一种基于图割的稠密三维场景重建算法

立体视觉技术是自主探测机器人在未知环境中获取信息的重要手段，通过对可视场景的稠密三维重建实现导航、定位及路径规划等一系列工作。本文在候选点匹配的基础上结合图割理论，首先在世界坐标系建立代表深度信息的网格节点，接着依据区域匹配算法对候选点进行初步筛选，去除大部分相关值较低的节点，建立简化的网格图，最后通过寻找图中最小割来实现能量函数的全局最小，完成稠密的三维场景重建。实验证明，相关阈值γ设为0.6时，简化网格图的重建精度和计算效率达到相对平衡。图割算法解决了候选点测量时潜在的匹配歧义问题，且对低纹理区域有较好的匹配效果。

A 3D Scene Graph Cut-Based Dense Reconstruction Algorithm

Stereo vision plays a very important role in environment perception of the autonomous exploration robot,by which the dense reconstruction result of the 3D scene can be derived.In this paper,the candidate matching algorithm combined with graph cut theory is proposed to solve the stereo correspondence problem.First,the mesh grid representing depth is constructed in the world coordinate system.Second,most of the candidates can be pruned according to an area matching algorithm,and the remaining nodes are used t...