Classification Taxonomy

Overview

This library uses a building-focused classification schema that differs from the standard ASPRS classification. Point cloud classifications are automatically remapped from ASPRS codes to LOD2 or LOD3 taxonomies designed for detailed architectural analysis.

Important

When you see buildings classified as Class 0, this is CORRECT! The LOD2 schema maps buildings to class 0 (wall) as the base category for detailed building analysis.

LOD2 Classification Schema (15 Classes)

LOD2 is designed for building structure analysis with basic architectural elements.

Building Elements

Class ID	Class Name	Description	ASPRS Source
0	wall	Building walls and main structure	6 (Building)
1	roof_flat	Flat roof surfaces	—
2	roof_gable	Gable/pitched roof	—
3	roof_hip	Hip roof	—
4	chimney	Chimney structures	—
5	dormer	Dormer windows	—
6	balcony	Balconies and terraces	—
7	overhang	Overhanging elements	—
8	foundation	Building foundation	—

Context Elements (Non-Building)

Class ID	Class Name	Description	ASPRS Source
9	ground	Ground surface, roads	2 (Ground), 11 (Road)
10	vegetation_low	Low vegetation, shrubs	3, 4 (Low/Med Veg)
11	vegetation_high	Trees, high vegetation	5 (High Vegetation)
12	water	Water bodies	9 (Water)
13	vehicle	Vehicles, mobile objects	17 (Bridge Deck)
14	other	Unclassified objects	0, 1, 8, 10-16, 18

LOD3 Classification Schema (30 Classes)

LOD3 provides detailed architectural classification including facade elements.

Structural Elements

Class ID	Class Name	Description
0	wall_plain	Plain wall without openings
1	wall_with_windows	Wall with window openings
2	wall_with_door	Wall with door openings
8	chimney	Chimney structures
19	pillar	Structural pillars
20	cornice	Decorative cornices
21	foundation	Building foundation

Roof Elements

Class ID	Class Name	Description
3	roof_flat	Flat roof surfaces
4	roof_gable	Gable/pitched roof
5	roof_hip	Hip roof
6	roof_mansard	Mansard roof
7	roof_gambrel	Gambrel roof
9	dormer_gable	Gable dormer
10	dormer_shed	Shed dormer
11	skylight	Roof skylights
12	roof_edge	Roof edges/gutters

Openings

Class ID	Class Name	Description
13	window	Standard windows
14	door	Entry doors
15	garage_door	Garage doors
22	basement_window	Basement windows

Facade Elements

Class ID	Class Name	Description
16	balcony	Balconies
17	balustrade	Balustrades/railings
18	overhang	Overhanging elements

Context Elements

Class ID	Class Name	Description
23	ground	Ground surface
24	vegetation_low	Low vegetation
25	vegetation_high	Trees
26	water	Water bodies
27	vehicle	Vehicles
28	street_furniture	Urban furniture
29	other	Other/unclassified

ASPRS to LOD Mapping

How It Works

When processing LiDAR files, the pipeline automatically remaps ASPRS standard classifications to LOD schemas:

# Example: ASPRS class 6 (Building) → LOD2 class 0 (wall)
original_classification = 6  # ASPRS Building
lod2_classification = 0      # LOD2 Wall

