Advancing Smart City Infrastructure: A Deep Learning-Based Framework for Real-Time Traffic Monitoring and Violation Detection Using YOLOv11

Abstract

Urban traffic congestion and violations of dedicated bus lanes in metropolitan cities, such as Jakarta, are significant challenges affecting the efficiency of public transportation systems. Traditional traffic monitoring methods are insufficient to address these issues, particularly in real-time violation detection. This research proposes an AI-based smart traffic monitoring framework using YOLOv11 for real-time detection of vehicle violations in TransJakarta’s Bus Rapid Transit (BRT) lanes. The study aims to improve urban mobility by enhancing the detection accuracy and speed of traffic monitoring systems. The methodology involves data collection from surveillance cameras, data annotation using Roboflow, and model training with YOLOv11, utilizing transfer learning and hyperparameter optimization. The system's performance is evaluated through precision, recall, F1-score, and mean Average Precision (mAP@0.5), as well as real-time inference speed. The results show that YOLOv11 achieves a mAP@0.5 of 0.946 and an F1-score of 0.898, demonstrating the model's high accuracy in detecting vehicle violations across different lighting conditions. Real-time inference is achieved at a rate of 35-40 FPS, making it suitable for deployment in real-world urban environments. This research concludes that the YOLOv11-based framework is an effective solution for automated traffic monitoring, offering significant implications for smart city development and intelligent transportation systems. Further research is needed to address lighting challenges and improve the system's scalability across various urban settings.