Multiview Point Cloud Registration Method for Nonspherical Objects Based on Manifold Clustering

Hui Chen; Yibo Wang; Heping Huang; Fei Yan; Yunfeng Huang

doi:10.3788/LOP222574

Laser & Optoelectronics Progress, Volume. 60, Issue 16, 1615005(2023)

Multiview Point Cloud Registration Method for Nonspherical Objects Based on Manifold Clustering

Hui Chen¹, Yibo Wang¹, Heping Huang², Fei Yan³, and Yunfeng Huang^1、*

Author Affiliations

¹College of Automation Engineering, Shanghai University of Electric Power, Shanghai 200090, China

²Zhengtai Instrument (Hangzhou) Co., Ltd., Hangzhou 310052, Zhejiang, China

³Shanghai Minghua Electric Power Science & Technology Co., Ltd., Shanghai 200437, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Comparison between Euclidean distance and geodesic distance. (a) Euclidean distance; (b) geodesic distance

Download full size

Fig. 2. Calculation method of geodesic distance. (a) Directed weighted graph in space; (b) weight between two points; (c) shortest path between two points

Download full size

Fig. 3. Flowchart of the thermal gradient method

Download full size

Fig. 4. Selection of neighbourhood feature points. (a) Curve with little surface change; (b) curve with large surface changes

Download full size

Fig. 5. Flowchart of fine registration

Download full size

Fig. 6. Cross-section of multiview point cloud registration. (a) Multiview point cloud registration model; (b) the initial cross-section of fine registration; (c) the results of the MAICP method; (d) the results of the LRS method; (e) the results of the JRMPC method; (f) the results of K-means method; (g) the results of the proposed method

Download full size

Fig. 7. Comparison of the local effect of cross-section. (a) The local magnification effect of the registration result of the Dragon model obtained by K-means method; (b) the local magnification effect of the registration result of the Dragon model obtained by the proposed method; (c) the local magnification effect of the registration result of the Chicken model obtained by JRMPC method; (d) the local magnification effect of the registration result of the Chicken model obtained by the proposed method

Download full size

Table 1. Point cloud data information
View table
Table 1. Point cloud data information
Dataset Bunny Dragon Happy Chicken
Number of views 10 15 15 16
Total points 362272 469193 1099005 418412

Table 2. Accuracy comparison of different methods

View table

Table 2. Accuracy comparison of different methods

Dataset	Error	Initial	MAICP	LRS	JRMPC	K-means	Proposed method
Bunny	E_R	0.0386	0.0378	0.0439	0.0236	0.0219	0.0096
Bunny	E_t	2.9426	2.1693	2.0452	1.5247	1.1615	0.8556
Dragon	E_R	0.0393	0.0420	0.2838	0.0346	0.0433	0.0111
Dragon	E_t	4.5040	2.9385	2.5304	2.6730	2.9957	1.2980
Happy	E_R	0.0429	0.0603	0.0512	0.0217	0.0258	0.0069
Happy	E_t	1.5065	1.9258	0.5326	0.4436	0.6094	0.1795
Chicken	E_R	0.0395	0.0425	0.0806	0.0274	0.0304	0.0170
Chicken	E_t	2.3417	2.1060	1.4076	1.1525	0.9927	0.6801

Table 3. Efficiency comparison of two kinds of geodesic distance matrix calculation methods
View table
Table 3. Efficiency comparison of two kinds of geodesic distance matrix calculation methods
Dataset Bunny Dragon Happy Chicken
Matrix order 2950×2950 1915×1915 2241×2241 3558×3558
Floyd method 91.09 33.83 48.11 155.07
Thermal gradient method 52.87 21.65 30.89 82.57

Table 4. Efficiency comparison of different registration methods
View table
Table 4. Efficiency comparison of different registration methods
Data MAICP LRS JRMPC K-means Proposed method
Bunny 75.59 55.96 344.89 3.11 475.83
Dragon 241.33 695.78 621.71 14.91 1649.59
Happy 272.05 289.00 942.32 34.75 1739.63
Chicken 263.67 317.03 751.89 15.33 1321.12

Table 5. Robustness comparison of different methods for the Bunny dataset

View table

Table 5. Robustness comparison of different methods for the Bunny dataset

Bunny	Error	MAICP	LRS	JRMPC	K-means	Proposed method
100	E_R	0.0548	0.0523	0.0210	0.0225	0.0091
100	E_t	2.2366	1.9532	1.5651	1.1869	0.8695
200	E_R	0.0693	0.0753	0.0311	0.0270	0.0109
200	E_t	2.4352	2.5266	1.6393	1.8636	1.0210
400	E_R	0.0780	0.0815	0.0420	0.0308	0.0136
400	E_t	3.2534	3.7306	1.9634	2.2453	1.0351

Table 6. Robustness comparison of different methods for the Dragon dataset

View table

Table 6. Robustness comparison of different methods for the Dragon dataset

Dragon	Error	MAICP	LRS	JRMPC	K-means	Proposed method
100	E_R	0.0698	0.3157	0.0344	0.0393	0.0134
100	E_t	3.0489	2.5354	1.6541	2.6498	1.5274
200	E_R	0.0823	0.4425	0.0443	0.0516	0.0229
200	E_t	3.5402	2.7454	2.9126	3.4135	1.7779
400	E_R	0.0984	0.5829	0.0569	0.0762	0.0597
400	E_t	4.1586	3.7205	3.6746	4.3379	1.8025

Table 7. Robustness comparison of different methods for the Happy dataset

View table

Table 7. Robustness comparison of different methods for the Happy dataset

Happy	Error	MAICP	LRS	JRMPC	K-means	Proposed method
100	E_R	0.0759	0.0536	0.0210	0.0383	0.0066
100	E_t	2.0891	0.7924	0.5609	0.6051	0.2216
200	E_R	0.0961	0.0771	0.0414	0.0341	0.0061
200	E_t	2.3149	1.1587	0.9549	0.7376	0.3141
400	E_R	0.1542	0.1428	0.0702	0.0490	0.0131
400	E_t	2.5044	1.6032	1.5121	1.4671	0.6505

Table 8. Robustness comparison of different methods for the Chicken dataset

View table

Table 8. Robustness comparison of different methods for the Chicken dataset

Chicken	Error	MAICP	LRS	JRMPC	K-means	Proposed method
100	E_R	0.0566	0.1217	0.0693	0.0304	0.0286
100	E_t	2.3039	1.6762	1.3274	1.2214	0.8881
200	E_R	0.0735	0.1970	0.0723	0.0342	0.0320
200	E_t	2.5464	1.9279	1.7106	1.3195	1.0615
400	E_R	0.0975	0.1832	0.0825	0.0418	0.0382
400	E_t	2.9942	2.3206	2.2112	1.7151	1.3654

Tools

Get Citation

Copy Citation Text

Hui Chen, Yibo Wang, Heping Huang, Fei Yan, Yunfeng Huang. Multiview Point Cloud Registration Method for Nonspherical Objects Based on Manifold Clustering[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1615005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites