Intelligent system for feature-based recognition of machining parts from points cloud

Feature recognition is a complex process consists of several steps. This process mostly requires the extraction of geometric features like normal vectors and curvatures, as well as segmentation of points cloud. There are two fundamental approaches can be used to calculate normal vectors and curvatures: analytical and numerical. The first approach is practically more complex. Nevertheless, the second approach has a drawback particularly in detecting sharp edges and transitions between different geometric objects. In this paper, a CAD intelligent system is proposed which is developed by enhancing the local Delaunay triangulation method. The new approach can automatically detect and eliminate noises in the points cloud and, thus, the direct neighboring points can be determined. Therefore, triangulations at sharp corners and edges, as well as transition between different surface regions can be optimized globally.