Automatic Extraction of Geometric Models from 3D Point Cloud Datasets

We present in this work a methodology for fitting 3D primitive geometric models to a non-organized point cloud datasets. First, a mathematical model for the plane, sphere, cylinder and cone primitives are introduced, then we use the Random Sample Consensus algorithm to detect and extract one or several of those primitives. As an additional step, the scene reconstruction using only the primitive models and constructive solid geometry can be generated. The proposed models can be used for both, to reduce the space required in their representation, from thousand of 3D points to a single equation, and to obtain the 3D reconstruction from single and composited geometric objects. Furthermore, the models can be used for render them in different graphic software tools like CAD, OpenGL, or Povray.