GitHub repository
Automatically classifies and labels urban point clouds using data fusion with public datasets and region growing techniques.
What is GitHub repository?
Urban PointCloud Processing is a Python-based toolkit for automatically classifying and labeling urban point clouds, such as those from LiDAR scans. It solves the problem of scarce labeled 3D training data by fusing point clouds with public geospatial datasets (e.g., elevation and topography registries) to generate initial labels for objects like buildings, trees, and street furniture.
Geospatial data scientists, urban planners, and researchers working with 3D point cloud data in urban environments, particularly those focused on automating asset detection and classification without extensive manual labeling.
Developers choose this project because it provides a practical, data-fusion-driven alternative to machine learning methods that require large labeled datasets, offering out-of-the-box tools for urban point cloud labeling with public data integration and region-growing post-processing.
Overview
Repository for automatic classification and labeling of Urban PointClouds using data fusion and region growing techniques.
Use Cases
Best For
- Automatically labeling street furniture (e.g., traffic signs, benches) in urban point clouds
- Generating training datasets for 3D object detection models using fused public data
- Extending building footprints in point clouds to include architectural details like balconies
- Classifying ground, roads, and vegetation in LiDAR scans for urban mapping
- Processing point clouds in the Netherlands using Rijksdriehoek coordinate system and AHN/BGT data
- Visualizing labeled point clouds in CloudCompare with custom color scales
Not Ideal For
- Projects outside the Netherlands that lack access to AHN and BGT public datasets
- Teams needing real-time or high-throughput point cloud processing due to computational fusion steps
- Developers preferring web-based or Python-native visualization over CloudCompare dependency
- Applications requiring object classes beyond standard urban assets like vehicles or pedestrians
Pros & Cons
Pros
Integrates AHN elevation and BGT topographic data to automatically label point clouds, eliminating the need for extensive manual annotation or ML training sets.
Labels a wide range of urban objects including ground, roads, buildings, trees, and street furniture like traffic signs and benches, as shown in the demo GIF.
Extends building facades to include protruding elements such as balconies, improving accuracy beyond basic footprint data from public registries.
Provides Jupyter notebook tutorials, like the 'Complete solution' notebook, that walk through entire workflows, making it easier to adopt.
Cons
Heavily reliant on Dutch-specific data sources (AHN, BGT) and the Rijksdriehoek coordinate system, requiring significant adaptation for international use.
Requires manual building of cccorelib and pycc from the CloudCompare-PythonPlugin, which depends on Qt and isn't available on PyPi, adding setup overhead.
Assumes point clouds are in LAS format and tiled following specific rules, limiting flexibility with other formats or untiled data.
Frequently Asked Questions
Related Projects
GitHub repositoryFrustum PointNets for 3D Object Detection from RGB-D Data
GitHub repository🔥RandLA-Net in Tensorflow (CVPR 2020, Oral & IEEE TPAMI 2021)
GitHub repositoryOfficial PyTorch implementation of Superpoint Transformer [ICCV'23], SuperCluster [3DV'24 Oral], and EZ-SP [ICRA'26]
GitHub repositoryLarge-scale Point Cloud Semantic Segmentation with Superpoint Graphs
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.