Real-Time Smart Traffic Monitoring System Using YOLOv12 and ByteTrack: An Integrated Approach for Vehicle Detection, Counting, Speed Estimation, and Incident Detection

Abstrak

Effective traffic management in modern urban environments demands real-time monitoring systems capable of extracting actionable insights from video surveillance data. This paper presents a comprehensive smart traffic monitoring system that integrates state-of-the-art deep learning techniques for vehicle detection, multi-object tracking, and traffic analysis. The proposed system employs YOLOv12, an attentioncentric object detector, for robust vehicle identification across diverse conditions, combined with ByteTrack for consistent multiobject tracking through occlusions. Building upon this detectiontracking foundation, we implement a configurable multi-line counting mechanism supporting arbitrary directional boundaries, speed estimation through pixel-to-meter calibration, congestion detection via joint density-velocity analysis, and incident identification using bounding box overlap metrics. The complete system is deployed through a FastAPI backend with WebSocket streaming to a web-based dashboard, enabling real-time visualization and interaction. Experimental evaluation demonstrates counting accuracy exceeding 94% across varied traffic scenarios, speed estimation RMSE below 3.5 km/h against radar ground truth, and sustained real-time processing at over 30 FPS for singlecamera deployment. The modular architecture and practical deployment considerations distinguish this work from purely algorithmic contributions, providing a foundation suitable for both research exploration and operational deployment.

Identifier

• DOI: https://doi.org/10.36227/techrxiv.177160628.82572286/v1
• OpenAlex: https://openalex.org/W7130702083