From drawing geometry to structured floorplan data.
CAD-to-Floorplan Parsing
A DXF parsing pipeline that infers walls, windows, doors, and room regions from geometry and topology instead of relying on unreliable CAD layer names.
Problem
CAD drawings are optimized for humans, not parsers. Walls may be parallel lines, doors may be blocks, furniture pollutes the drawing, layer names are inconsistent, and units are often missing or wrong.
Approach
- Normalize units and crop the relevant scene before expensive geometry processing.
- Use constrained triangulation to recover a robust outer boundary even when drawings contain small gaps.
- Build a region adjacency graph and infer walls, windows, doors, and rooms from geometry, thickness, and neighborhood rules.
My Contribution
I independently designed and implemented the full CAD-to-floorplan parsing pipeline from scratch.
On the algorithm side, I built the core modules:
- DXF/DWG format parsing with deep-learning-based door block classification
- Constrained Delaunay triangulation to recover robust outer boundaries from incomplete drawings
- Region adjacency graph construction with geometry-based rule inference for walls, windows, doors, and rooms
- Wall thickness estimation and door position refinement
Outcome
The parser converts noisy CAD linework into usable structural floorplan JSON without manual layer cleanup as a required preprocessing step.
