Scanning Buildings for Hidden Structural Risks: Designing the Facade Analyzer

I grew up in Istanbul, where building safety is not an abstract concern. Before you move into an apartment, you think about earthquake resistance. You think about cracks, structural stress, and whether the exterior gives away deeper problems. That's the reality that shaped the Facade Risk Analyzer.

The original idea was ambitious: use COLMAP to generate a 3D reconstruction of a building and overlay a heatmap of potential damages. A full photogrammetry pipeline that you could run with only a phone camera. But COLMAP on macOS was far more unreliable than expected. After multiple attempts, I realized the project needed to pivot if it was ever going to deliver something usable.

The pivot led to the current design: a multi-stage image pipeline that detects facade features, identifies risks, scores them, and generates a structured PDF report. The hardest part wasn't the scoring logic. It was rethinking the entire system architecture after abandoning COLMAP. Changing the foundation forced me to approach the project the way real engineers do: adapt the design, salvage what matters, and rebuild the rest.

What feels uniquely mine about this project is the decision to analyze exterior damage. Most people focus on interior inspection: walls, floors, furniture. But in a seismic city, the facade can be the first and strongest indicator of underlying structural risk. It's the part you see every day yet rarely scrutinize.

I was surprised by how accurate modern vision models are, especially GPT-4o. They picked up patterns that even I missed, and they did it consistently. For a prototype, that level of reliability exceeded my expectations.

If I revisit this project, I'll return to the original dream: a full 3D reconstruction with a damage heatmap generated by COLMAP or a similar pipeline. The current version is strong, but the long-term vision is still the same: using accessible tools to give people an early warning about the safety of the place they might call home.