Surface meshes obtained from a 3D laser-scanner typically include high-frequency noise and distortions due to scanning inaccuracies. Special discrete variants of anisotropic mean curvature flows have proven to remedy noise artifacts while maintaining or even improving quality of the surface mesh. Industrial surface models typically have a large number of variable nodes leading to long running times. This project will investigate to accelerate the compute time by using the latest Cuda library. Cuda allows to drastically reduce the compute time by shifting computing resources from the main processor to the highly parallel graphics card. The project investigates the underlying differential geometry of mean curvature flow, its discretization and its efficient implementation on a Cuda graphics card.
The project was part of Research in Industrial Projects for Students (RIPS) 2010 program organized by the Institute for Pure and Applied Mathematics (IPAM) and Matheon.