多视重建的研究现状

：对目前的多视重建技术进行了综述．先介绍了多视重建的分类和评价方法，然后分析了光照一致性约 束和可见性约束，再分别讨论了几个有代表性的算法，如深度图方法、空间雕刻法、变形模型方法、目标函数求解 法（包括水平集法、图割法）、种子生长法、概率估算法（马尔可夫场方法、EM 方法）等等，再探讨了这些算法的不 足之处．最后，提出了融合多种方法、多种线索的3 维重建新思路，以提高多视重建的完整性和精度．

A Review of Today's Multi-view Reconstruction

This paper reviews today’s algorithms about multi-view reconstruction. Firstly, the classification and evaluation methods for multi-view reconstruction are introduced. Secondly, some constraints, such as photo-consistency and visibility, are analyzed. Then, some typical algorithms are discussed, such as depth map, space carving, deformable model, objective function optimization (including the level set method, graph cuts method), seed-growing method, and probability estimation method (including Markov-field method, EM method). Some shortcomings in all of these algorithms are discussed. A new 3D reconstruction idea integrating some of aforementioned reconstruction methods and reconstruction cues is proposed to improve the integrity and accuracy of multi-view reconstruction.