From Digitized Data to NURB Surface Meshes

A common reverse engineering problem is to convert several hundred thousand points collected from the surface of an object via a digitizing process, into a coherent geometric model that is easily transferred to a CAD software such as a solid modeler for either design improvement or manufacturing and analysis. These data are very dense and makes data-set manipulation difficult and tedious. Many commercial solutions exist but involve time consuming interaction to go from points to surface meshes such as BSplines or NURBS (Non Uniform Rational BSplines). Our approach differs from current industry practice in that we produce a mesh with little or no interaction from the user. The user can produce bi-quadratic or bi-cubic BSpline surfaces and can choose the number of patches, tolerance etc. as parameters to the system. The BSpline surface is both compact and continuous. The former property reduces the large storage overhead, and the later eliminates the rippling effect that a scanner often produces. In addition, the nature of the BSpline allows one to easily and smoothly alter the surface, making re-engineering extremely feasible. The BSpline surface is created using the principle of higher orders least squares with smoothing functions at the edges. Both linear and cylindrical data sets are handled using a different set of parameterization schemes, a key factor in the correct solution of the resulting BSplines. Also, because of the BSpline’s continuous nature, a multiresolutional-triangulated mesh can quickly be produced. This last fact means that an STL file is simple to generate. STL files can also be easily used as input to the system.