A Lightweight Model for Driver Distraction Classification

Abstract

The number of traffic accidents has increased steadily in recent years all over the world. According to the US National Highway Traffic Safety Administration, 45 percent of traffic accidents are caused by a distracted driver. This work deals with the problem of automating the detection and classification of driver distraction and monitoring the driver in case of unsafe driving. A lightweight model is proposed for driver distraction classification that combines Human Pose Estimation followed by a common classifier (like RandomForest). MoveNet model is uded for Human Pose Estimation. It is an ultra-fast and accurate model that detects 17 keypoints on the human body. The keypoints are prepocessed and then feeded to the classifier. Our model is trained using a publicly available dataset, namely the StateFarm Distracted Driver Detection Dataset. Experimental results show that the keypoints of the the human pose can be used as good features for classification and give good performance compared with deep neural networks.