Potential Application of Deep Learning and UAS for Guardrail Safety Inspection

Abstract

Unmanned Aerial Systems (UAS) can provide valuable information about on-site compliance with safety regulations, especially identifying workers in areas without guardrails or fall arrest systems. Despite the advances in using Machine Learning (ML) and, more specifically, Deep Learning (DL) algorithms for detecting safety systems in construction, the literature indicates a gap regarding automatic guardrail recognition. Therefore, this paper proposes a set of criteria for data collecting and processing using UAS and DL for safety inspections in temporary guardrails while producing cast-in-place concrete wall systems. For this research, an exploratory case study was adopted as the research strategy, developed according to the following steps: (a) database image analysis, (b) field study on constructions, (c) formal meetings, and (d) survey carried out with ML/DL specialists. Results show the main failures in guardrails of cast-in-place concrete wall systems, analyzing which can be inspected using UAS visual assets and ML/DL techniques. Also, it indicates the more adequate construction stages to perform safety inspections on guardrails. These findings may guide future research using UAS and DL algorithms for inspecting guardrail safety systems to further contribute to managers' decision-making.