Three-Dimensional Point Cloud Registration Network Based on Deep Interactive Multi-Scale Receptive Field Feature Learning

Han Zhou; Xuchu Wang; Yue Yuan

doi:10.3788/AOS240529

Acta Optica Sinica, Volume. 44, Issue 14, 1415001(2024)

Three-Dimensional Point Cloud Registration Network Based on Deep Interactive Multi-Scale Receptive Field Feature Learning

Han Zhou¹, Xuchu Wang^1,2、*, and Yue Yuan¹

Author Affiliations

¹College of Optoelectronic Engineering, Chongqing University, Chongqing 400040, China

²Key Laboratory of Optoelectronic Technology & Systems (Chongqing University), Ministry of Education, Chongqing 400040, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Framework diagram of DIM-RFNet

Download full size

Fig. 2. Deep feature extraction module

Download full size

Fig. 3. Transformer module diagram

Download full size

Fig. 4. ORE structure diagram

Download full size

Fig. 5. Visualization results on ModelNet40 and ModelLoNet datasets obtained by proposed model

Download full size

Fig. 6. Visualization results on 3DMatch and 3DLoMatch datasets obtained by proposed model

Download full size

Fig. 7. Visualization results on Odometry KITTI dataset obtained by proposed model

Download full size

Table 1. Datasets used for experiments and their partition

View table

Table 1. Datasets used for experiments and their partition

Dataset	Partition
Dataset	Training set	Validation set	Testing set
ModelNet40 and ModelLoNet	5000 samples	1200 samples	1220 samples
3DMatch and 3DLoMatch	45 scenes	9 scenes	8 scenes
OdometryKITTI	Sequences 00-05	Sequences 06-08	Sequences 09-10

Table 2. Experimental parameters of the proposed method
View table
Table 2. Experimental parameters of the proposed method
Dataset L N $n_{X}$ $γ$ $Δ_{p o s}$ $Δ_{n e g}$ $r^{(c)}$ $r^{(n)}$ $r^{(o)}$ $r^{(m)}$
ModelNet40 5 3 384 64 0.1 1.4 0.018 0.060 0.040 0.040
3DMatch 5 3 256 24 0.1 1.4 0.036 0.130 0.037 0.040
OdometryKITTI 5 3 512 48 0.1 1.4 0.210 0.720 0.430 0.280

Table 3. Test results on ModelNet40 and ModelLoNet datasets obtained by compared methods

View table

Table 3. Test results on ModelNet40 and ModelLoNet datasets obtained by compared methods

Dataset	RRE of ModelNet40	RTE of ModelNet40	CD of ModelNet40	RRE of ModelLoNet	RTE of ModelLoNet	CD of ModelLoNet
ICP	21.236	0.286	0.12130	28.043	0.791	0.1997
Symmetric ICP	17.576	0.135	0.10340	24.718	0.857	0.1838
FGR （2016）	14.726	0.103	0.08520	23.416	0.703	0.1741
FMR （2020）	13.674	0.085	0.09830	22.574	0.685	0.1583
PointNetLK （2019）	12.733	0.121	0.08410	21.733	0.632	0.1342
DCP （2019）	11.795	0.171	0.01170	16.501	0.251	0.0268
OMNet （2021）	2.947	0.043	0.00150	8.947	0.133	0.0095
RPMNet （2021）	1.712	0.018	0.00850	7.342	0.124	0.0050
Predator （2021）	1.739	0.018	0.00120	5.235	0.132	0.0083
Regtr （2022）	1.473	0.014	0.00078	3.930	0.087	0.0037
DIM-RFNet	1.542	0.014	0.00077	3.975	0.085	0.0037

Table 4. Experimental results on 3DMatch and 3DLoMatch datasets obtained by compared methods

View table

Table 4. Experimental results on 3DMatch and 3DLoMatch datasets obtained by compared methods

