3D Architectural Modeling: Efficient RANSAC for n-gonal Primitive Fitting

Ahsan Abdullah, Reema Bajwa, Syed Rizwan Gilani, Zuha Agha, Saeed Boor Boor, Murtaza Taj, Sohaib Ahmed Khan

Abstract: We present a modeling approach to automatically fit 3D primitives to point clouds in order to generate a CAD like model. For detailed modeling we propose a new n-gonal 3D primitive and a novel RANSAC based fitting approach. Non-planar surfaces are modeled through surface of revolution with B-spline profiles. We first reduce the dimension by projecting the 3D data onto a 2D plane. Primitive fitting algorithm is then applied in this 2D space. Our approach compares favorably both with manually and automatically generated models. Not only is it much more time efficient than manual modeling, but it also gives significantly better output than state-of-the-art automatic methods. Since the focal technique of our approach is the fitting of detailed primitives, our results are ideal in the domain of architecture and preservation of heritage.