Large-scale Indoor 3D Reconstruction from 360° Images

Problem

Reconstruct large-scale indoor environments for virtual tours using 360° imagery while staying robust to sparse viewpoints and imperfect capture conditions.

Method

The system combines panoramic image alignment, multi-view geometry, learned depth estimation, and depth fusion to recover consistent scene structure across wide indoor spaces.

My Role

I designed the reconstruction pipeline, with a focus on camera pose refinement, depth fusion, and practical system behavior on larger datasets.

Focus

• Robustness under sparse views
• Stable alignment across panoramas
• Scalable processing for larger scenes