Level of Detail (LOD) Classification Reference

Comprehensive guide to LOD2 and LOD3 building classification systems for architectural analysis and 3D modeling.

---

🎯 Overview

Level of Detail (LOD) classification enables detailed building analysis from LiDAR point clouds. This library implements:

• LOD2: Building-focused taxonomy (15 classes) - Roof structures and major elements
• LOD3: Extended architectural taxonomy (30 classes) - Detailed facade elements
• ASPRS Integration: Automatic mapping from ASPRS classes
• ML-Ready: Optimized for deep learning applications

---

📐 LOD Levels Explained

LOD Hierarchy

Level	Description	Complexity	Use Cases
LOD0	2D footprint	Low	Urban planning, GIS
LOD1	Extruded blocks	Low	City-scale visualization
LOD2	Roof structures	Medium	Solar analysis, this library
LOD3	Facade details	High	Heritage documentation, this library
LOD4	Interior spaces	Very High	Indoor navigation

---

🏗️ LOD2 Classification (15 Classes)

Overview

LOD2 focuses on building structures with simplified architectural elements:

• Target: Roof shapes, major building components
• Classes: 15 classes (0-14)
• Applications: Solar potential, urban modeling, roof type analysis
• Point Density: Optimized for standard LiDAR (10-20 pts/m²)

Class Taxonomy

Structural Elements

Class	Name	Code	Description
0	wall	0	Wall surfaces (vertical)

Roof Types

Class	Name	Code	Description
1	roof_flat	1	Flat roofs (< 5° slope)
2	roof_gable	2	Gable/pitched roofs
3	roof_hip	3	Hip roofs (4+ slopes)

Roof Details

Class	Name	Code	Description
4	chimney	4	Chimney structures
5	dormer	5	Dormer windows

Facade Elements

Class	Name	Code	Description
6	balcony	6	Balconies, terraces
7	overhang	7	Roof overhangs, eaves

Foundation

Class	Name	Code	Description
8	foundation	8	Building base, foundation

Context (Non-Building)

Class	Name	Code	Description
9	ground	9	Ground surface
10	vegetation_low	10	Low vegetation (< 2m)
11	vegetation_high	11	High vegetation (trees)
12	water	12	Water surfaces
13	vehicle	13	Vehicles, mobile objects
14	other	14	Unclassified/other

Configuration

# config.yaml (V5) - LOD2 Mode
classification:
  mode: lod2 # Enable LOD2 classification

  # LOD2-specific parameters
  lod2:
    roof_slope_threshold: 5.0 # degrees (flat vs pitched)
    wall_verticality: 0.85 # Vertical surface threshold
    chimney_height_min: 1.0 # meters
    dormer_protrusion: 0.5 # meters
    balcony_protrusion: 0.8 # meters

  # ASPRS → LOD2 mapping
  asprs_to_lod2:
    enabled: true
    auto_refine_buildings: true # Refine Class 6 → wall/roof

Python API

from ign_lidar.classes import LOD2_CLASSES, ASPRS_TO_LOD2
import numpy as np

# Convert ASPRS classes to LOD2
asprs_classes = las.classification
lod2_classes = np.array([ASPRS_TO_LOD2.get(c, 14) for c in asprs_classes])

# Get class statistics
for class_id, class_name in LOD2_CLASSES.items():
    count = np.sum(lod2_classes == class_id)
    if count > 0:
        print(f"{class_name:20s}: {count:8d} points")

---

🎨 LOD3 Classification (30 Classes)

Overview

LOD3 provides detailed architectural classification including facade elements:

• Target: Windows, doors, architectural details
• Classes: 30 classes (0-29)
• Applications: Heritage documentation, detailed 3D modeling, facade analysis
• Point Density: Requires high-density LiDAR (20-50 pts/m²)

Class Taxonomy

Structural Elements (Walls)

Class	Name	Code	Description
0	wall_plain	0	Plain wall surface
1	wall_with_windows	1	Wall with window openings
2	wall_with_door	2	Wall with door openings

Roof Types (Detailed)

Class	Name	Code	Description
3	roof_flat	3	Flat roof (< 5°)
4	roof_gable	4	Gable/pitched roof
5	roof_hip	5	Hip roof (4 slopes)
6	roof_mansard	6	Mansard roof (French style)
7	roof_gambrel	7	Gambrel roof (barn style)

Roof Details

Class	Name	Code	Description
8	chimney	8	Chimney structures
9	dormer_gable	9	Gable dormer windows
10	dormer_shed	10	Shed dormer windows
11	skylight	11	Roof skylights
12	roof_edge	12	Roof edges, gutters

