ScholarWorks@성균관대학교

초록

The construction industry is one of the most dangerous industries worldwide, and scaffold-related accidents are a significant concern. Despite the widespread use of scaffolds, the safety regulations pertaining to scaffolding remain among the most frequently violated, leading to a high frequency of accidents. Advancements in deep learning offer promising avenues for automating safety monitoring. However, the field is hindered by the lack of accessible datasets for training models in worker-behavior recognition. This study introduces the scaffolding worker inertial measurement unit (IMU) time-series (SWIT) dataset, which is designed to enrich the development of deep learning models for the automated recognition of construction worker behaviors. The SWIT dataset addresses the limitations of existing datasets by incorporating a wide range of hazardous behaviors, regulatory violations, and emergency situations specific to scaffolding. The dataset was developed through a rigorous process involving the analysis of sensor positions from previous studies, studies on abnormal behavior recognition, and scaffolding-safety regulations. It comprises ten categories of behaviors, including hazardous actions, near-miss incidents, and activities that may lead to musculoskeletal disorders. By providing a comprehensive collection of annotated time-series data from IMU sensors, this dataset aims to facilitate the development of robust deep learning models for automated worker-behavior recognition. © 2025 Elsevier Ltd

키워드