mesh2plan

3D Mesh → Floor Plan Research

Comprehensive research into converting 3D mesh data into 2D floor plans. From interactive Three.js viewers (v1-v9) to advanced Python analysis pipelines (v10-v16).

Interactive Viewers (v1-v9)

Basic Mesh + Planes

Load OBJ mesh with MTL textures, display with plane overlays from JSON. Basic visualization with opacity controls.

Three.js • OBJ/MTL ⭐⭐⭐⭐⭐⭐

Enhanced Visualization

Improved UI, better plane rendering, enhanced controls. More polished visualization with better performance.

Three.js • WebGL ⭐⭐⭐⭐⭐⭐⭐

Advanced Plane Analysis

Enhanced plane detection, improved boundary rendering, better classification of wall/floor/ceiling elements.

Three.js • Geometry ⭐⭐⭐⭐⭐⭐⭐

Performance Optimized

Optimized rendering pipeline, improved memory usage, better handling of large meshes and complex geometries.

Three.js • Optimization ⭐⭐⭐⭐⭐⭐⭐

Interactive Floor Plans

2D floor plan generation, interactive editing, measurement tools. Bridge between 3D mesh and 2D plans.

Three.js • Canvas2D ⭐⭐⭐⭐⭐⭐⭐

Wall Detection

Advanced wall detection algorithms, opening identification, architectural element classification.

Three.js • ML ⭐⭐⭐⭐⭐⭐⭐

Room Segmentation

Automatic room detection, space classification, connectivity analysis. Multi-room floor plan generation.

Three.js • Clustering ⭐⭐⭐⭐⭐⭐⭐⭐

Export & Integration

SVG/PDF export, CAD integration, measurement accuracy improvements. Professional output formats.

Three.js • SVG ⭐⭐⭐⭐⭐⭐⭐⭐

Real-Time Browser

Real-time processing in browser, drag-drop mesh loading, side-by-side 3D/2D views, complete pipeline.

Three.js • WebWorkers ⭐⭐⭐⭐⭐⭐⭐⭐

Research Approaches (v10-v16)

v10

Voxelization + 2D Projection

Convert mesh to voxel grid, project to 2D, use morphological operations for wall cleanup. Robust and consistent results.

Python • OpenCV • Voxels ⭐⭐⭐⭐⭐⭐⭐

v11

Normal-based Wall Segmentation

Classify faces by normal direction, extract walls, use clustering and convex hull. Excellent performance on complex meshes.

Python • NumPy • DBSCAN ⭐⭐⭐⭐⭐⭐⭐⭐

v12

Contour Detection on Depth Maps

Render to height map, use edge detection and contours for walls. Good opening detection with computer vision.

Python • OpenCV • Canny ⭐⭐⭐⭐⭐⭐

v13

Alpha Shape Boundary Extraction

Extract floor points, use alpha shapes for non-convex boundaries. Sophisticated geometric approach.

Python • Alpha Shapes • Geometry ⭐⭐⭐⭐⭐⭐⭐

v14

Hybrid (v11 Walls + v12 Openings)

Combines v11's wall segmentation with v12's opening detection. Best of both worlds approach.

Python • Hybrid • Combined ⭐⭐⭐⭐⭐⭐⭐

v15

Flood-fill Room Segmentation

Voxelize and flood-fill from center to find connected rooms. Excellent room segmentation capabilities.

Python • Flood-fill • Voxels ⭐⭐⭐⭐⭐⭐⭐

v16

Confidence-Map Guided Extraction

Uses ARKit confidence maps and camera motion quality for guided extraction. Advanced AR-based approach.

Python • ARKit • Confidence ⭐⭐⭐⭐⭐⭐

L-Shape Research (v17-v21)

v17

Wall Intersection + Density Grid

First L-shape attempt. Vertex density grid confirms room footprint, builds rectilinear polygon from wall positions. Still keeps interior wall. 11.0m²

Python • Density Grid • L-Shape ⭐⭐⭐⭐⭐⭐

v18

Density Contour (Morphological)

Pure CV approach: density grid → morphological ops → contour tracing → rectilinear simplification. No wall histogram needed. Jagged output (14 walls). 11.3m²

Python • OpenCV • Morphology ⭐⭐⭐⭐⭐⭐

v19 ⭐

Smart Boundary Selection

Best result. Structural wall filtering: outermost 2 per axis, reject ALL interior X-walls. Partial Z-walls detect L-shape steps. Clean 6-vertex polygon. 10.3m² (target: 10.4m²)

Python • Structural Filtering • ⭐ Best ⭐⭐⭐⭐⭐⭐⭐⭐

v20

Hough Line Transform

Canny edge detection → HoughLinesP → cluster → boundary selection. Edges clearly show L-shape but clustering missed the step wall. 9.7m²

