An active surface reconstruction method based on growing neural gas

Abstract

This study proposes a novel active-surface-reconstruction method called ASGNG, which is an artificial neural network derived from the growing neural gas (GNG) algorithm. ASGNG can reconstruct triangle meshes with various resolutions from unorganized point cloud data of the original surface. Compared with similar algorithms such as Growing Self-Reconstruction Maps (GSRM) and Surface-reconstructing Growing Neural Gas (SGNG), ASGNG designs a new edge-penalty mechanism by introducing the concept of active edges and proposes an active-surface-creation mechanism. This mechanism enables ASGNG to reconstruct high-quality triangle mesh surfaces with almost no holes, minimizing the need for postprocessing. Experimental results demonstrate that ASGNG can efficiently process unorganized point cloud data with intricate topology. The topology and shape of the triangle meshes reconstructed by ASGNG closely resemble the original surface compared with some existing methods, with lower distance error and better mesh quality.