A Traffic Sign Recognition Method with Bi-Level Routing Attention
Liu Ziqiong, Wenju Li, Liu Cui, Xiaosong Gao · 2023
Traffic sign recognition has long been an essential part of automatic driving. Accurate traffic sign recognition algorithms are conducive to the development of automatic driving. However, traffic signs in natural scenes are often small targets, which contain less feature information and are difficult to detect and recognize. Aiming at the problem of low recognition accuracy of small targets in traffic sign recognition, a traffic sign recognition algorithm based on a Bi-Level Routing Attention is proposed. It utilizes a single-stage detector YOLOv5 as the basic network, with a double-layer routing attention mechanism introduced, realizing more flexible computing allocation and content awareness through double-layer routing, which could capture long-range data without incurring an increasing computational overhead and memory usage. Furthermore, our approach achieves noteworthy recognition accuracy by extracting fine-grained information of small targets. In particular, the experiment shows the appealing results on the TT100K dataset, which detection accuracy map of reaches 92.9%, and the frame processing rate is 119.8FPS.