Floor-Plan-Based Visual Localization
Overview
The field estimates where a camera is inside a building by aligning images, reconstructed geometry, or visual-inertial motion with a floor plan. Approaches include matching learned image-to-plan features, aligning geometric primitives such as walls and wireframes, generating and refining pose hypotheses with diffusion models, and using plan geometry to correct drift in visual-inertial SLAM. Benchmarks such as S3D, ZInD, Gibson, Hilti-Trimble, and SlugTrails test single-frame and sequential localization, including sparse observations and real-world trajectories. A central challenge is generalizing to unseen buildings and imperfect plans when structural cues are scarce; limited training data and the need to handle ambiguous poses remain important constraints.
Latest papers
10 papers3 papers in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.