基于候选点稠密匹配的三维场景重构方法

针对星球探测机器人在未知环境中三维场景重建存在的计算复杂度问题,提出基于网格候选点的三目立体匹配算法。在空间中建立代表深度信息的网格节点,并对深度方向的节点分布进行合理规划,确保候选点稠密匹配的准确性和高效性。候选点匹配解决了传统立体匹配算法中图像校正带来的实时性问题,同时采用三目视觉系统代替双目,通过另一组对应点的相似性测度对潜在歧义的少量候选点进行二次判决。实验证明,由于处理每组图像对不再需要进行极线校正,因此计算代价与传统的匹配算法相比有一定降低,而第三台摄像机有效消除了匹配歧义,计算量相对于双目系统也没有明显增加。

A 3D Scene Reconstruction Method Based on Candidate Dense Matching

Considering to the computational complexity of the 3D reconstruction for planetary robot in an unknown environment,a mesh-candidate-based trinocular stereo matching algorithm is proposed.The mesh nodes are established in the world coordinate system and axes of the nodes along the depth direction are refined to ensure the accuracy and efficiency of the candidate matching.The candidate matching solves the computational cost problem of epipolar rectification.With trinocular stereo vision system,the ambiguity is reduced by using re-judgment of the a few candidates.The experimental result proves that our method has better performance in computational efficiency than traditional methods,for the epipolar rectification is omitted.The accuracy is improved by using another image pair,and the cost is almost unchanged.