YOLO-Based Vehicle Detection and Classification for Multi-Lane Free Flow Systems

Abstract

Multi-Lane Free Flow (MLFF) systems have emerged as a promising approach to reducing congestion at toll gates by enabling uninterrupted toll transactions without requiring vehicles to stop. To support this implementation, this study proposes a YOLOv9-based vehicle detection and classification system optimized using grid search and the SGD optimizer. The proposed system is designed for edge computing deployment and evaluated in a real multi-lane toll environment using an NVIDIA Jetson AGX Orin device. Experimental results show that the optimized model achieves a mean Average Precision at IoU threshold 0.50 (mAP@50) of 92.6%, an F1-score of 92.0%, and an inference time of 70.8 ms, corresponding to 14.1 FPS. These results indicate that the proposed system provides a balanced trade-off between detection accuracy and computational efficiency for edge-based MLFF applications. Unlike previous studies that primarily evaluated vehicle classification models using offline datasets, this study includes implementation and testing in a real multi-lane toll scenario. The findings demonstrate the potential of YOLO-based vehicle detection and classification to support practical MLFF deployment, although further evaluation under more diverse traffic, lighting, and environmental conditions is still required.