# Example: ASPRS class 2 (Ground) → LOD2 class 9 (ground)
original_classification = 2  # ASPRS Ground
lod2_classification = 9      # LOD2 Ground
```python
# Output shows LOD2 classifications

ASPRS to LOD2 Mapping Table

ASPRS Code	ASPRS Name	LOD2 Class	LOD2 Name
0	Never classified	14	other
1	Unclassified	14	other
2	Ground	9	ground
3	Low Vegetation	10	vegetation_low
4	Medium Vegetation	10	vegetation_low
5	High Vegetation	11	vegetation_high
6	Building	0	wall
7	Low Point (noise)	10	vegetation_low
8	Model Key-point	14	other
9	Water	12	water
10	Rail	14	other
11	Road Surface	14	other
17	Bridge Deck	13	vehicle
18	High Noise	11	vegetation_high

ASPRS to LOD3 Mapping Table

ASPRS Code	ASPRS Name	LOD3 Class	LOD3 Name
0	Never classified	29	other
1	Unclassified	29	other
2	Ground	23	ground
3	Low Vegetation	24	vegetation_low
4	Medium Vegetation	24	vegetation_low
5	High Vegetation	25	vegetation_high
6	Building	0	wall_plain
7	Low Point (noise)	24	vegetation_low
8	Model Key-point	29	other
9	Water	26	water
10	Rail	29	other
11	Road Surface	23	ground
17	Bridge Deck	27	vehicle
18	High Noise	25	vegetation_high

Understanding Your Classifications

Example Output Analysis

When you inspect an enriched LAZ file, you might see:

Classification distribution in the file:
============================================================
Class   0:  2,644,762 points ( 12.30%) → WALLS (building structure)
Class   9:  9,024,921 points ( 41.96%) → GROUND (terrain)
Class  10:    292,956 points (  1.36%) → VEGETATION LOW (shrubs)
Class  11:  7,953,872 points ( 36.98%) → VEGETATION HIGH (trees)
Class  12:        674 points (  0.00%) → WATER
Class  14:  1,591,015 points (  7.40%) → OTHER (unclassified)

This is the expected behavior! Buildings are classified as Class 0 (wall) in the LOD2 schema.

Why This Approach?

The LOD-focused taxonomy provides several advantages:

1. Building-Centric Analysis: Optimized for architectural feature detection
2. Hierarchical Structure: Buildings are subdivided into structural components
3. ML Training: Better class balance for machine learning models
4. Semantic Richness: More meaningful categories for urban analysis

Checking Classifications

Using Python

import laspy
import numpy as np
from ign_lidar.classes import LOD2_CLASSES

# Load LAZ file
las = laspy.read("your_file.laz")

# Get classification distribution
unique, counts = np.unique(las.classification, return_counts=True)

# Print with LOD2 names
class_names = {v: k for k, v in LOD2_CLASSES.items()}
for cls, count in zip(unique, counts):
    name = class_names.get(cls, "unknown")
    percentage = (count / len(las.classification)) * 100
    print(f"Class {cls:3d} ({name:20s}): {count:10,} ({percentage:6.2f}%)")

Using CloudCompare

1. Open the enriched LAZ file in CloudCompare
2. Select the point cloud in the DB Tree
3. Go to Edit > Scalar Fields > Classification
4. The classification values correspond to LOD2/LOD3 schema, not ASPRS

Configuration Options

Disabling Remapping (Use ASPRS Classes)

If you need to preserve original ASPRS classifications, modify the configuration:

processor:
  lod_level: null # Disable LOD remapping
  preserve_asprs_classes: true

Custom Class Mapping

You can define custom mappings in ign_lidar/classes.py:

# Custom mapping example
CUSTOM_MAPPING = {
    6: 100,  # Map buildings to class 100 instead of 0
    2: 101,  # Map ground to class 101 instead of 9
    # ... etc
}

Implementation Details

The remapping occurs in ign_lidar/core/processor.py:

# Remap ASPRS classifications to LOD schema
labels_v = np.array([
    self.class_mapping.get(c, self.default_class)
    for c in classification_v
], dtype=np.uint8)

The mapping dictionaries are defined in ign_lidar/classes.py:

• LOD2_CLASSES: LOD2 taxonomy (15 classes)
• LOD3_CLASSES: LOD3 taxonomy (30 classes)
• ASPRS_TO_LOD2: ASPRS → LOD2 mapping
• ASPRS_TO_LOD3: ASPRS → LOD3 mapping

Related Documentation

• LOD3 Classification - Level of Detail concepts
• Enriched LAZ Output - Understanding output files
• API Reference - Python API for class definitions

References

• ASPRS LAS Specification 1.4: Standard LiDAR classification codes
• CityGML LOD Specification: Level of Detail standards for 3D city models
• OGC 3D Portrayal Service: 3D geospatial visualization standards