Scholarlink

The standardized removal of personal protective equipment (PPE) by medical staff is critical for ensuring medical safety. Current manual visual inspection methods in healthcare institutions are constrained by operator fatigue, inconsistent evaluation criteria, and lack of real-time monitoring. Leveraging advancements in human pose estimation and video behavior recognition, this paper introduces a novel video-segmentation-based framework with a multi-scale feature fusion architecture. This paper employs Enhanced Regularized Residual Inverted Block (RRIB) pre-activation and Channel-Spatiotemporal Attention Module (CSAM) to achieve lightweight discriminative modeling. Experimental results on public datasets demonstrate significant improvements over state-of-the-art methods: the proposed model achieves an AUC of 92.56% (1.72% higher than I3D, 4.17% than C3D, and 2.62% than TimeSformer), recall of 91.87%, and F1-score of 92.15%. These metrics validate its superior discriminative ability and generalization performance for detecting PPE removal sequence irregularities while reducing computational complexity by 38% compared to baseline models.

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 Detection Method for the Dismantling Sequence of Protective Equipment for Medical Staff Based on Video Segmentation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Qiang Zhang,
  • Xinzi Xu,
  • Lixin Yang,
  • Ying Qi,
  • Teng Wan,
  • Rui Xu,
  • Jie Li,
  • Lei Hu

摘要

The standardized removal of personal protective equipment (PPE) by medical staff is critical for ensuring medical safety. Current manual visual inspection methods in healthcare institutions are constrained by operator fatigue, inconsistent evaluation criteria, and lack of real-time monitoring. Leveraging advancements in human pose estimation and video behavior recognition, this paper introduces a novel video-segmentation-based framework with a multi-scale feature fusion architecture. This paper employs Enhanced Regularized Residual Inverted Block (RRIB) pre-activation and Channel-Spatiotemporal Attention Module (CSAM) to achieve lightweight discriminative modeling. Experimental results on public datasets demonstrate significant improvements over state-of-the-art methods: the proposed model achieves an AUC of 92.56% (1.72% higher than I3D, 4.17% than C3D, and 2.62% than TimeSformer), recall of 91.87%, and F1-score of 92.15%. These metrics validate its superior discriminative ability and generalization performance for detecting PPE removal sequence irregularities while reducing computational complexity by 38% compared to baseline models.