Leaf Segmentation on Dense Plant Point Clouds with Facet Region Growing

Area of Science:

• Botany
• Computer Vision
• Computational Geometry

Background:

• Leaves are crucial for plant photosynthesis and growth monitoring.
• Existing image-based leaf segmentation struggles with dense canopies and species limitations.
• Accurate individual leaf segmentation is vital for plant phenotyping.

Purpose of the Study:

• To develop an effective and efficient individual leaf segmentation method for dense plant point clouds.
• To overcome limitations of current species-specific and occlusion-sensitive segmentation techniques.
• To enable precise measurement of leaf traits and plant growth status.

Main Methods:

• A novel approach combining facet over-segmentation and facet region growing for point cloud data.
• Three-step process: point cloud pre-processing, facet over-segmentation, and facet region growing.
• Utilizes 3D point cloud data for detailed leaf surface analysis.

Main Results:

• Achieved over 90% average precision and recall in segmenting individual leaves.
• Demonstrated effectiveness on diverse ornamental plants like Epipremnum aureum, Monstera deliciosa, and Calathea makoyana.
• Validated applicability across different 3D imaging systems (stereo vision, Kinect v2).

Conclusions:

• The proposed method offers a robust solution for individual leaf segmentation from 3D point clouds.
• The technique shows broad applicability for regular surface and object segmentation tasks.
• This advancement supports improved plant phenotyping and agricultural monitoring.