RSO-YOLO: Traffic Sign Detection Method Based on Reparameterization and Shallow Feature Enhancement

Traffic sign detection is a critical task in autonomous driving and intelligent transportation systems. However, existing detection methods struggle to achieve an effective balance between accuracy and inference latency, especially in applications requiring small object detection and real-time perfo...

Full description

Bibliographic Details
Published in:IEEE Open Journal of Intelligent Transportation Systems
Main Authors: Guowen Chang, Qingzhan Zhao, Xuewen Wang
Format: Article
Language:English
Published: IEEE 2026-01-01
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11520316/
Description
Summary:Traffic sign detection is a critical task in autonomous driving and intelligent transportation systems. However, existing detection methods struggle to achieve an effective balance between accuracy and inference latency, especially in applications requiring small object detection and real-time performance. To address this challenge, this paper proposes an improved YOLO-based traffic sign detection method called RSO-YOLO. This approach enhances shallow feature extraction capabilities by optimizing network structure, deepening the downsampling process, effectively reducing feature loss and improving small object detection accuracy. To further enhance inference efficiency, this paper combines reparameterization techniques with large-kernel convolutions to expand the receptive field while reducing computational costs, significantly improving the detector’s inference speed. Experimental results on the GTSDB and TT100K datasets show that the RSO-YOLO-S model achieves 98.2% and 83.6% in mAP50 metrics, representing improvements of 3.8 and 11 percentage points compared to baseline models, while maintaining inference latencies of 11.6ms and 30.3ms. Ablation studies verify the contribution of each module to model performance improvement. This research successfully balances detection accuracy and inference latency while ensuring model lightweight properties, providing a more efficient and reliable solution for visual perception tasks in intelligent transportation systems.
ISSN:2687-7813