Wrong-Way Driving Detection with YOLOv9 and Convolutional Neural Networks

Nowadays, wrong-way driving has emerged as a significant problem for road safety, thereby demanding robust detection systems. Specifically, this paper provides a solution to wrong-way driving detection by utilizing a deep learning model combining You Only Look Once (YOLOv9) to identify the vehicle and propose a Convolutional Neural Network (CNN) to categorize the behavior. The developed system uses the ability of YOLOv9 to detect objects on the road and then classifies the vehicles using a CNN classifier to determine the direction of travel. The training and validation dataset contains images of moving vehicles in the correct and incorrect directions, as marked by the annotators. The proposed model uses vehicle orientation concerning the road direction and motion trajectory for wrong-way driving detection. The experimental outcomes demonstrate the proposed method’s efficiency, achieving a precision of 0.91, recall of 0.89, F1 score of 0.90, and an accuracy of 0.97. Error analysis, measured by Percent Bias (PBIAS), indicates a minimal overprediction bias of 3.09%, proving that the system is robust and capable of accurately detecting wrong-way driving. This work is relevant to the development of Intelligent Transportation System (ITS) by proposing a solution for wrong-way driving accidents that is practical and effective on a large scale.