3D Gaussian Splatting aided Localization for Large and Complex Indoor-Environments

Recent breakthroughs in deep learning, including 3D Gaussian Splatting (3DGS) (Kerbl et al., 2024), have significantly advanced both the performance and visual quality of the reconstruction. Within our work, we focus on 3D mapping of complex, large-scale indoor environments such as construction sites and factory halls. This initiative is driven by a project within the Cluster of Excellence Integrative Computational Design and Construction for Architecture (IntCDC) at the University of Stuttgart, which aims to enable autonomous indoor construction for new or preexisting buildings (IntCDC, 2024a). Typical construction tasks, including material handling and element assembly, require highly accurate mapping approaches to enable precise localization of both building components and the construction robots. Image-based localization methods are particularly valuable due to the widespread availability and low cost of cameras, which are now standard equipment on most modern robots.