Dataset		3DMatch					3DLoMatch
Number of samples		5000	2500	1000	500	250	5000	2500	1000	500	250
FMR /%	PerfectMatch （2019）	95.0	94.3	92.9	90.1	82.9	63.6	61.7	53.6	45.2	34.2
	FCGF （2019）	97.4	97.3	97.0	96.7	96.6	76.6	75.4	74.2	71.7	67.3
	D3Feat （2020）	95.6	95.4	94.5	94.1	93.1	67.3	66.7	67.0	66.7	66.5
	SpinNet （2021）	97.6	97.2	96.8	95.5	94.3	75.3	74.9	72.5	70.0	63.6
	Predator （2021）	96.6	96.6	96.5	96.3	96.5	78.6	77.4	76.3	75.7	75.3
	CofiNet （2021）	98.1	98.3	98.1	98.2	98.3	73.1	73.5	75.5	75.3	69.9
	Regtr （2022）	97.8	97.4	96.9	96.1	95.6	74.3	74.4	74.2	73.8	72.9
	GeoTransformer （2023）	97.9	97.9	97.9	97.9	97.6	88.3	88.6	88.8	88.6	88.3
	RoReg （2023）	98.2	97.9	98.2	97.8	97.2	82.1	82.1	81.7	81.6	80.2
	DIM-RFNet	97.7	97.6	97.6	97.6	97.5	88.4	88.6	88.9	88.6	88.4
IR /%	PerfectMatch （2019）	36.0	32.5	26.4	21.5	16.4	11.4	10.1	8.0	6.4	4.8
	FCGF （2019）	56.8	54.1	48.7	42.5	34.1	21.4	20.0	17.2	14.8	11.6
	D3Feat （2020）	39.0	38.8	40.4	41.5	41.8	13.2	13.1	14.0	14.6	15.0
	SpinNet （2021）	47.5	44.7	39.4	33.9	27.6	20.5	19.0	16.3	13.8	11.1
	Predator （2021）	58.0	58.4	57.1	54.1	49.3	26.7	28.1	28.3	27.5	25.8
	CofiNet （2021）	49.8	51.2	51.9	52.2	52.2	24.4	25.9	26.7	26.8	26.9
	Regtr （2022）	57.3	55.2	53.8	52.7	51.1	27.6	27.3	27.1	26.6	25.4
	GeoTransformer （2023）	71.9	75.2	76.0	82.2	85.1	43.5	45.3	46.2	52.9	57.7
	RoReg （2023）	81.6	80.2	75.1	74.1	75.2	39.6	39.6	34.0	31.9	34.5
	DIM-RFNet	71.8	75.2	76.0	82.2	85.1	43.7	45.3	46.1	52.9	57.8
RR /%	PerfectMatch （2019）	78.4	76.2	71.4	67.6	50.8	33.0	29.0	23.3	17.0	11.0
	FCGF （2019）	85.1	84.7	83.3	81.6	71.4	40.1	41.7	38.2	35.4	26.8
	D3Feat （2020）	81.6	84.5	83.4	82.4	77.9	37.2	42.7	46.9	43.8	39.1
	SpinNet （2021）	88.6	86.6	85.5	83.5	70.2	59.8	54.9	48.3	39.8	26.8
	Predator （2021）	89.0	89.9	90.6	88.5	86.6	59.8	61.2	62.4	60.8	58.1
	CofiNet （2021）	89.3	88.9	88.4	87.4	87.0	67.5	66.2	64.2	63.1	61.0
	Regtr （2022）	92.0	91.2	89.7	90.6	90.4	64.8	64.4	64.2	62.3	59.7
	GeoTransformer （2023）	92.0	91.8	91.8	91.4	91.2	75.0	74.8	74.2	74.1	73.5
	RoReg （2023）	92.9	93.2	92.7	93.3	91.2	70.3	71.2	69.5	67.9	64.3
	DIM-RFNet	92.0	91.7	91.6	91.4	91.2	74.9	74.9	74.3	74.2	73.6

Table 5. Experimental results on OdometryKITTI datasets

View table

Table 5. Experimental results on OdometryKITTI datasets

Method	RTE /cm	RRE /（°）	RR /%
3DFeat-Net （2019）	25.9	0.57	96.0
FCGF （2019）	9.5	0.30	96.6
D3Feat （2020）	7.2	0.30	99.8
Predator （2021）	6.8	0.27	99.8
GLORN （2022）	6.2	0.27	99.8
GeoTransformer （2023）	7.4	0.27	99.8
DIM-RFNet	6.7	0.25	99.8

Table 6. Ablation experimental results on ModelLoNet and ModelNet40 datasets

View table

Table 6. Ablation experimental results on ModelLoNet and ModelNet40 datasets

ENSF	DNSF	ORE and ORD	RRE of ModelNet40	RTE of ModelNet40	CD of ModelNet40	RRE of ModelLoNet	RTE of ModelLoNet	CD of ModelLoNet
×	×	×	12.745	0.312	0.1243	21.416	0.752	0.1351
√	×	×	11.762	0.163	0.0112	16.523	0.253	0.0269
√	√	×	8.993	0.143	0.0099	13.475	0.201	0.0207
×	×	√	1.738	0.019	0.0085	7.408	0.152	0.0106
√	×	√	1.723	0.015	0.0081	7.351	0.148	0.0102
√	√	√	1.542	0.014	0.0008	4.975	0.115	0.0079

Tools

Get Citation

Copy Citation Text

Han Zhou, Xuchu Wang, Yue Yuan. Three-Dimensional Point Cloud Registration Network Based on Deep Interactive Multi-Scale Receptive Field Feature Learning[J]. Acta Optica Sinica, 2024, 44(14): 1415001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites