EVENT DOSSIER
The researchers proposed a hierarchical structure-from-motion framework called HiSfM, aimed at addressing the failure issues caused by ambiguity in traditional methods. This method first uses geometric-induced heuristic algorithms to form strong local communities, and then constructs compact and robust skeletons that connect these communities by打包 non-intersecting edge-generated trees. These skeletons serve as anchors for the nature of the scene, used to reconstruct stable frameworks. On this basis, the system absorbs remaining image information through efficient registration and triangulation steps to further refine the results. Experiments show that HiSfM performs well on benchmark tests and general datasets involving ambiguity, significantly reducing running time compared to previous methods and effectively improving the integrity of the reconstructed results.
提出名为 HiSfM 的层级粗到精细结构 - 运动框架，通过构建支架提高鲁棒性和效率。该方法首先利用几何诱导启发式形成强局部社区，随后用紧凑且强健的骨架连接这些社区，该骨架由打包边不相交生成树（EDST）构成并经由两视图消歧器验证。在此基础上重建稳定支架作为场景本质的锚点，再通过高效配准和三角化吸收剩余图像以进行进一步细化。实验表明，HiSfM 在针对模糊性的基准测试和通用数据集上，有效防止了由模糊性导致的失败，相比先前方法大幅减少了运行时间，并提高了完整性。