3D Reconstruction and Woker Recognition Techniques Based on Monocular Depth Estimation for Small and Medium-Sized Construction Site Applications

초록

The environment of the construction site is complex and there are various risk factors, making it difficult to recognize multiple workers and heavy equipment. The technology proposed in this study supplemented the limitations of the existing 2D depth information analysis and 3D point cloud reconstruction by applying a single RGB camera and neural network-based algorithm. This study verified reliable spatial reconstruction performance without using expensive sensors through experiments at actual construction sites, highlighting the possibility of practical application to small and medium-sized construction sites. Through quantitative evaluation using IoU indicators, the 3D point cloud-based object recognition technology showed an average improvement of about 19.34% compared to the 2D method. This research finding is expected to contribute to further developing safety management in the construction sector through the integration of the safety management system in various environmental variables in the future.

키워드