Vehicle Classification and Traffic Density Analysis using RT-DETR

Abstract: Traffic congestion and vehicle monitoring have become major challenges in modern urban transportation systems, requiring efficient real-time traffic analysis and management solutions. This project presents an intelligent Vehicle Classification and Traffic Density Analysis System using the advanced RT-DETR deep learning architecture for fast and accurate vehicle detection in traffic environments. Unlike traditional CNN-based models, RT-DETR uses transformer-based attention mechanisms to improve detection accuracy in crowded and complex road conditions. The system can identify multiple vehicle categories such as cars, buses, trucks, motorcycles, bicycles, and auto-rickshaws from images and videos. A dual-platform deployment using Flask and Streamlit supports both testing and large-scale monitoring applications. The application includes secure user authentication, traffic monitoring dashboards, and automated traffic reporting features. Uploaded traffic videos are processed using frame-based inference techniques to estimate traffic density levels such as Low, Medium, and High. Confidence-based filtering helps reduce false detections caused by occlusion, lighting variations, and dense traffic conditions. The system utilizes OpenCV and PyTorch for efficient image and video processing in real-time environments. Experimental results show that the RT- DETR model provides high detection accuracy, stable performance, and an effective AI-driven solution for smart traffic monitoring and intelligent transportation systems.

Keywords: Vehicle Classification, Traffic Density Analysis, RT-DETR, Deep Learning, Computer Vision, Intelligent Transportation System, Object Detection, Real-Time Monitoring, Flask, Streamlit, OpenCV, PyTorch, Traffic Surveillance, Smart City, Vehicle Detection.