Traffic Sign Recognition Algorithm Based on Improved Real-Time Detection Transformer

Accurate detection of traffic signs is of great significance in the fields of road safety, intelligent transportation, and autonomous driving. This paper proposes the MY-DETR traffic sign detection method, which incorporates three innovative modules: EDB-SOP, AIFI-HiLo, and Inner-IoU. The EDB-SOP enhances robustness to small targets, occlusion, and complex backgrounds through multi-branch feature extraction and spatial optimization pooling technology. The AIFI-HiLo module integrates low-frequency and high-frequency attention mechanisms, optimizing the processing of multi-scale and multi-frequency features, and improving detection accuracy and robustness. The Inner-IoU loss function focuses on the inner overlapping area of the bounding box, improving the accuracy of traditional IoU in cases of low overlap. Experimental results show that MY-DETR significantly outperforms some mainstream algorithms on multiple traffic sign datasets, especially in terms of m<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$A$</tex> P50 and mAP50-95, providing a more efficient and accurate traffic sign detection solution for complex scenarios.