Openings

Class	Name	Code	Description
13	window	13	Window openings
14	door	14	Door openings
15	garage_door	15	Garage door openings

Facade Elements

Class	Name	Code	Description
16	balcony	16	Balconies, terraces
17	balustrade	17	Railings, balustrades
18	overhang	18	Roof overhangs
19	pillar	19	Pillars, columns
20	cornice	20	Cornices, moldings

Foundation

Class	Name	Code	Description
21	foundation	21	Building foundation
22	basement_window	22	Basement windows

Context (Non-Building)

Class	Name	Code	Description
23	ground	23	Ground surface
24	vegetation_low	24	Low vegetation
25	vegetation_high	25	Trees, tall vegetation
26	water	26	Water surfaces
27	vehicle	27	Vehicles
28	street_furniture	28	Street furniture
29	other	29	Unclassified

Configuration

# config.yaml (V5) - LOD3 Mode
classification:
  mode: lod3 # Enable LOD3 classification

  # LOD3-specific parameters
  lod3:
    # Roof classification
    roof_slope_threshold: 5.0
    mansard_slope_break: 30.0 # degrees
    gambrel_slope_break: 45.0

    # Wall classification
    wall_verticality: 0.85
    window_depth: 0.15 # meters (recess)
    door_height_min: 1.8 # meters

    # Facade elements
    balcony_protrusion: 0.8 # meters
    pillar_radius: 0.3 # meters
    cornice_height: 0.2 # meters

    # Detail detection
    min_feature_size: 0.1 # meters
    point_density_required: 20 # pts/m²

Python API

from ign_lidar.classes import LOD3_CLASSES, ASPRS_TO_LOD3

# Convert ASPRS to LOD3
lod3_classes = np.array([ASPRS_TO_LOD3.get(c, 29) for c in asprs_classes])

# Refine building points (Class 6) with geometric analysis
from ign_lidar.preprocessing.classification_refinement import refine_lod3_buildings

lod3_refined = refine_lod3_buildings(
    points=points,
    initial_classes=lod3_classes,
    point_density=25.0,  # pts/m²
    detect_windows=True,
    detect_doors=True,
    detect_balconies=True
)

---

🔄 Class Mapping Tables

ASPRS → LOD2 Mapping

ASPRS_TO_LOD2 = {
    0: 14,   # Never classified → other
    1: 14,   # Unclassified → other
    2: 9,    # Ground → ground
    3: 10,   # Low Vegetation → vegetation_low
    4: 10,   # Medium Vegetation → vegetation_low
    5: 11,   # High Vegetation → vegetation_high
    6: 0,    # Building → wall (requires refinement)
    7: 10,   # Low Point (noise) → vegetation_low
    9: 12,   # Water → water
    10: 14,  # Rail → other
    11: 14,  # Road Surface → other
    17: 13,  # Bridge Deck → vehicle
    18: 11,  # High Noise → vegetation_high
    67: 14,  # Unknown → other
}

ASPRS → LOD3 Mapping

ASPRS_TO_LOD3 = {
    0: 29,   # Never classified → other
    1: 29,   # Unclassified → other
    2: 23,   # Ground → ground
    3: 24,   # Low Vegetation → vegetation_low
    4: 24,   # Medium Vegetation → vegetation_low
    5: 25,   # High Vegetation → vegetation_high
    6: 0,    # Building → wall_plain (requires refinement)
    7: 24,   # Low Point (noise) → vegetation_low
    9: 26,   # Water → water
    10: 29,  # Rail → other
    11: 23,  # Road Surface → ground
    17: 27,  # Bridge Deck → vehicle
    18: 25,  # High Noise → vegetation_high
    67: 29,  # Unknown → other
}

LOD2 → LOD3 Mapping

LOD2_TO_LOD3 = {
    0: 0,    # wall → wall_plain
    1: 3,    # roof_flat → roof_flat
    2: 4,    # roof_gable → roof_gable
    3: 5,    # roof_hip → roof_hip
    4: 8,    # chimney → chimney
    5: 9,    # dormer → dormer_gable
    6: 16,   # balcony → balcony
    7: 18,   # overhang → overhang
    8: 21,   # foundation → foundation
    9: 23,   # ground → ground
    10: 24,  # vegetation_low → vegetation_low
    11: 25,  # vegetation_high → vegetation_high
    12: 26,  # water → water
    13: 27,  # vehicle → vehicle
    14: 29,  # other → other
}

---

🎯 Classification Workflow

LOD2 Workflow

LOD3 Workflow

---

📊 Use Cases

LOD2 Applications

1. Solar Potential Analysis

   • Roof type identification
   • Slope calculation
   • Area estimation

2. Urban Morphology

   • Building type classification
   • Roof shape distribution
   • Urban pattern analysis

