Plant Point Cloud Segmentation

Automated segmentation and volumetric analysis of individual plants from 3D point cloud scans (3D camera) using spatial clustering and voxel-based measurements.

Features

• Automatic Segmentation: Automatically detects and separates individual plants using X-axis distribution analysis
• Robust Cleaning: Removes background walls, pots, and outliers using color and spatial filtering
• Volumetric Analysis: Multi-resolution voxel-based volume and surface area calculation
• Visual Outputs: Generates 3D meshes (OBJ), point clouds (PLY), and diagnostic visualizations
• Batch Processing: Process multiple scans automatically
• Interactive Mode: User-friendly prompts

For plant phenotyping, growth analysis, and agricultural research.

Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Setup

1. Clone the repository:

git clone https://github.com/yourusername/Plant-Point-Cloud-Segmentation.git
cd Plant-Point-Cloud-Segmentation

2. Install dependencies:

pip install -r requirements.txt

Usage

Step 1: Segmentation

Segment a multi-plant point cloud into individual plants:

python segmentation.py

Follow the interactive prompts:

• Mode 1: Single file processing
• Mode 2: Batch processing (entire folder)

Input: Multi-plant PLY point cloud
Output:

• individual_plants/plant_01.ply through plant_0N.ply
• plants_cleaned.ply (cleaned combined point cloud)
• segmentation_histogram.png (visualization)

Plant point cloud Processing. (A) Raw point cloud from 3D-Camera; (B) Cleaned point cloud before segmentation

Step 2: Volumetric Analysis

Calculate volume and surface area for each plant:

python volumetric_analysis.py

Input: Individual plant PLY files (from Step 1)
Output (per plant, at 3 resolutions):

• plant_01_coarse_voxels_10mm.obj - 3D voxel mesh (viewable in Meshlab/Blender)
• plant_01_medium_voxels_5mm.obj - Medium resolution mesh
• plant_01_fine_voxels_3mm.obj - Fine resolution mesh
• plant_01_*_voxels_*mm.ply - Voxel centers as point clouds
• plant_measurements.csv - Summary with volume (cm³) and area (cm²)

(A) Individual plant point cloud; (B) 3D voxel of plant

Example Workflow

# 1. Segment plants from the scan
python segmentation.py
> Input PLY file: my_scan.ply
> Number of plants: 7
> Output directory: individual_plants

# 2. Analyze volumes
python volumetric_analysis.py
> Input directory: individual_plants
> Output directory: results
> Base voxel size: 5

Output Examples

Segmentation Results

individual_plants/
├── plant_01.ply (15,234 points)
├── plant_02.ply (18,901 points)
├── plant_03.ply (14,567 points)
└── ...

Volumetric Analysis Results

results/
├── plant_01_cleaned.ply
├── plant_01_coarse_voxels_10mm.obj
├── plant_01_medium_voxels_5mm.obj
├── plant_01_fine_voxels_3mm.obj
└── plant_measurements.csv

CSV Output:

Plant	Points	Coarse Vol (cm³)	Medium Vol (cm³)	Fine Vol (cm³)
plant_01	15,234	1,245.3	1,189.7	1,156.2
plant_02	18,901	1,567.8	1,498.4	1,467.9

Parameters

Segmentation Parameters

Parameter	Default	Description
num_plants	7	Expected number of plants in scan
nb_neighbors	20	Neighbors for outlier removal
std_ratio	2.0	Standard deviation threshold

Volumetric Analysis Parameters

Parameter	Default	Description
input_unit	mm	Units of input files (m/cm/mm)
voxel_size_mm	5.0	Base voxel size (coarse=2x, fine=0.5x)
dilate_iters	1	Dilation iterations to fill gaps

Requirements

• Python >= 3.8
• open3d >= 0.13.0
• numpy >= 1.19.0
• scipy >= 1.5.0
• matplotlib >= 3.3.0
• scikit-learn >= 0.24.0

See requirements.txt for a complete list.

File Format

Input Point Cloud Requirements

• Format: PLY (binary or ASCII)
• Coordinates: X, Y, Z in millimeters (configurable)
• Colors: RGB values (0-255 or 0-1)
• Structure: Multi-plant scene with pots visible

Expected Coordinate System

• X-axis: Horizontal (plant separation axis)
• Y-axis: Depth (front-to-back)
• Z-axis: Vertical (height)

Algorithms

Segmentation Method

1. Background Removal: Depth-based filtering using white pot labels
2. Pot Removal: Y-coordinate analysis with color filtering
3. Outlier Removal: Statistical outlier detection (Open3D)
4. Automatic Segmentation: Histogram-based peak/valley detection in X-axis
5. Boundary Optimization: Automatic merging/splitting to match plant count

Volumetric Method

1. Cleaning: Color filtering (green/yellow) + DBSCAN spatial clustering
2. Voxelization: Convert point cloud to 3D occupancy grid
3. Dilation: Fill small gaps using 6-neighborhood morphology
4. Volume: Count occupied voxels × voxel volume
5. Surface Area: Face counting method at voxel boundaries

Troubleshooting

Issue: "No PLY files found"

Solution: Ensure the input directory contains .ply files and the path is correct.

Issue: "Zero volume results"

Solution:

• Check input units match your data (mm/cm/m)
• Verify PLY files contain valid point data
• Try disabling color filtering if plants aren't green

Issue: "Segmentation creates wrong number of plants"

Solution:

• Adjust num_plants parameter
• Check if plants are clearly separated in the X-axis
• Verify background/pots are properly removed

Issue: "Too many/few points after cleaning"

Solution: Adjust cleaning parameters:

• nb_neighbors: Higher = more aggressive cleaning
• std_ratio: Lower = more aggressive cleaning

Citation

If you use this code in your research, please cite:

@software{plant_point_cloud_segmentation,
  title={Plant Point Cloud Segmentation},
  author={Worasit Sangjan},
  year={2024},
  url={https://github.com/WorasitSangjan/Plant-Point-Cloud-Segmentation}
}

License

This project is open source under the MIT License. See the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Contact

For questions or issues, please open an issue on GitHub or contact [worasitsangjan.ws@gmail.com]

Acknowledgments

• Built with Open3D
• Inspired by plant phenotyping research in precision agriculture