Design of partial overlap point cloud registration network driven by overlap score and matching matrix

Method	MIE（R）	MIE（t）	MAE（R）	MAE（t）
ICP^［15］	24.641 3	0.252 5	12.147 0	0.117 1
FGR^［16］	14.011 4	0.097 8	8.274 4	0.046 2
DeepGMR^［17］	14.361 2	0.158 9	7.091 4	0.077 5
IDAM^［6］	14.289 1	0.190 9	7.496 6	0.087 7
RPMNet^［5］	1.423 9	0.013 9	0.730 4	0.006 5
ROPNet^［8］	1.154 1	0.010 8	0.592 8	0.005 1
GAP-Net^［18］	1.487 1	0.015 2	—	—
GTGMM^［19］	—	—	0.560 6	0.010 2
Ours	0.906 8	0.008 2	0.462 6	0.003 9

Table 2. Experimental results of different algorithms on the ModelLoNet40 dataset
View table
View in Article
Table 2. Experimental results of different algorithms on the ModelLoNet40 dataset
Method MIE（R） MIE（t）
RPMNet^［5］ 7.342 0.124
Predator^［7］ 5.235 0.132
ROPNet^［8］ 5.045 0.049
REGTR^［20］ 3.930 0.087
UDPReg^［21］ 3.578 0.069
HPNet^［22］ 4.128 0.787
Ours 3.304 0.031

Table 3. Experimental results of different algorithms on the ShapeNetCore.v2 dataset

View table

View in Article

Table 3. Experimental results of different algorithms on the ShapeNetCore.v2 dataset

Method	RMSE（R）	RMSE（t）	MAE（R）	MAE（t）
IDAM^［6］	1.746	0.014	0.757	0.006
ReAgent^［23］	1.791	0.028	1.522	0.010
ROPNet^［8］	0.673	0.006	0.320	0.003
DORNet^［24］	1.376	0.009	0.734	0.006
UTOPIC^［11］	0.682	0.008	0.281	0.002
MGFPCR^［25］	0.581	0.002	0.253	0.001
Ours	0.337	0.004	0.097	0.001

Table 4. Experimental results of generalization performance of different algorithms on the ModelNet40 dataset

View table

View in Article

Table 4. Experimental results of generalization performance of different algorithms on the ModelNet40 dataset

Method	MIE（R）	MIE（t）	MAE（R）	MAE（t）
ICP^［15］	25.487 9	0.250 4	12.897 6	0.118 3
FGR^［16］	15.331 7	0.103 6	9.411 0	0.049 1
DeepGMR^［17］	18.360 8	0.199 8	9.142 5	0.099 7
IDAM^［6］	17.361 4	0.215 3	8.982 3	0.102 2
RPMNet^［5］	1.678 1	0.016 9	0.874 9	0.007 9
ROPNet^［8］	1.252 3	0.012 4	0.668 7	0.005 9
REGTR^［20］	0.965 3	0.009 7	0.509 7	0.004 6
DLF^［26］	—	—	0.478 1	0.004 3
GANPCR^［27］	—	—	5.158 1	0.043 1
Ours	0.915 4	0.009 0	0.473 1	0.004 3

Table 5. Experimental results of generalization performance of different algorithms on the Stanford Bunny dataset

View table

View in Article

Table 5. Experimental results of generalization performance of different algorithms on the Stanford Bunny dataset

Method	RMSE（R）	RMSE（t）	MAE（R）	MAE（t）
ICP^［15］	13.32	0.049 2	10.72	0.024 2
Go-ICP^［28］	12.92	0.042 9	4.52	0.028 2
DCP^［29］	6.44	0.040 6	4.78	0.037 4
ROPNet^［8］	3.76	0.028 9	2.33	0.015 9
Ours	2.80	0.019 7	2.20	0.015 1

