Reproduce “BIMCaP: BIM-based AI-supported LiDAR-Camera Pose Refinement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Miguel Arturo Vega Torres, Anna Ribic, Borja García de Soto, André Borrmann
BIMCaP, a framework from the Technical University of Munich and NYUAD, refines LiDAR-camera sensor poses and improves 3D mapping accuracy by integrating sparse LiDAR data and RGB images with Building Information Models (BIMs). The approach reduced translational error by over 4 cm compared to a state-of-the-art method, enhancing 3D reconstruction in complex indoor construction environments.
Key insights
- Building Information Models (BIMs) provide essential geometric and semantic ground truth that can effectively correct accumulated drift in low-cost sensor data (LiDAR and RGB camera) within GPS-denied indoor environments.
- A hybrid approach combining linear interpolation with a deep learning model (CompletionFormer) effectively generates dense and accurate depth maps from sparse LiDAR and RGB images, which is critical for robust 3D semantic projection.
- An AI-supported bundle adjustment, incorporating custom cost functions for geometric consistency and alignment with BIM-derived semantic elements (floor, ceiling, walls, columns), can significantly refine approximated sensor poses and improve overall 3D map quality in challenging indoor scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.