A Framework for 3D Modeling of Construction Sites Using Aerial Imagery and Semantic NeRFs

The rapid evolution of drone technology has revolutionized data acquisition in the construction industry, offering a cost-effective and efficient method to monitor and map engineering structures. However, a significant challenge remains in transforming the drone-collected data into semantically meaningful 3D models. 3D reconstruction techniques usually lead to raw point clouds that are typically unstructured and lack the semantic and geometric information of objects needed for civil engineering tools. Our solution applies semantic segmentation algorithms to the data produced by NeRF (Neural Radiance Fields), effectively transforming drone-captured 3D volumetric representations into semantically rich 3D models. This approach offers a cost-effective and automated way to digitalize physical objects of construction sites into semantically annotated digital counterparts facilitating the development of digital twins or XR applications in the construction sector.