Adaptive Meshing and Detail-reduction of 3d-point Clouds from Laser Scans

3D laser scanners become more and more popular especially for measuring construction sites, in the field of architecture and for preservation of monuments. As these scanners can only record discrete data sets (point clouds), it is necessary to mesh these sets for getting closed 3D models and take advantage of 3D graphics acceleration of modern graphics hardware. The meshing process is a complex issue and in the last years there were lots of algorithms developed to solve this problem. In this work a continuous Level-of-Detail (LOD) algorithm will be presented, from which a simplified 3D surface model can be created, which uses only as much triangles as needed. The grade of simplification is user-defined by an error tolerance similar to the Hausdorff distance. This algorithm is not a dynamic (view-dependent) LOD mesh simplification, but a non-redundant approximation of point clouds. It is a simple and extendable meshing algorithm, which is made up of a mixture of some techniques adapted from popular terrain LOD algorithms. The algorithm has been implemented in an OpenGL based editing tool for 3D point clouds called PointMesh, which will be explained in detail. We present examples from an excavation site (church rest of Lorsch Abbey, Germany), the inner part of the Heidelberg Tun (German: Großes Fass, an extremely large wine vat) and the Old Bridge gate, Heidelberg. For different cut-outs a performance comparison of the regular mesh to various adaptive 4-8-meshes with different error tolerances is given in terms of frame rates.