Abstract: Through using image information fusing silhouette and stereo, a new robust energy model for multiview 3D reconstruction 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 requirements and the effectiveness of this method with multiview 3D reconstructions at higher resolutions.

Key words: multiview 3D reconstruction, energy model, global optimization, convex problem, successive overrelaxation

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。

关键词: 多视三维重构, 能量模型, 全局优化, 凸问题, 超松弛法