Optimal Transport for Mesh Repair

Text

In photogrammetry-based 3D-Scanning, a multitude of photographs of an object are taken and three-dimensional structures are estimated via Structure from Motion. These 3D-Data can then be triangulized to create 3D-Meshes. However, the results usually contain artefacts or some parts may be missing. Repairing a mesh is a cumbersome process when done by hand and requires speciality software when done automatically. This, for example includes the smoothing of surfaces, outlier removal, and closing of holes. Optimal transport has enjoyed a variety of applications in recent years. In this study, we want to concentrate on 3D-point clouds as data-source. Wasserstein barycenters offer a way of transforming features from one 3D point cloud to another. We test whether this can be used to repair meshes or parts thereof as they are generated in photogrammetry-based 3D-scanning processes. Automatically completing the scan with parts of the same object which in turn can be duplicated and pasted would make mesh repair a simpler process. To make this user-friendly special attention should be paid to runtime of the algorithm and ease of use. Related approaches have already shown that Wasserstein barycenters can be used to transform features, this can be seen in Figure 1. We adapt this algorithm to be tested for the repair of partial meshes resulting from 3D Scans.

Titel der Veranstaltung

24th Scientific-Professional Colloquium on Geometry and Graphics

Veranstalter*in