Worker Safety at Heights using Deep Learning: Detection of Helmet and Harness Compliance with DETR Architecture

Mrs.T.Jyothi,M.Tech,

Electronics and communication engineering,Annamacharya institute of technology and sciences,

Tirupati,@gmail.com

  Galiveedu Reddy Prasad,B.Tech

Electronics and communication engineering,Annamacharya institute of technology and sciences,Tirupati,

reddygaliveedu71@gmail.com

 Murikinati Ashok Kumar Reddy,  B.Tech,

Electronics and communication engineering,Annamacharya institute of technology and sciences,Tirupati,

ashokkumarreddy7981@gmail.com

 Narike Praveen Kumar,B.Tech,

Electronics and communication engineering,Annamacharya institute of technology and sciences,Tirupati,

narikepraveenkumar@gmail.com

Gaddamolla sreedevi,B.Tech,

Electronics and communication engineering,Annamacharya institute of technology and sciences,Tirupati,

sreedevigaddamollasreedevi@gmail.com

ABSTRACT:Worker safety at construction sites and industrial environments, especially at elevated working areas, remains a major concern due to frequent accidents caused by the absence of personal protective equipment (PPE) such as safety helmets and harnesses. Manual monitoring of PPE compliance is inefficient, error-prone, and not scalable for large worksites. This paper proposes a deep learning–based automated vision system for real-time detection of helmet and safety harness usage using the Detection Transformer (DETR) architecture. The proposed model leverages transformer-based global feature modeling to accurately identify workers and verify PPE compliance under complex backgrounds, varying lighting conditions, and occlusions. The system is trained and evaluated on annotated construction-site datasets and demonstrates robust detection performance with improved accuracy and reduced false positives compared to traditional CNN-based detectors. The proposed approach enhances proactive safety monitoring and supports intelligent surveillance systems for accident prevention in hazardous work environments.