Three-Dimensional Modeling of Weed Plants Using Low-Cost Photogrammetry

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• Accurate plant phenotyping is crucial for agricultural research and management.
• 3D modeling techniques are rapidly advancing for plant morphological characterization.
• Challenges remain in achieving precise reconstruction of plant end-details.

Purpose of the Study:

• To adapt low-cost Structure from Motion (SfM) and MultiView Stereo (MVS) techniques for 3D plant modeling.
• To reconstruct and validate 3D models of three weed species with diverse structures.
• To assess the accuracy of SfM-based 3D modeling for weed phenotyping.

Main Methods:

• Acquired sequential digital images of plants from multiple angles.
• Applied SfM algorithms for initial 3D reconstruction and dense point cloud generation.
• Utilized MultiView Stereo (MVS) to create detailed 3D polygon meshes of plant geometry.

Main Results:

• 3D models showed good consistency with ground truth measurements (plant height, leaf area, biomass).
• SfM-based 3D modeling accurately estimated plant height and leaf area.
• The image processing using SfM was found to be relatively fast.

Conclusions:

• SfM and MVS offer a valuable, budget-friendly methodology for high-detail weed phenotyping.
• The developed system demonstrates potential for plant reconstruction in various conditions, including outdoors.
• Future work should explore time-cost efficiencies and varying levels of detail in different approaches.