Reconstruction of tubular structures from 2.5D point clouds: A mesophotic gorgonian coral case study

Abstract

A method for the surface reconstruction of 3D tubular branched structures characterized by low informative point clouds (i.e., 2.5D) is proposed. These specific clouds can arise when using photogrammetry techniques on complex subjects in challenging scanning environments (e.g., underwater gorgonian coral at mesophotic depths). The core idea behind the proposed Sphere Skeleton Approach (SSA) is to approximate the assumed tubular shapes via merged spheres having variable radii and centered in the points of the medial skeleton. To assess the generality and robustness of the proposed SSA, additional experiments have been conducted on 2.5D point clouds that were synthetically generated from 3D model benchmarks. Hausdorff distances between the target and the reconstructed 3D models are used to quantitatively compare the SSA performances to a classical meshing algorithm. Early results highlight the capability to outperform existing approaches in reconstructing objects from 2.5D clouds.