Diverse data-driven approaches for automated 3D indoor scene reconstruction: a comprehensive analysis

With the increasing prevalence of indoor activities and the growing demand for refined spatial information, 3D indoor scene reconstruction has become a core supporting technology for geospatial services, digital twins, and smart cities. This paper presents a comprehensive survey of data-driven approaches for automated 3D indoor modeling. We first systematically summarize the major data modalities, including floor plans, point clouds, RGB-D images, images, and textual descriptions, along with representative open datasets and key spatial data standards. We then provide a structured review of the full technical pipeline, covering permanent building component identification, scene reconstruction, spatial topology construction, and model texture optimization, as well as state-of-the-art learning-based and generative AI methods. Finally, we conduct in-depth comparisons of accuracy, efficiency, and automation across different sensing and modeling paradigms, and discuss the extended values of these techniques for geoscience applications such as indoor navigation, emergency response, building monitoring, and digital twin management. This review aims to clarify the technical evolution, current challenges, and future trends of data-driven indoor 3D reconstruction, providing a systematic reference for researchers in geoinformatics, computer vision, and digital construction.