| Traffic monitoring and violation detection are critical components of modern intelligent transportation systems. In this paper, we present a real-time traffic analysis system that integrates object detection, tracking, and behavior analysis within a unified pipeline. The proposed system is based on the YOLOv5 object detection framework, where the standard CSPDarknet53 backbone is replaced with EfficientNet-B3 to enhance feature extraction. Vehicle tracking is performed using the SORT algorithm, enabling consistent identification of objects across frames and supporting trajectory-based analysis. The system is applied to multiple tasks, including vehicle detection, speed estimation, behavior-based violation detection, and traffic congestion monitoring. Experimental evaluation is conducted on both a benchmark dataset and a custom traffic dataset consisting of 750 annotated images. Results show that the proposed backbone modification improves localization performance, particularly in terms of mAP@50-95, while maintaining real-time processing speeds of approximately 30 FPS. Although the system relies on established components, the contribution lies in their integration into a practical real-time pipeline and in the analysis of backbone replacement within the YOLOv5 framework. The results demonstrate the feasibility of deploying such systems for real-world traffic monitoring applications, while also highlighting limitations related to dataset size and camera calibration. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.