Scholarlink

<p>Traffic columns and traffic cones, as crucial traffic guidance tools, are often used together to maintain the orderliness of road traffic. However, due to their small size and frequent use in large quantities and high density, it is difficult for autonomous vehicles to accurately capture and locate them. Therefore, we choose the YOLOv8n network as the baseline and propose a novel strip-shaped object recognition model, namely SOR-YOLO. Firstly, we utilize the SGE attention mechanism and GhostConv module to construct a lightweight grouped enhanced C2f module, namely LGEC2f module, which reduces the number of parameters and computations of the network. Secondly, we integrate the MSCA module into the SPPF module to form the MSCSPPF module, thus enhancing the network’s ability to recognize strip-shaped objects. Finally, we replace the loss function of the YOLOv8n network with the WIoU loss function to accelerate the convergence of the model. Experiments show that SOR-YOLO improves the mAP50 and mAP75 by 2.7% and 3.0% respectively, and the mAP50-95 by approximately 1.1% compared to YOLOv8n in the recognition of traffic columns and traffic cones, demonstrating higher recognition accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SOR-YOLO: a strip-shaped object detection network for traffic columns and traffic cones

  • 期刊论文

  • 发表日期：

    2025年4月23日

查看全文
  • JiaHao Wang,
  • Yongqiang Wang,
  • Congling Zhou,
  • Hao Wang

摘要

Traffic columns and traffic cones, as crucial traffic guidance tools, are often used together to maintain the orderliness of road traffic. However, due to their small size and frequent use in large quantities and high density, it is difficult for autonomous vehicles to accurately capture and locate them. Therefore, we choose the YOLOv8n network as the baseline and propose a novel strip-shaped object recognition model, namely SOR-YOLO. Firstly, we utilize the SGE attention mechanism and GhostConv module to construct a lightweight grouped enhanced C2f module, namely LGEC2f module, which reduces the number of parameters and computations of the network. Secondly, we integrate the MSCA module into the SPPF module to form the MSCSPPF module, thus enhancing the network’s ability to recognize strip-shaped objects. Finally, we replace the loss function of the YOLOv8n network with the WIoU loss function to accelerate the convergence of the model. Experiments show that SOR-YOLO improves the mAP50 and mAP75 by 2.7% and 3.0% respectively, and the mAP50-95 by approximately 1.1% compared to YOLOv8n in the recognition of traffic columns and traffic cones, demonstrating higher recognition accuracy.