Driver Behavior Tracking: A Hierarchical Classification Approach

Abstract

Promoting road safety remains a paramount concern in contemporary transportation systems, necessitating proactive measures to detect potential driver behaviors that may pose safety risks. This paper introduces an innovative methodology for evaluating driver safety, which integrates various AI models, including YOLOv8, logistic regression, a fine- tuned VGG13 classification model used for emotion classification, and CNN for identifying 15 common driver behaviors. The incorporation of hierarchical classifiers ensures robust, efficient, and accurate assessments by leveraging data from multiple sources, including publicly available datasets and custom datasets designed to capture an abundance of behaviors. Ten video streams were passed to the hierarchical classifier. Moreover, the proposed approach significantly reduced the number of processed frames compared to non-hierarchical methods. The accuracy obtained from the hierarchical classifier is 71%.