Development of a Fall Detection and Safety Communication System Using Small Language Models

Abstract

Fall incidents are a leading cause of injuries and fatalities in the construction industry, significantly impacting worker safety and productivity. Although many Al-based automated fall detection methods have been introduced, existing systems lack continuous communication support and often fail to address critical scenarios such as isolated work zones or high-risk tasks involving limited oversights. To address these limitations, this study proposes a new fall detection system for the construction industry using small language models (SLMs). Incorporating real-time conversation support within the system improves communication with emergency care teams and increases its utility in high-risk environments. We present the system architecture, integrate lightweight machine learning models, and evaluate the system's performance using the TinyLlama and Phi-3 models. Our assessment, which emphasizes relevance, correctness, and response speed, provides essential insights into effectively integrating language models into high-reliability systems for the construction industry.