CAD2OSM Project Guide

This repository converts CAD floor plans into osmAG maps: standard OSM XML with indoor room geometry, passage topology, and optional semantic room names.

Citation

@misc{zhang2025generationindooropenstreet,
      title={Generation of Indoor Open Street Maps for Robot Navigation from CAD Files}, 
      author={Jiajie Zhang and Shenrui Wu and Xu Ma and Sören Schwertfeger},
      year={2025},
      eprint={2507.00552},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2507.00552}, 
}

Paper: Generation of Indoor Open Street Maps for Robot Navigation from CAD Files

Data Availability and Reproducibility

This repository provides the source code and configuration files for the CAD-to-osmAG pipeline. The exact CAD floor plans used in the experiments are not redistributed for the following reasons:

1. The ShanghaiTech campus floor plans are non-public institutional building data and cannot be released without authorization.
2. The ArchWeb floor plans were obtained from ArchWeb and remain subject to ArchWeb's copyright and current Terms and Conditions, which do not permit us to redistribute or republish the downloaded CAD files. Researchers interested in these plans should obtain them independently from ArchWeb and comply with its current terms of use.

To support reproducibility without embedding protected data, this repository includes processing code, configuration templates, and placeholder input paths. Users should provide only CAD files they are authorized to use from a local path outside the repository. Original protected CAD files and their derived artifacts must not be committed to this repository.

Reproducible Quick Start

The protected experimental CAD files are intentionally not tracked. To run the pipeline, copy the tracked configuration template to a Git-ignored local configuration and replace its placeholder with the path to a DXF file you are authorized to use. Keep that source file outside the repository.

# Create a local configuration, then edit its `cases.example.dxf` value
cp config/repro_local_cad.example.yaml config/repro_local_cad.yaml
# Replace /path/to/authorized/input.dxf in config/repro_local_cad.yaml

# Build the AreaGraph executable
cmake -S area_graph_segment -B area_graph_segment/build
cmake --build area_graph_segment/build --target area_graph_segmentation -j

# Preview the configured local case without running the pipeline
python3 run_pipeline.py --config config/repro_local_cad.yaml --case example --dry-run

# Run one case end to end: DXF -> SVG/bounds -> PNG -> osmAG
python3 run_pipeline.py --config config/repro_local_cad.yaml --case example

Outputs are written to runs/local-cad/<case>/:

Output	Purpose
effective_config.yaml	The fully merged parameters actually used for this run
run_manifest.json	Input files, generated outputs, commands, and run metadata
preprocess/*.svg, *.bounds.json, *.png	CAD preprocessing artifacts
area_graph/*_osmAG.osm	Main osmAG result
area_graph/*_roomGraph.png	Visual segmentation result

The optional text naming stage is disabled by default:

python3 run_pipeline.py --config config/repro_local_cad.yaml --case example --with-text

Pipeline Stages

DXF -> SVG + bounds.json -> PNG -> AreaGraph segmentation -> osmAG.osm -> optional text naming

DWG conversion is still available through cad2osm/script/core_process/dwg2dxf_oda.py, but it requires ODA File Converter and is not part of the default reproducibility path.

Parameter Map

Configuration priority is:

built-in defaults < config global defaults < case profile < case overrides < CLI overrides

Key parameters live in the selected YAML configuration and are copied into each run's effective_config.yaml.

Parameter	Unit	Used By	CLI Override
map_preprocessing.resolution	m/pixel	PNG scale, AreaGraph thresholds, OSM export, area statistics	--resolution
map_preprocessing.door_width / corridor_width	m	Dynamic alpha calculation	--set map_preprocessing.door_width=...
root_node.latitude/longitude/pixel_x/pixel_y	deg / px	OSM coordinate conversion	--root-lat, --root-lon, --root-pixel-x, --root-pixel-y
polygon_processing.simplify.tolerance	px	OSM polygon simplification	--set polygon_processing.simplify.tolerance=...
polygon_processing.small_room_filter.min_area	m²	Room filtering	--min-room-area
area_graph.alpha.*	px / mode	Room split granularity	--set area_graph.alpha.fixed_value=...
area_graph.vori_config.*	px or m as named	Voronoi/topology cleanup	--set area_graph.vori_config.<key>=...

area_graph_segment/config/params.yaml remains the direct executable default. The unified runner writes an effective config and passes it to area_graph_segmentation --config.

Main Components

Module	Purpose	Entry
run_pipeline.py	Reproducible DXF-to-osmAG orchestration	python3 run_pipeline.py --config <config> --case <case>
cad2osm/script/core_process	CAD preprocessing utilities	imported by run_pipeline.py or used directly
area_graph_segment	AreaGraph segmentation and osmAG export	area_graph_segment/build/bin/area_graph_segmentation
cad2osm/script/text_extract_module	Optional DXF text extraction and room naming	text_extractor.py

Dependencies

pip install ezdxf svgwrite svgpathtools cairosvg pillow numpy opencv-python pyproj pyyaml
sudo apt-get install g++ cmake qtbase5-dev libcgal-dev

For deeper module-level details, see area_graph_segment/README.md, cad2osm/README-zh.md, and cad2osm/script/text_extract_module/README.md.