Decentralized traffic management with Federated Edge AI: a reinforced transnet model for real-time vehicle object detection and collaborative route optimization

Urbanization and rising vehicle density have introduced significant challenges to traditional traffic management systems, including high latency, limited scalability, and concerns over data privacy. This study proposes a solution to these issues by developing a decentralized traffic management framework, DST-FedRL, which integrates Federated Edge AI, Deep Spatial-Temporal Transformer Networks (DST-net), and Reinforcement Learning. The aim is to enhance real-time vehicle detection, optimize traffic routes, and alleviate congestion while ensuring the privacy of sensitive data. The proposed system processes traffic data locally on edge devices, minimizing the need for data transfer and preserving privacy. The framework leverages Python-based tools, TensorFlow for model training, and reinforcement learning libraries for optimizing traffic routes. Federated learning is utilized to allow collaborative model updates across devices without the need to share raw data, while DST-net enables accurate vehicle detection under diverse and challenging conditions. The system was trained using the Road Traffic Video Monitoring Dataset, which includes a variety of traffic scenarios to ensure robust performance. The results show that DST-FedRL outperforms traditional methods, achieving a 97.8% vehicle detection accuracy, a 30% reduction in travel time, and a 27.9% decrease in fuel consumption. This study demonstrates that DST-FedRL is an effective, scalable, and privacy-preserving solution for intelligent traffic management in smart cities, offering significant improvements in efficiency and sustainability for urban mobility systems.