AK-Medoids-based Framework for Traffic Monitoring in Low Illumination Conditions

A novel unsupervised framework employs YOLOv4 (You Only Look Once, Version 4) to enable real-time object detection on roadways, significantly enhancing intelligent traffic monitoring. The system efficiently identifies vehicles, pedestrians, and other road objects in various challenging conditions, such as low-light environments, varying weather, and high-traffic density. By adopting an unsupervised approach, the framework eliminates the need for large labeled datasets, ensuring scalable deployment across diverse road networks. With integrated difference image analysis and K-medoids clustering, the framework boosts detection accuracy, achieving an impressive 92% detection accuracy-representing a 7% improvement over YOLOv4 and a 12% improvement over YOLOv3. Difference images highlight significant frame-to-frame changes, essential in low-illumination scenarios, while K-medoids clustering refines object grouping and reduces detection noise. These techniques complement YOLOv4‘s object detection capabilities, providing a robust solution for real-time monitoring. With edge computing for low-latency performance, the system is well-suited for large-scale deployment in Intelligent Transportation Systems. By combining unsupervised methods with advanced object detection, this framework delivers a scalable, efficient, and adaptive solution for continuous traffic monitoring, contributing to improved road safety.