Research on Traffic Sign Detection Algorithm Using Enhanced Real-Time Detection Transformer

Automatic detection and recognition of traffic signs are essential for enhancing traffic safety and efficiency in intelligent transportation systems (ITS). Traditional object detection methods often face challenges in cluttered backgrounds and small object detection, with high computational complexity. To overcome these issues, this study proposes SE-DETR, a novel traffic sign detection algorithm based on the RT-DETR network structure. SE-DETR introduces three key modules: APBlock, EDR-CF, and AIFI-DPB. APBlock reduces the dimensional complexity of extracted features and aggregates multi-scale outputs, improving small object detection. EDR-CF combines atrous convolution and efficient upsampling to enhance performance in complex and occluded environments. AIFI-DPB employs attention mechanisms to capture global context and long-range dependencies, boosting detection accuracy. Experimental results demonstrate that SE-DETR outperforms RT-DETR, achieving superior performance in small object detection and complex background adaptation. SE-DETR improves mAP50 and mAP50-95 by 3.7% and 3.6%, respectively, while reducing computational cost to 39.0 GFLOPs. This makes SE-DETR a more efficient and effective solution for real-time traffic sign detection.