Texture- and geometry-based approaches for the classification of 3D heritage

Abstract

The continuous evolution in the last years of remote sensing technologies and methodologies for Cultural Heritage 3D documentation allowed to multiply photogrammetric and laserscanning acquisitions. At the same time, to exploit the real potential of this significant amount of data, the need for reliable and efficient methods to classify (i.e. semantically segment) point clouds or meshes has become a priority. This article explores the use of Machine and Deep Learning methods as support for studies, monitoring, and restoration purposes. More specifically, three different approaches based on texture, geometry, and texture plus geometry features are presented and compared.