CoFiNet论文 英文视频讲解文稿
需积分: 5 23 浏览量
2023-10-30
21:18:30
上传
评论
收藏 5.08MB DOC 举报
given the groundtruth correspondence set C bar consisting of all the groundtruth
correspondences, the recovered rigid transformation should minimize the distances between
every corresponding points
we follow the path of correspondence based methods like fcgf D3feat and predator
where we first establish correspondences between a pair of point clouds from the overlept
regions and then estimate poses with unextracted correspondences, we mainly focus on the
correspondence retrieval more specifically
given a pair of partially overlapped point clouds，named the PX and PY
on the first phase we need to learn a function F， a condition of both point clouds to establish
correspondences, here in this case, we extract four correspondences from the point cloud pair
among which three are in liars shown in green lines
we use C tilda to represent the extracted correspondence set, then we use a function G to
estimate the relative pose based on the correspondences
like svd or ransac to get the estimated rotation r tilda and translation t tilda
资源评论
