On Reconstructing 3D Feature Boundaries

Raw point data collected by 3D scanning techniques are usually rendered by building an interpolating robust polygonal mesh. This approach is accurate and fast but provides no means for large scale subsequent modifications. Only local interactive or non-interactive tools are provided that are usually targeted to correcting small imperfections and eliminating noise effects. CAD applications require robust and editable CAD models to support processes such as reproduction, design modification and redesign. In this paper we present a curve approximation method used in a feature-based approach to building feature-based CAD models from 3D point clouds. This approach is based on discovering features on the point cloud by detecting local changes in the morphology of the point cloud. This results in a number of regions that represent object features. The boundaries of the features are approximated by a collection of piecewise cubic rational Bezier curves that best fit the detected border point cloud and are G1 continuous.