Traffic Sign Detection Based on Lightweight RT-DETR

Traffic sign detection is crucial for the functionality of autonomous driving systems, directly impacting the safety and compliance of vehicle operation with traffic regulations. It is not only a vital component of autonomous driving technology but also a pivotal technology for advancing and refining autonomous systems. This paper proposes an improvement to the RT-DETR object detection algorithm, substituting the original ResNet network with a lightweight ShuffNetv2 network as the backbone of RT-DETR. Employing ShuffNetv2 as the primary network backbone reduces computational load and parameter quantity while maintaining the learning capability of RT-DETR and enhancing detection speed. Experimental results show that the improved RT-DETR achieves an average precision (MAP) of 98.2% on the CCTSDB2021 dataset, with an inference speed of 19.2 ms. Compared to using ResNet-50 as the backbone, computational cost is reduced by 55.93<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, the parameter count decreases by 63.38%, and inference speed is increased by 21%.