TSNS-YOLO: An Improved Traffic Sign Detection Network for Natural Scenes Based on YOLOv7
Xinke Wang, Jingyu Sun · 2024
Traffic sign detection is an important part of intelligent transportation systems (ITS) and autonomous driving. Many factors in traffic sign detection tasks can affect the effective detection of traffic signs, including lighting variations, weather conditions, occlusion, deformation, and vandalism. In addition, traffic sign objects often account for only a small portion of the entire image, and detecting small traffic signs adds to the difficulty of traffic sign detection. An improved traffic sign detection network for natural scenes based on YOLOv7 is proposed to address these issues, which is called TSNS-YOLO. Firstly, an improved receptive field block (IRFB) module is added in multiple layers after the backbone network to expand the receptive field to make the network more suitable for detecting datasets containing smaller traffic signs. Secondly, the GSConv module is used in the neck to ensure the detection effect while reducing the number of network parameters. Thirdly, SimAM is also introduced in the network to enhance the detection effect. The experimental results show that the improved network detection based on YOLOv7 in this paper is better than several existing methods, and the detection precision and detection speed are also improved over the original YOLOv7 network. TSNS-YOLO achieved the Mean Average Precision (MAP) of 85.54% and 99.02% on the CCTSDB2021 dataset and the GTSDB dataset, respectively, while the average detection time reached the real-time detection requirement in both cases.