AI-Based Construction Site Safety Behaviour Detection and Management System

Construction is regarded as one of the hazardous industries worldwide. A major cause of on-site accidents is unsafe human behaviour, particularly the failure to use personal protective equipment (PPE) such as a helmet. However, manual inspections are time-consuming and cannot be consistently covered throughout the entire project lifecycle, which weakens worker wellbeing, a key part of construction sustainability. To address this issue, this research develops a system based on artificial intelligence (AI) to detect helmet non-compliance in real time.

An AI model was trained to identify the presence of helmets on the heads of construction workers in site images. Specifically, the training used 6,519 real construction site photos collected from Google Images and Baidu Images, which were split into training (70%), testing (20%) and validation (10%) sets. Images were labelled into two groups: workers with helmets and workers without helmets.

The outcomes were of good performance. The trained model achieved a precision of 93.91% (i.e., accuracy when flagging non-compliance), a recall of 88.07% (i.e., the proportion of actual violations successfully detected), and an mAP@0.5 of 94.50% (an overall indicator of detection accuracy). Based on these findings, it was then built into a system with three parts: a detection engine that processes real-time site images to find helmets, a user interface that displays detection results on screen, and a project management module that tracks daily safety statistics and maintains records. As a result, managers can track compliance rates and obtain early warnings to support their decisions.

This research develops an AI-based construction site safety behaviour detection and management system to replace the traditional manual checks. It will be expected to help decrease the risks of accidents, enhance safety, and give project managers information on unsafe behaviour data to make more effective decisions and risk management.