A hybrid intelligent system for 3D reconstruction from a single line drawing

In automatic reconstruction of 3D objects from single line drawings, existing systems are all single-track, containing one general solution for all drawings. This paper proposes a method in which an input drawing is first classified based on dominant features which exist in the drawing, including symmetry, orthogonality and parallelism. The reconstruction is then performed by experts to deal with each class specifically. Drawing classification is done using the technique of support vector machine classification. A specific set of features are selected to form an optimal regularity set for each class, and used in the formulation of the objective function for effective reconstruction. Experimental results show that the proposed system can improve the reconstruction accuracy and efficiency than that of a single-track general 3D reconstruction system.