Scholarlink

<p>Traffic sign detection plays a pivotal role in enhancing the safety and efficiency of autonomous driving systems. Nevertheless, accurately detecting small, occluded, and visually complex traffic signs remains a significant challenge in real-world environments. To address these issues, this article proposes a Heterogeneous Attention YOLO model (HA-YOLO), a novel framework built upon the YOLOv8 model, designed to improve detection robustness under challenging conditions. The proposed approach incorporates High-level Screening-feature Pyramid Networks (HSFPN) for adaptive multiscale feature fusion, thereby reducing the loss of fine-grained details crucial for small or partially obscured signs. The proposed model also integrates Spatial and Channel Synergistic Attention (SCSA) and Triplet Attention are embedded into the C2f module, forming the C2f-Synergistic Attention (CSA) and C2f-Triplet Attention (CTA) submodules. These attention mechanisms enable the network to focus on salient spatial and semantic information, alleviating false positives and missed detections in complex scenarios. Under the tested conditions, experimental evaluations on TT-100K, CCTSDB, and GTSDB datasets demonstrate that HA-YOLO outperforms the baseline YOLOv8m by 5.1%, 0.6% and 15.9% in mAP@50, respectively, validating its effectiveness in real-world traffic sign detection tasks. The code of the proposed method will be publicly available at <a href="https://github.com/dengy1207/HAYOLO">https://github.com/dengy1207/HAYOLO</a>.</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

A heterogeneous attention YOLO model for traffic sign detection

  • 期刊论文

  • 发表日期：

    2025年4月21日

查看全文
  • Yan Deng,
  • Lidong Huang,
  • Xinsu Gan,
  • Youfang Lu,
  • Shunxiang Shi

摘要

Traffic sign detection plays a pivotal role in enhancing the safety and efficiency of autonomous driving systems. Nevertheless, accurately detecting small, occluded, and visually complex traffic signs remains a significant challenge in real-world environments. To address these issues, this article proposes a Heterogeneous Attention YOLO model (HA-YOLO), a novel framework built upon the YOLOv8 model, designed to improve detection robustness under challenging conditions. The proposed approach incorporates High-level Screening-feature Pyramid Networks (HSFPN) for adaptive multiscale feature fusion, thereby reducing the loss of fine-grained details crucial for small or partially obscured signs. The proposed model also integrates Spatial and Channel Synergistic Attention (SCSA) and Triplet Attention are embedded into the C2f module, forming the C2f-Synergistic Attention (CSA) and C2f-Triplet Attention (CTA) submodules. These attention mechanisms enable the network to focus on salient spatial and semantic information, alleviating false positives and missed detections in complex scenarios. Under the tested conditions, experimental evaluations on TT-100K, CCTSDB, and GTSDB datasets demonstrate that HA-YOLO outperforms the baseline YOLOv8m by 5.1%, 0.6% and 15.9% in mAP@50, respectively, validating its effectiveness in real-world traffic sign detection tasks. The code of the proposed method will be publicly available at https://github.com/dengy1207/HAYOLO.