Smart Junctions: Yolov5, DeepSORT, and Direction Analysis for Traffic Enhancement

In realm of urban traffic management, this study proposes an innovative approach to combat the escalating challenge of traffic congestion and enhance transportation efficiency. The Adaptive traffic management system leverages cutting-edge technologies, specifically utilizing You Only Look Once (YOLO) version 5 for real-time vehicle recognition and Deep Simple Online Real-Time Tracking (DeepSORT) for robust vehicle tracking. Focusing on Hyderabad's intricate road network, where over 3.2 million vehicles navigate. This work aims to optimize traffic flow and alleviate congestion. The system categorizes vehicles into four types (bike, car, bus, and truck). This study employs advanced techniques, including Gaussian filtering for noise reduction in video feeds, YOLOv5, and DeepSORT. After thorough pre-processing, Vehicle detection and tracking system performs direction analysis too. The output is structured data in Comma Separated Value (CSV) format, encompassing vehicle IDs, categories, and directional paths. Notably, this model labeled as Vehicle Direction Analysis (VDA), achieves 87.93% accuracy, surpassing YOLOv5n and YOLOv5x with 77.84% and 84.37% respectively. This comprehensive and systematic approach, combining DeepSORT sophisticated tracking capabilities with YOLOv5' s object identification architecture, contributes to data-driven decision-making for effective urban traffic management.