多视三维重构中连续能量模型的凸优化  

Convex optimization of continuous energy model in multiview 3D reconstruction

作　　者：刘光帅[1] 李柏林[1] 何朝明[1] 

机构地区：[1]西南交通大学机械学院,成都610031

出　　处：《计算机工程与应用》2013年第1期19-22,66,共5页Computer Engineering and Applications

基　　金：国家自然科学基金(No.50675180);四川省科技计划项目(No.2009GZ0007);中央高校基本科研业务费专项资金资助项目(No.SWJTU09BR104)

摘　　要：利用融合了轮廓线及体视的序列图像信息,提出了一个面向多视三维重构的稳健能量模型;为了适配于可视性约束,提出一种针对该能量模型的连续全局优化方法;为了保证栅格连通性选择的一致性及独立性,实施了全局连续优化的超松弛离散化。实例证明,该方法的实用性好,极大地减少了算法处理的内存开销,实现了在更高分辨率上有效的多视重构。Through using image information fusing silhouette and stereo, a new robust energy model for multiview 3D recon-struction is proposed. In order to match the visibility constraints, a method that globally optimized this energy model is suggested. For the sake of being consistent and independent of the choice of grid connectivity, the discretization based on successive over relaxation of a continuous optimization technique is carried out. The experimental results show the reduction of memory require-ments and the effectiveness of this method with multiview 3D reconstructions at higher resolutions.

关 键 词：多视三维重构 能量模型 全局优化 凸问题 超松弛法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]