Quality preserving fusion of 3D triangle meshes

Surface representations in the shape of 3D triangle meshes of heterogeneous quality are readily generated using sensors such as laser scanners. We propose an algorithm to fuse two registered, i.e. aligned in rotation and translation, meshes while using only vertices contained in the input data. Mesh quality is determined for each input mesh depending on the assumed sensor model. In the first step, redundant faces are removed using local comparison of mesh quality. Finally, the meshes are fused by inserting new faces, using the remaining vertices, resulting in a mesh with locally optimal quality regarding both input datasets. Since no new vertices are introduced, further processing can rely on the original measurements and any associated meta data such as sensor specific information. Results are shown for fusing overlapping parts of datasets representing a ballroom and the Stanford Bunny.