Geometry-aware point cloud clustering for spherical-component aggregate modeling

Area of Science:

• Computer Vision
• Geometric Modeling
• 3D Reconstruction

Background:

• Aggregate objects (e.g., grape bunches) pose challenges for traditional 3D reconstruction due to component occlusion and missing data.
• Existing methods often produce a holistic shape, failing to distinguish individual components within an aggregate.

Purpose of the Study:

• To develop a method for obtaining independent mesh models of individual components from aggregate point clouds.
• To address the limitations of current shape reconstruction techniques for clustered objects.

Main Methods:

• The proposed method assumes individual components can be approximated as spheres.
• It employs geometry-aware clustering to identify and segment components.
• Optimal component position and size are determined, with merging for overlapping detections.

Main Results:

• Successfully segmented individual components from various aggregate types.
• Demonstrated effectiveness in recovering distinct shapes from occluded and incomplete point cloud data.
• Generated independent mesh models for each component.

Conclusions:

• The spherical approximation and geometry-aware clustering provide an effective solution for modeling individual components within aggregates.
• This method advances 3D reconstruction capabilities for clustered objects.