ForAlign is a Python-based algorithm designed for automated registration of terrestrial laser scanning (TLS) and drone laser scanning (DLS) point clouds in forest environments. The algorithm leverages structural features for robust alignment and operates within a Jupyter Notebook environment, primarily utilizing the Open3D library.
For compatibility, please use Python 3.8 - 3.11.
ForAlign consists of two key registration stages:
- Coarse Alignment (matching_optimization.ipynb): Identifies and matches trees across TLS and DLS point clouds using structural graph-based constraints.
- Fine Alignment (finetune_optimization.ipynb): Refines the registration by optimizing the transformation through a voxel-based point-to-mesh approach.
Each stage is implemented in separate files, allowing users to execute the full pipeline or run specific alignment steps independently.
Each notebook includes detailed documentation and comments to guide users through the process.
Note: When using Coarse Alignment, ensure that the Digital Terrain Model (DTM) is separated, with TLS and DLS each containing ground and off-ground point clouds (e.g., using the CSF filter). The point clouds must be in PCD format.
Due to the diverse nature of forest point cloud data, we recommend adjusting parameter settings based on the specific characteristics of your dataset.
For detailed guidance, please refer to our published paper and the comments provided in the code files.
Before running ForAlign, ensure that you have the necessary dependencies installed.
If you encounter compatibility issues, refer to the tested versions listed below.
Install the required libraries using the following command:
pip install open3d==0.18.0 numpy==1.24.3 matplotlib==3.7.1 scipy==1.15.1 \
scikit-learn==1.3.0 hdbscan==0.8.33 networkx==3.1 pulp==2.8.0 rasterio==1.3.9 pandas==2.0.3ForAlign has been tested in the following environment:
- Python Version:
3.11.11(Anaconda) - Operating System:
Windows 64-bit (MSC v.1929 AMD64)
| Library | Version |
|---|---|
| Open3D | 0.18.0 |
| NumPy | 1.24.3 |
| SciPy | 1.15.1 |
| Scikit-learn | 1.3.0 |
| HDBSCAN | 0.8.33 |
| PuLP | 2.8.0 |
| Rasterio | 1.3.9 |
| Pandas | 2.0.3 |
If you experience any issues, consider using the tested versions above.
To complement publicly available forest point cloud datasets, we provide two real-world TLS and DLS datasets collected in Japanese plantation forests, along with three synthetic datasets.
Each dataset contains TLS and DLS point clouds of the same forest scene.
The synthetic datasets were generated using:
- Arbaro for simulating 3D tree models.
- Blender for constructing 3D forest environments.
- Helios++ for LiDAR simulation.
To access the datasets, please fill out the following survey form.