Triangulating Point-Set Surfaces

by
Carlos Eduardo Scheidegger

Advised by
Claudio Silva

We introduce a meshing algorithm for Point-Set Surfaces that uses no intermediate representation and produces an adaptive triangulated mesh directly. Our algorithm naturally handles noisy input and lack of normal information without the need for an explicit filtering phase. The algorithm produces a high-quality triangulation, where triangle size respects the geometry of the object rather than the sampling of the object. Our technique is based on using the moving least squares (MLS) paradigm as the underlying surface representation and a novel advancing front algorithm as the triangulation scheme. The advancing front algorithm uses an adaptive surface prediction scheme so that triangle size changes gracefully even when there are large changes in surface curvature throughout the point set. Our results show that our technique produces high-quality triangulations for models that other techniques fail due to irregular sampling, noise, or underlying geometric features.