| Title |
State-Based Interpretation of Scaffold Fall Risk Using Finite-State Machines and Hierarchical Reasoning |
| Authors |
이주형(Lee, Joo-Hyeong) ; 유무영(Yoo, Moo-Young) |
| DOI |
https://doi.org/10.5659/JAIK.2026.42.8.373 |
| Keywords |
Scaffold Fall Accidents; State-Based Risk Interpretation; Finite-State Machine; Policy-Based Re-Evaluation; Construction Safety |
| Abstract |
Falls from scaffolding remain a major cause of fatal accidents on construction sites, especially during mobile and system scaffold operations.
Fall risk is affected by interacting factors including guardrail installation, edge work, and rainfall. However, most vision-based safety studies
focus on detecting individual hazards and do not provide a structured mechanism for integrated risk-state interpretation. This study proposes a
state-based risk assessment framework using an agent-inspired policy-based re-evaluation architecture integrating computer vision, a Finite-State
Machine (FSM), and a Vision?Language Model (VLM). Visual data are used to identify hazard variables (guardrail, edge work, rain). The
hazard vector is mapped to predefined risk states (S0?S3) via explicit FSM transition rules. A continuous risk score is introduced to
represent relative risk intensity, and the VLM generates natural-language explanations for each state. Experiments on scaffold-work image
datasets show that rainfall shifts the risk-state distribution toward higher levels. The FSM classification and risk score exhibit consistent
escalation patterns, and generated explanations remain logically consistent with hazard inputs. The framework provides a structured,
interpretable approach for scaffold fall risk assessment under multi-hazard conditions. |