BranchMatch: point cloud registration for individual apple trees with limited overlap based on local structure...

Area of Science:

• Agricultural Engineering
• Computer Vision
• 3D Imaging

Background:

• Accurate 3D point cloud reconstruction is vital for fruit tree management.
• Existing methods often require high overlap or multiple viewpoints, increasing complexity and cost.

Purpose of the Study:

• To develop an efficient and accurate point cloud registration method for individual apple trees.
• To enable precise 3D reconstruction using minimal data acquisition.

Main Methods:

• Proposed BranchMatch method for low-overlap viewpoint acquisition and registration (180° apart).
• Utilizes key branch segments, spatial/geometric features, and a dynamically weighted discriminant function.
• Employs an optimized iterative closest point algorithm for refinement and prevention of over-registration.

Main Results:

• Achieved a 90% registration success rate on individual apple trees with two low-overlap point clouds.
• Demonstrated low average rotation (1.93 mrad) and translation (4.33 mm) errors, with a pointwise error of 2.70 mm.
• Significantly reduced computational costs compared to multi-site high-overlap methods while maintaining accuracy.

Conclusions:

• BranchMatch offers an efficient and reliable solution for individual tree registration.
• The method improves 3D reconstruction accuracy and completeness for fruit tree management.
• Suitable for tall-spindle individual apple trees during dormancy with minimal data acquisition.