View-graph construction framework for robust and efficient structure-from-motion
作者：
Highlights：
• We use an incremental manner to construct view graph, rather than computing the relative geometry in each edge independently.
• By iteratively performing local reconstruction, the verified maximum spanning tree is extended into our view graph.
• The relative geometry is estimated by local reconstruction, which has less ambiguity and is more robust than matrix decomposition.
• By embedding our view graph, both the reconstruction efficiency and robustness are greatly improved.
• Our view graph has a wide applicability, and in principle it can be inserted into any SfM systems.
摘要
•We use an incremental manner to construct view graph, rather than computing the relative geometry in each edge independently.•By iteratively performing local reconstruction, the verified maximum spanning tree is extended into our view graph.•The relative geometry is estimated by local reconstruction, which has less ambiguity and is more robust than matrix decomposition.•By embedding our view graph, both the reconstruction efficiency and robustness are greatly improved.•Our view graph has a wide applicability, and in principle it can be inserted into any SfM systems.
论文关键词：Structure-from-motion,View-graph construction,Epipolar geometry computation
论文评审过程：Received 12 December 2019, Revised 14 October 2020, Accepted 21 October 2020, Available online 21 October 2020, Version of Record 2 March 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2020.107712
