Recently three-dimensional (3D) graphics has been applied in areas such as building and mechanical design, with the help of CAD tools to first construct the two-dimensional (2D) and 3D models. Building synthetic environments usually involves lots of work in constructing virtual reality worlds, and making authoring more efficient is one of the most important goals in virtual reality. We propose a prototype system that integrates several modules such as image preprocessing, vectorization, editing, and 3D model generation, to transform a 2D architecture design diagram into a 3D building model. We also propose an improved method that can efficiently recognize slant lines in addition to vertical and horizontal lines. To do this, since a line has a constant slope, we use back-tracking in our method and utilize a line slope consistency check to prune wrong targets.