Scholarlink

<p>Construction sites are high-risk environments where improper delegation and unauthorized presence often lead to fatal accidents. This paper presents a dual-verification system based on deep learning algorithms to improve construction worker safety through two levels: (i) face verification for the authentication of registered workers’ identities and (ii) object verification to determine if the worker is performing their certified trade. A Convolutional Neural Network (CNN) classifier accurately identified eight registered workers from a synthetic database of 2,400 facial images. Meanwhile, a YOLOv8s-based detection system identifies tools in the same image database, which are classified into five major construction trades. The decision rule combines both determinants to classify each worker as “SAFE” or “UNSAFE.” Unlike existing literature that tackles identity and/or behaviour separately, this system combines both to provide real-time authentication and verification, thereby addressing a critical gap in safety enforcement. Under controlled validation conditions, the proposed framework successfully achieved 99% closed-set face recognition accuracy and 90% trade classification accuracy, highlighting its effectiveness. The proposed system facilitates the identity verification of a closed-set environment and offers an automated system that oversees compliance of workers and their tasks.</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 deep learning based dual-verification framework for ensuring construction workers safety

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月8日

查看全文
  • Megha Nain,
  • Shilpa Sharma,
  • Sandeep Chaurasia,
  • Niyaz Ahmad Wani

摘要

Construction sites are high-risk environments where improper delegation and unauthorized presence often lead to fatal accidents. This paper presents a dual-verification system based on deep learning algorithms to improve construction worker safety through two levels: (i) face verification for the authentication of registered workers’ identities and (ii) object verification to determine if the worker is performing their certified trade. A Convolutional Neural Network (CNN) classifier accurately identified eight registered workers from a synthetic database of 2,400 facial images. Meanwhile, a YOLOv8s-based detection system identifies tools in the same image database, which are classified into five major construction trades. The decision rule combines both determinants to classify each worker as “SAFE” or “UNSAFE.” Unlike existing literature that tackles identity and/or behaviour separately, this system combines both to provide real-time authentication and verification, thereby addressing a critical gap in safety enforcement. Under controlled validation conditions, the proposed framework successfully achieved 99% closed-set face recognition accuracy and 90% trade classification accuracy, highlighting its effectiveness. The proposed system facilitates the identity verification of a closed-set environment and offers an automated system that oversees compliance of workers and their tasks.