Curvature-based pebble segmentation as a foundation for 3D roundness and orientation analysis

Quantifying pebble size, shape, and roundness is fundamental to
understanding sediment transport and abrasion in fluvial systems, yet
remains challenging in natural, densely packed settings.  Most existing
approaches rely on 2D imagery and therefore fail to capture true
three-dimensional morphology. Here, we present a curvature-based instance
segmentation framework for reconstructed surface meshes and demonstrate its
role as a key step enabling 3D roundness and orientation analysis.

In our approach, individual pebbles are detected directly from 3D surface
reconstructions using curvature features, without prior shape assumptions.
Validation against high-resolution reference models yields a high detection
precision of 0.98, with remaining errors mainly due to under-segmentation
in overly smooth reconstructions.  Estimates of 3D pebble orientation are
strongly controlled by the represented surface area, highlighting both the
potential and current limitations of orientation retrieval from incomplete
surface segments.

We illustrate how reliable segmentation allow downstream 3D shape and
roundness analyses that are not accessible in 2D, including curvature-based
surface metrics and volumetric descriptors. Example fluvial scenes
demonstrate that segmentation quality directly controls the stability of
roundness estimates and their geomorphic interpretation. Our results
establish curvature-based 3D pebble segmentation as a methodological
foundation for reproducible analyses of pebble shape, roundness, and
orientation in natural river systems.