| Title |
Proposal and Verification of the Faster R-CNN Regarding the Worker and Machine Interference Scope Detection Model to Prevent On-site Safety Accidents |
| Authors |
왕택보(Wang, Zepu) ; 김장순(Kim, Jang-Soon) ; 함남혁(Ham, Nam-Hyuk) ; 김재준(Kim, Jae-Jun) |
| DOI |
https://doi.org/10.5659/JAIK.2022.38.4.217 |
| Keywords |
Machine Learning; Construction Safety Management; Deep Learning; Faster R-CNN; Visual Inspection Model; Image Analysis |
| Abstract |
Safety management of construction projects have a significant impact on the construction project’s schedule and the control carried out on
site. Current site safety monitoring methods are highly dependent on manual labor; human errors can occur through missing content. This
study aims to resolve these issues by applying machine learning visual detection algorithms to identify unsafe behaviors of workers at
construction sites, to enhance external monitoring of workers and to relatively reduce the occurrence of safety accidents. A proposed method
combines an object detection algorithm and spatial localization relationship definition. Only the machinery and workers at the construction site
need to be accurately detected and the definition of spatial location relationship can be used to identify dangerous behaviors. A monitoring
network framework suitable for this study was constructed with the environmental characteristics and image features of a construction site.
The machines and workers were detected from construction images based on the Faster R-CNN algorithm for a computer to obtain the visual
detection data from the construction site. Three spatial concepts were defined to determine the position relationships of machines and workers
in these images. The detected location information of machines and workers at the construction site were combined and presented in a
visualized form. Based on the results of this research, it confirmed that the method and performance were suitable for construction site safety
management, which is expected to contribute to the speed, level of accuracy and risk warning with the application of automated progress
monitoring methods. |