Scholarlink

<p>Effective detection of traffic signs during vehicle operation is a critical component of intelligent driving technologies. It is a critical component for ensuring adherence to traffic regulations and enhancing overall safety. However, traffic sign detection is often affected by environmental factors, leading to false detections and missed detections. To address these challenges, we introduce TS-YOLO, a traffic sign detection model optimized within the YOLOv8n framework. The proposed model incorporates a 160 × 160 detection head to expand the receptive field and improve small-object detection capability. Structural re-parameterization is applied to the neck of the network to reduce computational complexity during inference, thereby improving real-time performance. In addition, partial convolution modules are integrated into the neck using a two-stage approach, which processes features via partial convolution and leverages the principles of the PConv module to extract partial feature information, reducing computational cost while improving efficiency. Furthermore, an AP-FasterBlock module is integrated into YOLOv8n to capture fine-grained features during detection. This module adopts depthwise separable convolution, combining depthwise and pointwise convolutions, which not only reduces the number of parameters but also strengthens small-object detection. To further enhance TS-YOLO’s performance under foggy conditions, a fog generation model combining AdaBins with an optical model is proposed to augment the TT100K dataset with foggy-weather information. Experimental results demonstrate that TS-YOLO achieves a 1.3% improvement in precision, a 2.6% increase in recall, a 1.8% gain in mean average precision (mAP), and a 12 FPS boost compared with YOLOv11n. These findings confirm that TS-YOLO offers superior detection accuracy and real-time performance.</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

TS-YOLO: a small traffic sign detection algorithm for various harsh driving conditions in bad weather

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月6日

查看全文
  • Ziyin Zhang,
  • Bingyin Feng,
  • Hao Liu

摘要

Effective detection of traffic signs during vehicle operation is a critical component of intelligent driving technologies. It is a critical component for ensuring adherence to traffic regulations and enhancing overall safety. However, traffic sign detection is often affected by environmental factors, leading to false detections and missed detections. To address these challenges, we introduce TS-YOLO, a traffic sign detection model optimized within the YOLOv8n framework. The proposed model incorporates a 160 × 160 detection head to expand the receptive field and improve small-object detection capability. Structural re-parameterization is applied to the neck of the network to reduce computational complexity during inference, thereby improving real-time performance. In addition, partial convolution modules are integrated into the neck using a two-stage approach, which processes features via partial convolution and leverages the principles of the PConv module to extract partial feature information, reducing computational cost while improving efficiency. Furthermore, an AP-FasterBlock module is integrated into YOLOv8n to capture fine-grained features during detection. This module adopts depthwise separable convolution, combining depthwise and pointwise convolutions, which not only reduces the number of parameters but also strengthens small-object detection. To further enhance TS-YOLO’s performance under foggy conditions, a fog generation model combining AdaBins with an optical model is proposed to augment the TT100K dataset with foggy-weather information. Experimental results demonstrate that TS-YOLO achieves a 1.3% improvement in precision, a 2.6% increase in recall, a 1.8% gain in mean average precision (mAP), and a 12 FPS boost compared with YOLOv11n. These findings confirm that TS-YOLO offers superior detection accuracy and real-time performance.