Table 6. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.01

View table

View in Article

Table 6. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.01

Method	MIE（R）	MIE（t）	MAE（R）	MAE（t）
ICP^［15］	25.521 9	0.251 5	12.911 8	0.119 0
FGR^［16］	29.281 3	0.182 8	18.198 6	0.188 7
DeepGMR^［17］	18.807 5	0.202 7	9.367 2	0.099 2
IDAM^［6］	16.980 5	0.212 5	8.936 6	0.100 3
RPMNet^［5］	1.756 9	0.017 7	0.917 7	0.008 4
ROPNet^［8］	1.573 2	0.015 5	0.838 6	0.007 4
MFGNet^［31］	12.266 8	0.110 6	6.341 9	0.055 3
FINet^［32］	7.206 8	0.085 3	3.733 3	0.040 6
REGTR^［20］	1.795 9	0.017 1	0.931 0	0.008 1
OGMM^［33］	14.482 5	0.189 4	7.290 1	0.086 1
RGM^［34］	1.736 6	0.015 9	0.937 4	0.007 6
GeoTransformer^［35］	1.435 0	0.018 1	0.725 2	0.008 1
Ours	1.251 4	0.012 2	0.654 3	0.005 9

Table 7. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.02

View table

View in Article

Table 7. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.02

Method	MIE（R）	MIE（t）	MAE（R）	MAE（t）
ICP^［15］	30.709 6	0.462 0	15.704 2	0.221 5
FGR^［16］	26.311 1	0.224 0	53.048 3	0.438 2
DCP^［29］	30.709 6	0.462 0	15.704 2	0.221 5
IDAM^［6］	13.651 7	0.167 9	7.102 7	0.084 7
RPMNet^［5］	1.862 3	0.019 6	0.970 4	0.009 4
PREDATOR^［7］	2.806 5	0.027 2	4.782 2	0.050 0
ROPNet^［8］	2.004 6	0.019 7	1.064 0	0.009 5
MFGNet^［31］	12.725 1	0.119 6	6.568 9	0.060 2
MLIPPCR^［37］	1.783 7	0.018 0	0.942 7	0.008 8
Ours	1.797 7	0.017 4	0.954 5	0.008 4

Table 8. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.03

View table

View in Article

Table 8. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.03

Method	MIE（R）	MIE（t）	MAE（R）	MAE（t）
ICP^［15］	30.964 1	0.467 9	15.742 4	0.225 8
FGR^［16］	47.788 3	0.533 5	29.070 7	0.254 2
DCP^［29］	13.890 5	0.176 3	7.010 2	0.088 5
IDAM^［6］	13.593 9	0.169 6	7.074 3	0.085 2
RPMNet^［5］	2.231 8	0.023 5	1.160 0	0.011 2
PREDATOR^［7］	7.704 3	0.082 8	4.055 8	0.039 6
ROPNet^［8］	2.568 6	0.025 1	1.359 3	0.012 2
MFGNet^［31］	13.857 8	0.129 4	7.110 9	0.065 0
MLIPPCR^［37］	2.403 0	0.024 4	1.270 5	0.011 9
Ours	2.277 5	0.022 0	1.200 4	0.010 7

Table 9. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.04

View table

View in Article

Table 9. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.04

Method	MIE（R）	MIE（t）	MAE（R）	MAE（t）
ICP^［15］	31.172 0	0.467 4	15.800 5	0.227 0
FGR^［16］	56.843 5	0.617 3	35.528 5	0.293 7
DCP^［29］	13.970 7	0.176 5	7.042 2	0.088 6
IDAM^［6］	13.891 5	0.173 9	7.145 9	0.087 4
RPMNet^［5］	2.610 3	0.028 0	1.357 7	0.013 3
PREDATOR^［7］	10.222 5	0.107 0	5.330 9	0.051 1
ROPNet^［8］	3.154 2	0.031 0	1.673 4	0.015 1
MFGNet^［31］	14.702 9	0.137 5	7.588 6	0.069 2
MLIPPCR^［37］	2.998 5	0.030 5	1.573 8	0.014 8
Ours	2.577 7	0.024 8	1.382 1	0.012 1

Table 10. Real time experimental results of different algorithms on the ModelNet40 dataset

View table

View in Article

Table 10. Real time experimental results of different algorithms on the ModelNet40 dataset