3. 3D City Models

   • CityGML LOD2 generation
   • Simplified visualization
   • Performance-optimized models

LOD3 Applications

1. Heritage Documentation

   • Detailed facade capture
   • Architectural element inventory
   • Historical building analysis

2. Facade Analysis

   • Window-to-wall ratio
   • Architectural style detection
   • Ornament classification

3. Detailed 3D Modeling

   • CityGML LOD3 generation
   • Textured building models
   • Virtual reality applications

---

⚙️ Processing Modes

LOD2 Processing

# Process tiles in LOD2 mode
ign-lidar-hd process \
  --config-name lod2_classification \
  input_dir=data/tiles/ \
  output_dir=output/lod2/ \
  classification.mode=lod2

LOD3 Processing

# Process tiles in LOD3 mode (requires high-density data)
ign-lidar-hd process \
  --config-name lod3_classification \
  input_dir=data/tiles_hd/ \
  output_dir=output/lod3/ \
  classification.mode=lod3 \
  classification.lod3.min_point_density=20

Adaptive Mode

# Automatically choose LOD2 or LOD3 based on point density
classification:
  mode: adaptive

  adaptive:
    lod3_density_threshold: 20.0 # pts/m²
    lod2_density_threshold: 10.0
    fallback_to_asprs: true # Use ASPRS if density < 10

---

💻 Advanced Python API

LOD Classification with Refinement

from ign_lidar.preprocessing.classification_refinement import (
    refine_lod2_buildings,
    refine_lod3_buildings
)

# Load point cloud
las = laspy.read("tile.laz")
points = np.column_stack([las.x, las.y, las.z])
asprs_classes = las.classification

# Initial conversion
lod2_classes = np.array([ASPRS_TO_LOD2.get(c, 14) for c in asprs_classes])

# Refine with geometric analysis
lod2_refined = refine_lod2_buildings(
    points=points,
    classes=lod2_classes,
    compute_normals=True,
    detect_roof_types=True,
    detect_chimneys=True,
    detect_dormers=True
)

# Further refine to LOD3 if density sufficient
point_density = estimate_point_density(points)
if point_density > 20.0:
    lod3_refined = refine_lod3_buildings(
        points=points,
        classes=lod2_refined,
        detect_windows=True,
        detect_doors=True,
        detect_balconies=True,
        detect_cornices=True
    )

Export Class Statistics

from ign_lidar.classes import LOD3_CLASSES

def print_lod3_statistics(classes):
    """Print detailed LOD3 class statistics."""
    print("\n=== LOD3 Classification Report ===\n")

    # Group by category
    categories = {
        'Walls': [0, 1, 2],
        'Roofs': [3, 4, 5, 6, 7],
        'Roof Details': [8, 9, 10, 11, 12],
        'Openings': [13, 14, 15],
        'Facade': [16, 17, 18, 19, 20],
        'Foundation': [21, 22],
        'Context': [23, 24, 25, 26, 27, 28, 29]
    }

    for category, class_ids in categories.items():
        print(f"\n{category}:")
        for class_id in class_ids:
            if class_id in LOD3_CLASSES.values():
                class_name = [k for k, v in LOD3_CLASSES.items() if v == class_id][0]
                count = np.sum(classes == class_id)
                if count > 0:
                    pct = 100 * count / len(classes)
                    print(f"  {class_name:20s}: {count:8d} pts ({pct:5.2f}%)")

print_lod3_statistics(lod3_refined)

---

📚 See Also

• ASPRS Classification Reference
• Ground Truth Classification
• Classification Workflow
• Architectural Styles

---

💡 Best Practices

LOD2

1. Check point density - Requires at least 10 pts/m²
2. Compute normals accurately - Essential for roof type detection
3. Validate roof slopes - Verify against architectural standards
4. Use building masks - Focus refinement on building points

LOD3

1. Verify high density - Requires 20+ pts/m²
2. Enable all refinements - Window, door, balcony detection
3. Tune thresholds - Adjust for regional architectural styles
4. Post-process results - Manual verification recommended
5. Consider computational cost - LOD3 is 5-10x slower than LOD2

General

1. Start with LOD2 - Simpler, faster, more robust
2. Use adaptive mode - Automatically choose based on data quality
3. Validate outputs - Check with QGIS or CloudCompare
4. Document parameters - Keep track of thresholds used
5. Test on samples - Validate before batch processing

---

Standards: CityGML LOD2/LOD3, ASPRS LAS 1.4
Updated: October 17, 2025 - V5 Configuration