Wrong-Way Driving Detection for Enhanced Road Safety using Computer Vision and Machine Learning Techniques

This paper describes a real-time vehicle detection and tracking system using deep learning techniques together with computer vision. A system based on the YOLOv4 model is developed using the Lukas-Kanade optical flow technique to locate the vehicles considering real-life traffic situations that involve atmospheric disturbances. The subsystem first assesses the trajectory of vehicles concerning the intended direction of traffic flow. Any deviation from this norm triggers a signal that will enable traffic managers to intervene in time to prevent the possibility of traffic congestion. The initial findings confirmed the system's evident ability to reduce the incidence of wrong-way driving, such that its enforcement was concentrated on specific highway sections, targeting high average accuracy rates and reducing the overall risk of base rate accidents. This paper tackles the issues related to the existing problems of surveillance systems, such as the quick detection of unusual traffic patterns and the ability to respond to critical situations quickly. Furthermore, the combination of cutting-edge AI technologies represents a practical and easy approach to implementing intelligent transport systems that are safe, cost-effective, efficient, and novel. Future work will extend the system’s adaptability to various traffic environments, refine its performance under challenging conditions, and explore advanced deep-learning models. This research ultimately contributes to smarter traffic monitoring technologies, fostering safer travel environments, mitigating road hazards, and reducing congestion for a more efficient and secure transportation ecosystem.