Method	512（Points）	1 024（Points）	2 048（Points）
ICP^［15］	0.128	0.162	0.231
Go-ICP^［28］	14.053	14.140	14.212
FGR^［16］	0.172	0.187	0.185
PointNetLK^［38］	0.059	0.074	0.084
ROPNet^［8］	0.022	0.031	0.053
Ours	0.040	0.052	0.086

Table 11. The impact of each improvement step of this algorithm on the ModelNet40 dataset on the model
View table
View in Article
Table 11. The impact of each improvement step of this algorithm on the ModelNet40 dataset on the model
Method AO TO
MIE（R） MIE（t） MIE（R） MIE（t）
Baseline 1.203 1 0.013 7 1.573 2 0.015 5
Method 1 1.141 3 0.013 1 1.419 8 0.014 0
Method 2 1.022 2 0.011 2 1.349 8 0.013 0
Method 3 0.914 7 0.010 3 1.251 5 0.012 2

Table 12. Overlap precision and recall ablation experiment of this algorithm on the ModelNet40 dataset

View table

View in Article

Table 12. Overlap precision and recall ablation experiment of this algorithm on the ModelNet40 dataset

Method	重叠点（0.05/0.07）				代表性重叠点（0.05/0.07）
	AO		TO		AO		TO
	OP	OR	OP	OR	OP	OR	OP	OR
Baseline	0.810/0.939	0.836/0.822	0.806/0.936	0.838/0.824	0.913/0.987	0.382/0.347	0.908/0.984	0.382/0.348
Method 1	0.810/0.939	0.835/0.823	0.805/0.936	0.837/0.824	0.912/0.986	0.381/0.347	0.908/0.984	0.382/0.348
Method 2	0.826/0.954	0.853/0.837	0.820/0.950	0.853/0.838	0.919/0.989	0.384/0.348	0.914/0.986	0.385/0.349
Method 3	0.826/0.954	0.854/0.837	0.820/0.951	0.854/0.838	0.933/0.992	0.391/0.349	0.927/0.989	0.391/0.350

Tools

Get Citation

Copy Citation Text

Jianbing YI, Xin CHEN, Feng CAO, Shuxin YANG, Jingyong WANG. Design of partial overlap point cloud registration network driven by overlap score and matching matrix[J]. Optics and Precision Engineering, 2024, 32(22): 3366

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: Jun. 3, 2024

Accepted: --

Published Online: Mar. 10, 2025

The Author Email: Jianbing YI (yijianbing8@jxust.edu.cn)

DOI:10.37188/OPE.20243222.3366

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Experimental results of different algorithms on the ModelNet40 dataset

Table 1. Experimental results of different algorithms on the ModelNet40 dataset

Table 2. Experimental results of different algorithms on the ModelLoNet40 dataset

Table 2. Experimental results of different algorithms on the ModelLoNet40 dataset

Table 3. Experimental results of different algorithms on the ShapeNetCore.v2 dataset

Table 3. Experimental results of different algorithms on the ShapeNetCore.v2 dataset

Table 4. Experimental results of generalization performance of different algorithms on the ModelNet40 dataset

Table 4. Experimental results of generalization performance of different algorithms on the ModelNet40 dataset

Table 5. Experimental results of generalization performance of different algorithms on the Stanford Bunny dataset

Table 5. Experimental results of generalization performance of different algorithms on the Stanford Bunny dataset

Table 6. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.01

Table 6. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.01

Table 7. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.02

Table 7. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.02

Table 8. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.03

Table 8. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.03

Table 9. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.04

Table 9. Experimental results of different algorithms under Gaussian noise with a standard deviation of 0.04

Table 10. Real time experimental results of different algorithms on the ModelNet40 dataset

Table 10. Real time experimental results of different algorithms on the ModelNet40 dataset

Table 11. The impact of each improvement step of this algorithm on the ModelNet40 dataset on the model

Table 11. The impact of each improvement step of this algorithm on the ModelNet40 dataset on the model

Table 12. Overlap precision and recall ablation experiment of this algorithm on the ModelNet40 dataset

Table 12. Overlap precision and recall ablation experiment of this algorithm on the ModelNet40 dataset