Description
An algorithm for connecting a set of points in 3D with a triangulated surface, which also works well for sparsely sampled point clouds, has been presented in this poster and documented in this thesis.Densely sampled regions of a point cloud can be reconstructed fast because they map to a unambiguous local triangulation. This can be achieved by constructing and matching umbrellas (closed triangle fans) locally. Based on a condition related to the above paper, such an umbrella can be quickly tested whether it fulfills the criterion of matching its neighbors, in order to produce a unique manifold triangulated surface, which is required for nice rendering. As most points in noise-free point sets are densely sampled, points can be processed in parallel on the GPU, to permit real-time rendering as a surface, as on-demand meshing in the view frustum. This partitioning makes also out-of-core visualization of huge point sets possible.
Tasks
- Construct umbrellas at each point with successive nearest neighbor points
- Optimize selection of the umbrella fulfilling the matching condition
- Parallel implementation of this point processing on the GPU
- Stitch mesh boundaries from successive frames in the view frustum
