Automated floor-plan references from terrestrial laser scans: horizontal band extraction and evaluation of density-based CAD export strategies.

This preprint presents planta_auto, a modular Python pipeline for Scan-to-BIM floor-plan reference generation from terrestrial laser scans in E57 format. The workflow aligns the point cloud to the inferred ground (RANSAC), extracts a configurable horizontal height band, and exports PLY, PNG, and optional DXF outputs in metric coordinates. Four CAD export modes are implemented and compared on a shared 2D density map of the slice: (1) point underlay for manual tracing, (2) multi-level density isolines, (3) morphological mask contours, and (4) a decoupled scikit-image post-processor on the exported PNG. The tool provides CLI and GUI interfaces, automated regression checks, and a frozen v1.0.0 baseline separating stable slice export from v2.x experimental vectorization. Evaluation on a real hotel scan (P20250106; Hotel Opatov, CTU Prague; ~54 million points; ~80 m footprint) shows that the core E57 → slice → PLY pipeline operates reliably at production scale. Qualitative CAD review indicates that mask-based export yields few closed polylines (8) but poor perimeter fidelity, whereas multi-level skimage contouring produces many polylines (799), consistent with the structural limitation of multi-curve isoline methods. Quantitative benchmark metrics were outside the scope of this version. Document type: preprint (not peer-reviewed). Software: planta_auto v1.0.0 (baseline) / v2.x (experimental). License: CC BY 4.0.