Application of Hazard Identification Method in Complex Construction Scene Based on YOLO-v8
The study on unsafe behavior and construction progress detection in complex construction scenes based on machine vision is still in the initial stage.This paper proposes an unsupervised learning detection model for unsafe behavior and construction progress in complex construction scenes based on the YOLO-v8 model.Self-su-pervised learning annotation of datasets is carried out through MaskCut module to reduce the workload of data set annotation.DRM and CRM modules are applied to segment occlusion features layer by layer and reconstruct background.In addition,the YOLO-v8 basic model is trained with the self-supervised annotation data set of MaskCut module to detect and study 15 common unsafe behaviors and construction progress in 8 typical construc-tion scenarios.The experimental results show that the proposed method has high detection performance for a vari-ety of unsafe behaviors in complex construction scenes,and the average accuracy rate reaches 84.63%.The pro-posed method has high application potential for a variety of common construction unsafe behaviors and fusion of occlusion object decoupling and reconstruction.And it has high reference value for the research of construction progress detection based on machine vision.
complex construction scenesource of dangerconstruction progressYOLO-v8
NETL
NSTL
万方数据
