Individual Component Detection of a Scaffolding Assembly for Vision-Based Safety Check
Journal
Construction Research Congress 2024, CRC 2024
Journal Volume
4
ISBN
9780784485293
Date Issued
2024-01-01
Author(s)
Abstract
Fall from heights accounts for 50% of the accidents on construction sites, and 20% are scaffolding-related. Therefore, ensuring the safety of scaffolding installation can significantly reduce accidents onsite. Scaffoldings are erected by assembling components such as footboards, standards, ledgers, and cross-bracings. A safety check must ensure all necessary components in a scaffolding unit are installed in the correct order. While vision-based scaffolding detection has become more accurate with the recent success of deep learning detection algorithms, the detection of individual scaffolding units and their components is still challenging due to the scaffolding component size and shape. The size and shape factors pose many difficulties for the vision algorithms to detect thin structures and components of similar looks. To address the difficulties, this study proposes a method to segment the scaffolding units and components from a point cloud using a deep learning-based 3D semantic segmentation model. After segmentation, a rule-based approach can be applied to check the missing components. The method has been implemented in a construction project, and the preliminary results confirm its applicability for drawing workers' attention to the missing scaffolding components, thereby improving the construction site's safety.
Type
conference paper