notesum.ai

Published at December 6

GS-Matching: Reconsidering Feature Matching task in Point Cloud Registration

cs.CV

Released Date: December 6, 2024

Authors: Yaojie Zhang¹, Tianlun Huang¹, Weijun Wang¹, Wei Feng²

Aff.: ¹Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences; ²Guangdong Provincial Key Laboratory of Construction Robotics and Intelligent Construction

Arxiv: http://arxiv.org/pdf/2412.04855v1

	3DMatch FPFH		3DMatch FCGF		3DLoMatch FPFH		3DLoMatch FCGF
		RE(deg)		RE(deg)		RE(deg)		RE(deg)	Time
	RR(%)	/TE(cm)	RR(%)	/TE(cm)	RR(%)	/TE(cm)	RR(%)	/TE(cm)	(s)
Deep Learned
3DRegNet [30]	26.31	3.75/9.60	77.76	2.74/8.13	-	-/-	-	-/-	0.07
DGR [10]	32.84	2.45/7.53	88.85	2.28/7.02	19.88	5.07/13.53	43.80	4.17/10.82	1.53
DHVR [26]	67.10	2.78/7.84	91.93	2.25/7.08	-	-/-	54.41	4.14/12.56	3.92
PointDSC [2]	77.39	2.05/6.43	92.85	2.05/6.50	27.74	4.11/10.45	55.36	3.79/10.37	0.10
Traditional
SM [27]	55.88	2.94/8.15	86.57	2.29/7.07	6.06	6.19/12.62	33.52	4.28/11.01	0.03
RANSAC [15]	65.29	3.52/10.98	89.62	2.50/7.55	15.34	6.05/13.74	46.38	5.00/13.11	0.97
GC-RANSAC [4]	71.97	2.43/7.03	89.53	2.25/6.93	17.46	4.43/10.75	41.83	3.90/10.44	0.55
TEASER [40]	75.79	2.43/7.24	87.62	2.38/7.44	25.88	4.83/11.71	42.22	4.65/12.07	0.07
FGR [49]	40.91	4.96/10.25	78.93	2.90/8.41	-	-/-	19.99	5.28/12.98	0.89
SC2 [9]	83.26	2.09/6.66	93.16	2.09/6.51	38.46	4.04/10.32	58.62	3.79/10.37	0.11
MAC [46]	83.92	2.11/6.79	93.72	2.03/6.53	41.27	4.06/10.64	60.08	3.75/10.60	1.87
SC2-Plus [chen2023sc]	87.18	2.10/6.64	94.15	2.04/6.50	41.27	3.86/10.06	61.15	3.72/10.56	0.28
SVC (w.o GS) [47]	88.66	2.18/6.87	94.58	2.07/6.60	45.76	4.04/10.62	67.77	3.93/10.83	0.25
Ours	90.82	2.19/6.96	95.01	1.99/6.31	52.39	3.99/10.35	71.70	3.71/10.37	0.20