| Author: | Kong, Ting |
| Title: | Explainable AI-driven UAVS for real-time fall-from-height risk monitoring in construction |
| Advisors: | Li, Heng (BRE) |
| Degree: | Ph.D. |
| Year: | 2026 |
| Subject: | Falls (Accidents) -- Prevention Drone aircraft Construction industry -- Safety measures Artificial intelligence Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Building and Real Estate |
| Pages: | xvi, 144 pages : color illustrations |
| Language: | English |
| Abstract: | Falls from height (FFH) have been consistently identified as a leading cause of injuries and fatalities in the construction industry, with most fall-related incidents attributable to the absence of or inadequate safety protections (e.g., unprotected open edges). Typically, traditional approaches heavily rely on manual inspections, which pose numerous challenges (i.e., being time-consuming) in real-life construction sites due to the nature of their complex and dynamic working conditions. To overcome these shortcomings, emerging technologies such as Building Information Modelling (BIM), Deep Learning (DL), and Computer Vision (CV) offer promising solutions for the automatic detection of hazards to improve safety performance in construction. Despite their success, the following limitations hinder their deployment and applications on construction sites: (1) existing deep learning models are complex non-linear statistical models, which suffer significantly from the "black box" problem, making it difficult for on-site managers to trust the outcomes produced by such AI systems; (2) prior computer vision approaches primarily rely on 2D/3D images, which often cannot localize hazards precisely within 3D models or associate the hazards with specific building components; and (3) many studies tend to be retrospective and focus on examining the conditions that contribute to unsafe behaviour from a psychological perspective, which are unable to visually comprehend the dynamic and complex conditions that influence unsafe behaviour. To address these limitations, our research aims to investigate the following research question: How could we improve the accuracy and interpretability of deep learning models in monitoring Fall-From-Height (FFH) risk in construction sites by using Unmanned Aerial Vehicles (UAVs) and computer vision systems? To this end, we adopted a design science research (DSR) paradigm to guide the conceptual framework design, model development, and evaluation. Following such a design science approach, we design, implement, and evaluate an XAI-enabled UAV-based computer vision system that (1) detects unprotected edges and guardrails, (2) precisely localizes hazards related to FFH within a three-dimensional (3D) construction site model and links the hazards to specific building components, and (3) predicts workers approaching unprotected edges. Real-life projects are used to validate the effectiveness and feasibility of the proposed approach. The research findings presented in this research have demonstrated that the proposed explainable Artificial Intelligence (XAI)-based computer vision system is able to accurately identify hazards. We suggest that our research can be used to improve worker safety and reduce operational costs by replacing labour-intensive inspections with automated monitoring, delivering a transparent, resilient, and practical safety solution for workers, safety managers, regulators, and the construction industry. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14573