1.35 million people died on the highways globally in 2019. Prevention must be taken to reduce road accidents by controlling one of the factors contributed to the increasing number of the cases, i.e., distracted drivers. Previous researches have used Deep Learning classification-based technology in detecting a distracted driver in a vehicle. However, there is potential for improvement in the investigation and development of detecting an action of a distracted driver focusing on looking elsewhere. This study aims to detect a distracted driver who is looking elsewhere using Deep Learning-based classification. The method proposed uses Jupyter Notebook and Python to program and run ResNet 50 network. The State Farm dataset, which consists of 10 types of driving behavior is also used in this study. The model has been evaluated based on confusion metrics, accuracy, precision, recall, and F1 score criterion. As a result, the model achieved 94% of accuracy in the classification of distracted driver looking elsewhere. The images of distracted driver were identified, and a notification will appear at the video that contains a distracted driver.