Scholarlink

<p>This paper proposes RSPV-YOLO, an advanced model for helmet detection in two-wheeler riders, built on YOLOv8. The model addresses common challenges such as false positives, missed detections, and difficulties in identifying small targets in complex traffic conditions. Key contributions include the C2f-RCAB module in the backbone, integrating a channel attention mechanism to enhance helmet feature extraction, especially in small-target and complex-background scenarios. In addition, the lightweight SPPELAN module replaces the SPPF, improving multi-scale receptive fields and detection efficiency. In the neck structure, the pixel-attention-guided fusion module (PagFM) and VoV-GSCSP module optimize shallow and deep feature fusion while reducing computational overhead. Experimental results show RSPV-YOLO outperforms YOLOv8, achieving mAPs of 85.4% and 93.6% on the TWHD and EBHD datasets. With 2.74M parameters and an inference speed exceeding 50 FPS, RSPV-YOLO offers a strong balance between accuracy and real-time performance, making it suitable for automated traffic monitoring and law enforcement.</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

RSPV-YOLO: two-wheeler helmet detection method based on residual channel attention and pixel-guided multi-scale fusion

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Junlong An,
  • Wei Chen,
  • Yunfeng Ni,
  • Fan Chang

摘要

This paper proposes RSPV-YOLO, an advanced model for helmet detection in two-wheeler riders, built on YOLOv8. The model addresses common challenges such as false positives, missed detections, and difficulties in identifying small targets in complex traffic conditions. Key contributions include the C2f-RCAB module in the backbone, integrating a channel attention mechanism to enhance helmet feature extraction, especially in small-target and complex-background scenarios. In addition, the lightweight SPPELAN module replaces the SPPF, improving multi-scale receptive fields and detection efficiency. In the neck structure, the pixel-attention-guided fusion module (PagFM) and VoV-GSCSP module optimize shallow and deep feature fusion while reducing computational overhead. Experimental results show RSPV-YOLO outperforms YOLOv8, achieving mAPs of 85.4% and 93.6% on the TWHD and EBHD datasets. With 2.74M parameters and an inference speed exceeding 50 FPS, RSPV-YOLO offers a strong balance between accuracy and real-time performance, making it suitable for automated traffic monitoring and law enforcement.