Multi-Camera Vehicle Recognition and Tracking for Urban Traffic Surveillance using Deep Learning

Urban traffic monitoring is essential for intelligent transportation systems and the infrastructure of smart cities, enabling real-time analysis of traffic patterns, enforcement of regulations, and rapid incident detection. Existing vehicle tracking frameworks, such as FairMOT-MCVT, were primarily developed for highway scenarios with minimal occlusions and uniform traffic, limiting their effectiveness in complex urban environments characterized by congestion, frequent occlusions, diverse vehicle types, and varying lighting and weather conditions. This paper presents an enhanced FairMOT-MCVT framework specifically designed for urban surveillance by integrating real-time vehicle detection, license plate recognition, and re-identification into a unified pipeline. The system employs YOLOv5 for vehicle detection, a CRNN-based model for license plate recognition, and advanced re-identification modules with attention mechanisms to maintain identity consistency across multiple non-overlapping cameras. Experiments on the CCPD and VeRi-776 datasets show that YOLOv5 achieved a precision of 87%, recall of 80%, and F1 score of 0.84 for vehicle detection; CRNN reached 75% precision, 73% recall, and 0.745 F1 for license plate recognition; and FairMOT’s re-identification embeddings attained 75% precision, 73% recall, and 0.74 F1. The results demonstrate that the proposed framework is accurate, robust, and capable of realtime performance, making it suitable for practical urban applications such as traffic flow analysis, vehicle counting, law enforcement, and incident detection, thus advancing the capabilities of intelligent transportation systems and smart city infrastructure.