Edge-Enabled Detection for Adaptive Traffic Signal Control: A Comparative Study of YOLO-Based Models in Intelligent Transportation Networks

Traffic congestion remains a major challenge in urban environments, requiring intelligent systems that can adapt to real-time conditions. Object detection is central to these systems, providing the vehicle and infrastructure data needed for effective traffic management. This paper compares five YOLO (You Only Look Once) models (v5, v8, v10, v11, v12) on real-world traffic video, evaluating inference time, detection accuracy, confidence scores, and memory usage. Results show clear trade-offs: YOLOv5 offers the fastest inference but with higher error rates, YOLOv8 maximizes sensitivity at the cost of more false positives, YOLOv10 delivers the strongest precision with fewest wrong guesses, and YOLOv11/12 provide stable predictions but slower speeds. These findings highlight that no single version dominates; instead, model selection should be guided by deployment needs, balancing speed, accuracy, and resource constraints in real-time traffic management.