A Deep Learning-Driven System for Real-Time Detection, Identification, and Documentation of Unsafe Behaviors in Construction: Case Study

9 pages•Published: June 2, 2026

Mohammad Dabash, Dyala Aljagoub, Edgar Small, Ri Na and Abdulaziz Banawi

Abstract

This pilot study presents an integrated deep-learning framework that not only detects but also identifies and records noncompliance with personal protective equipment (PPE) in real time on construction sites. The framework utilizes an object detection model, achieving a mean average precision (mAP) of 93%, and employs a rolling average supported by a real-time object tracking algorithm to minimize false positives in complex site environments. Detected violations trigger a deep learning facial recognition model that identifies the individual. All relevant data, including time of occurrence, nature of violation, and identity, are then stored in a Structured Query Language (SQL) database for subsequent analysis. This system addresses a critical research gap by going beyond detection to create a comprehensive record of unsafe behaviors, thereby enabling targeted interventions and data-driven safety enhancements. Despite its promising results, limitations such as occlusions and a relatively small dataset remain, suggesting that future work should incorporate larger, more diverse datasets to further refine and validate the approach.

Keyphrases: construction safety, construction workers, deep learning, object detection, you only look once (yolo)

In: Wesley Collins, Anthony Perrenoud and John Posillico (editors). Proceedings of Associated Schools of Construction 62nd Annual International Conference, vol 7, pages 970-978.

Links:	https://easychair.org/publications/paper/xD43
	https://doi.org/10.29007/q3bd

BibTeX entry

@inproceedings{ASC2026:Deep_Learning_Driven_System,
  author    = {Mohammad Dabash and Dyala Aljagoub and Edgar Small and Ri Na and Abdulaziz Banawi},
  title     = {A Deep Learning-Driven System for Real-Time Detection, Identification, and Documentation of Unsafe Behaviors in Construction: Case Study},
  booktitle = {Proceedings of Associated Schools of Construction 62nd Annual International Conference},
  editor    = {Wesley Collins and Anthony Perrenoud and John Posillico},
  series    = {EPiC Series in Built Environment},
  volume    = {7},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2632-881X},
  url       = {/publications/paper/xD43},
  doi       = {10.29007/q3bd},
  pages     = {970-978},
  year      = {2026}}

Download PDF Open PDF in browser