AI-Based Scaffolding Deficiency Detection As A Proactive Construction Safety Intervention: A Conceptual Review Of Methodological Approaches

Abstract

Falls from height and structural collapse associated with scaffolding remain among the most persistent contributors to construction fatalities worldwide, and the problem is sharper still in resource-constrained settings where inspection capacity is thin on the ground. This paper sets out a conceptual outline for an AI-based scaffolding deficiency detection system designed as a proactive safety intervention rather than a post-incident diagnostic. Drawing on a thematic review of computer vision methods applied to scaffolding inspection between 2019 and 2026, the study maps what has been tried, where the evidence converges, and where it still falls short. Four structural deficiency classes grounded in SANS 10085-1:2024 (missing cross-brace, missing guardrail, corroded or damaged members, and missing base plate) form the target taxonomy, and a multi-model pipeline combining YOLO26, Mask R-CNN, and a Hough Transform hybrid is proposed. The paper reflects critically on dataset construction, class imbalance, annotation subjectivity, hardware and edge-deployment feasibility, and cross-jurisdictional generalisation.