Defining Point Cloud Boundaries Using Pseudopotential Scalar Field Implicit Surfaces

Identifying smooth and meaningful object boundaries of noisy 3D point-clouds presents a challenge. Rather than rely on the points of the cloud itself, we identify a smooth implicit surface to represent the boundary of the cloud. By constructing a scalar field using a semantically-informative pseudopotential function, we take an arbitrary-resolution iso-surface and apply standard computer vision morphological transformations and edge detection on 2D slices of the pseudopotential field. When recombined, these slices comprise a new point-cloud representing the 3D boundary of the object as determined by the chosen isosurface. Our method leverages the strength and accessibility of 2D vision tools to identify smooth and semantically significant boundaries of ill-defined 3D objects, and additionally provides a continuous scalar field containing insight regarding the internal structure of the object. Our method enables a powerful and easily implementable pipeline for 3D boundary identification, particularly in domains where natural candidates for pseudopotential functions are already present.