A volumetric approach for 3D surface reconstruction

In the last years, measuring sensors like the TOF or optical-based terrestrial laser scanners have been more and more increasingly used, given their capability to acquire the 3D geometry of the surveyed object. Applications for these systems span among different fields, such as industry, medicine, land management, heritage and VR environments. Regardless long-range or close-range laser scanner was employed, two following steps need to be performed in order to reconstruct the object shape: range image registration and integration. The former allow for “tailoring” together the acquired point clouds, representing a “view” of the object surface as sampled by the laser sensor. Then, after all views have been aligned each other, a unique representation of the object surface is generated through the integration of those 3D views. At this stage several factors prevent from building the descriptive surface by simple connection of the range images: non-uniform density sampling, measuring and registration errors. To solve for such modeling issues, a volumetric approach has been devised for the generation of a mesh, i.e to create a surface representation from range data acquired by an optical triangulation laser scanner. The developed method is based on the “consensus surface” concept introduced by Wheeler, Sato and Ikeuchi, by which some kind of errors of the range images can be better identified and corrected. Then it has been refined by integration with the so-called “Marching cubes” algorithm, a well used surface generation procedure in the field of Computer Graphics. Finally, the proposed method has been completed with the development of a tool for hole-filling , though its application is limited to little holes with enough regular edges. Pros and cons along with the results of our meshing algorithm, applied to a little statue, will be presented as well.