Python • OpenCV • Hough ⭐⭐⭐⭐⭐⭐⭐

v21

Raycasting from Centroid

Density mask → 720 rays from centroid → find room boundary → simplify to rectilinear. Room mask correct but polygon simplification broken. 3.0m²

Python • Raycasting • Experimental ⭐⭐⭐⭐⭐

Improved Variants (v20b-v21b)

v20b

Hough Improved

Lowered Hough thresholds + v19's structural wall selection. Now detects L-shape step wall. 5 walls, 3 doors, 1 window. 9.5m²

Python • Hough + Structural ⭐⭐⭐⭐⭐⭐⭐

v21b ⭐

Contour Extraction (Improved Raycast)

Threshold=1 + morphological close + fill holes → contour extraction → D-P simplification. Clean 6-vertex L-shape. 10.4m² — EXACT MATCH!

Python • Contour • ⭐ Exact Match ⭐⭐⭐⭐⭐⭐⭐⭐⭐

v22 ⭐

Hybrid (Mask + Hough Walls)

Combines v21b mask for shape + v20b Hough for square wall positions. Best of both worlds. 10.7m²

Python • Hybrid • Square Walls ⭐⭐⭐⭐⭐

Advanced Research (v23-v26)

v23

Signed Distance Field (SDF)

KDTree-based SDF from projected vertices. Multi-threshold contour extraction for optimal area. Captures room shape but overestimates due to vertex spread. 12.5m²

Python • KDTree • SDF • Contour ⭐⭐⭐⭐⭐⭐

v24

Occupancy Grid + Graph Cut

Density-based occupancy grid with iterative morphological boundary refinement using rectilinear structural elements. Undershoots area. 5.7m²

Python • Morphology • ICM ⭐⭐⭐⭐⭐

v25

Structured Edge Detection

Fuses Sobel, Laplacian, and morphological gradient edges. NMS + Hough wall extraction + structural polygon. Clean 6-vertex L-shape. 11.8m²

Python • Multi-Edge • NMS • Hough ⭐⭐⭐⭐⭐⭐⭐

v26

Poisson-style Surface Slice

Delaunay triangulation + alpha filtering for concave hull boundary. Multi-alpha search for optimal room outline. 11.2m²

Python • Delaunay • Alpha Shape ⭐⭐⭐⭐⭐⭐

Multiroom Research (v27+)

v27

MULTIROOM

Multiroom Floor Plan Detection

First multiroom approach. Room mask + interior wall classification + connected component splitting. Per-room polygon extraction with wall snapping. Door detection at wall gaps.

Edge + Hough + Mask Split → 16 rooms detected

v27e

NEW

Density-Based Room Detection

Walls = high density, rooms = low density gaps. Threshold + directional morphological closing + connected components. No Hough lines needed.

Density Gaps → 5 rooms (multiroom), 1 room (single)

v27f

NEW

Watershed Room Segmentation

OpenCV watershed on density image. Wall ridges as barriers, distance-transform room centers as seeds. Handles irregular room shapes naturally.

Watershed → 6 rooms (multiroom), 2 rooms (single)

v27g

NEW

Watershed + Distance Transform + Hallway Detection

Distance transform seeds + watershed + skeleton-based hallway carving. Correctly identifies 3 rooms + 1 hallway. Polygon extraction with wall snapping.

Watershed+Skeleton → 3R+1H (multiroom), 1R (single)

v28

MULTIROOM

Wall Grid Cell Classification

Grid cells between Hough walls, classify by interior occupancy. Problem: too many walls → tiny cells, too few → everything merges.

Wall Grid → 1 room (too coarse grid)

v28b

MULTIROOM

Edge-Based Wall Detection

Density ratio between adjacent cells detects walls. Union-find merging. Better separation but rooms have gaps between them.

Edge Graph → 4R+2H (NMS=0.3)

v29

⭐ BEST MULTIROOM

Mask Cut — Wall Scoring

Fill apartment mask, cut along strongest validated walls. Wall scoring = strength × max_run. Boundary exclusion, min separation, fragment merging.

Mask Cut → 4R+1H (12.3, 9.8, 5.3, 4.4m² + 6.1m² hall)

v30

MULTIROOM

Architectural Floor Plan

v29 detection + clean architectural rendering. Thick wall rectangles, door arcs, scale bar, pastel room fills, room classification.

Architectural → 4R+1H, 37.9m² total

v32

⭐⭐ BEST MULTIROOM

Strip Merge

X cuts → vertical strips (left/center/right). Center strip = hallway. Z cuts split strips into rooms. Door detection via mask adjacency. Matches target layout!

Strip merge → 3R+1H+1C, 34.7m² total