摘要

A novel method is proposed to automatically recover a 3D polyhedral object from a single 2D line drawing in perspective projection. In the algorithm, a camera calibration based on partial orthogonality is performed in order to retrieve multiple pinhole cameras. Then, 3D orientations of edge lines are estimated by a factor graph inference. Finally, 3D reconstruction is conducted through hybrid optimization by combining beam search with simulated annealing. Parameterization based on inverse depths is proposed to reduce the dimension of variables in optimization. Experimental results show that the algorithm proposed in the study can reconstruct reasonable 3D objects from perspective-projected line drawings, which cannot be handled by existing approaches. Qualitative and quantitative comparisons are provided to illustrate the advantages of the proposed method.