3D-Grounded Floorplan Localization Method for Large-Scale Buildings
A novel method for floorplan localization has been introduced by researchers, specifically designed for extensive buildings featuring rasterized floorplans, addressing the shortcomings of current techniques. Named SceneAligner, this approach reconstructs a gravity-aligned three-dimensional scene from a diverse collection of images and translates it into a two-dimensional density map, serving as a proxy for the floorplan. The localization process aligns this proxy with the provided floorplan using a two-dimensional similarity transformation. This innovative technique effectively bridges the disparity between density maps and floorplans, facilitating reliable localization in practical settings. The findings are detailed in a publication on arXiv under ID 2605.22581.
Key facts
- SceneAligner performs floorplan localization in the wild using 3D scene reconstruction.
- It works with rasterized floorplans and large-scale buildings.
- The method uses a gravity-aligned 3D scene projected into a 2D density map.
- Localization is formulated as aligning the density map with the floorplan via a 2D similarity transform.
- The approach handles unconstrained image collections.
- Existing methods assume controlled small-scale environments and precise vectorized floorplans.
- The research is published on arXiv with ID 2605.22581.
- The paper is categorized as a cross-type announcement.
Entities
Institutions
- arXiv