Scholarlink

<p>S.</p><p>To address the issue of ineffective monitoring of personnel movement and posture in offshore platform surveillance, we propose a detection method based on the YOLOv7-Pose cascaded with the YOLOv7 model for human pose-operational target association. By using YOLOv7-Pose, the method can be adopted for recognizing human targets and their joint points, postures, and movements in the offshore platform operational scene. The cascaded YOLOv7 model is used to detect possible operational targets from the obtained human posture recognition data set. The association between human posture and operational targets is established then. However, there are practical challenges to its use on offshore platform scenarios such as target variety, density, overlap, obscuration, and small size. Therefore, The YOLOv7 model’s recognition accuracy for operational targets is improved through the introduction of the PConv2D_BN_SiLU module and CBAM attention mechanism, as well as the use of a preferred loss function for optimization. Test experiments on the field dataset verify the feasibility and effectiveness of human posture recognition, operational target recognition, and the association detection method between the two counterparts.</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

Deep learning-based method for detecting the association between the personnel operating attitudes and the operational targets on offshore platform

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Feng Sun,
  • Shihai Zhang,
  • Chongnian Qu,
  • Zhenwei Li

摘要

S.

To address the issue of ineffective monitoring of personnel movement and posture in offshore platform surveillance, we propose a detection method based on the YOLOv7-Pose cascaded with the YOLOv7 model for human pose-operational target association. By using YOLOv7-Pose, the method can be adopted for recognizing human targets and their joint points, postures, and movements in the offshore platform operational scene. The cascaded YOLOv7 model is used to detect possible operational targets from the obtained human posture recognition data set. The association between human posture and operational targets is established then. However, there are practical challenges to its use on offshore platform scenarios such as target variety, density, overlap, obscuration, and small size. Therefore, The YOLOv7 model’s recognition accuracy for operational targets is improved through the introduction of the PConv2D_BN_SiLU module and CBAM attention mechanism, as well as the use of a preferred loss function for optimization. Test experiments on the field dataset verify the feasibility and effectiveness of human posture recognition, operational target recognition, and the association detection method between the two counterparts.