基于 Grassmann 流形上二次分配的鲁棒仿射点匹配
计算机视觉与模式识别
2024-10-10 v5
摘要
基于 Grassmann 流形的鲁棒仿射匹配（RoAM）是一种用于点云仿射配准的新算法。该算法基于最小化 Grassmann 流形上两个元素之间的 Frobenius 距离。为此，使用了二次分配问题（QAP）的一种不定松弛，并研究并比较了多种仿射特征匹配方法。实验表明，RoAM 比先前的方法对噪声和点不一致具有更强的鲁棒性。
引用
@article{arxiv.2303.02698,
title = {Robust affine point matching via quadratic assignment on Grassmannians},
author = {Alexander Kolpakov and Michael Werman},
journal= {arXiv preprint arXiv:2303.02698},
year = {2024}
}
备注
9 pages, 23 figures ; GitHub repository at https://github.com/sashakolpakov/roam ; added comparison to GrassGraph (https://doi.org/10.1109/TIP.2019.2959722); notably, GrassGraph quickly loses accuracy on our test examples with noise and occlusion ; to appear in Pattern Recognition Letters