Fast adaptive multimodal feature registration (FAMFR): an effective high-resolution point clouds registration workflow for cultural heritage interiors

Heritage Science

Table 7 Average registration results. Comparison with FAMFR in column Recall\(_C\) (values in brackets)

Average		Recall\(_C\) (FAMFR)[%]	\(S_{Vg}\)	\(S_{Vs}\)	RMSD	Time
	Ground truth	100 (−4.8)	0.572	0.550	0.469	–
Feature-based	SIFT + FPFH	15.9 (+79.3)	0.963	0.949	4.612	158 [s]
	SIFT + PFHRGB	12.7 (+82.5)	0.973	0.963	4.715	132 [s]
	SIFT + RIFT	9.9 (+85.3)	0.980	0.974	4.776	47 [s]
Deep learning	DCP	6.9 (+88.3)	0.987	0.981	4.813	236 [ms]
	PointNetLK	10 (+85.2)	0.979	0.973	4.737	269 [ms]
	DeepGMR	33.6 (+61.6)	0.925	0.900	3.924	137 [ms]
	GeoTransformer	56.6 (+38.6)	0.868	0.827	2.927	332 [ms]
	Predator	60.5 (+34.7)	0.850	0.810	2.810	1867 [ms]
	Proposed solution	95.2 (–)	0.598	0.615	0.767	17 [s]