Point Cloud Registration Based on Surface Feature Degree and Improved Dung Beetle Optimization Algorithm

Junchao Zhu; Siyuan Song; Fangfang Han; Minghui Zhang

doi:10.3788/LOP241656

Laser & Optoelectronics Progress, Volume. 62, Issue 6, 0615010(2025)

Point Cloud Registration Based on Surface Feature Degree and Improved Dung Beetle Optimization Algorithm

Junchao Zhu*... Siyuan Song, Fangfang Han and Minghui Zhang |Show fewer author(s)

Author Affiliations

School of Electrical Engineering and Automation, Tianjin University of Technology, Tianjin 300384, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Point cloud surface normal vector variation diagram. (a) Flat surface variation; (b) dramatic surface change

Download full size

Fig. 2. Flow chart of point cloud registration based on ST-DBO algorithm

Download full size

Fig. 3. Unregistered point cloud data. (a) Bunny; (b) dragon; (c) toy; (d) airplane; (e) chair

Download full size

Fig. 4. Images corresponding to collected point cloud data. (a) Toy 01; (b) toy 02

Download full size

Fig. 5. Registration results of different feature extraction methods in point cloud model. (a) SIFT; (b) ISS; (c) Harris3d; (d) proposed method

Download full size

Fig. 6. Comparison of effects of different registration algorithms. (a) ICP; (b) SSA; (c) PSO; (d) GWO; (e) DBO; (f) ST-DBO

Download full size

Fig. 7. Registration results of different feature extraction methods in noise point cloud model. (a) SIFT; (b) ISS; (c) Harris3d; (d) proposed method

Download full size

Fig. 8. Comparison of the effects of different registration algorithms in the noise point cloud model. (a) ICP; (b) SSA; (c) PSO; (d) GWO; (e) DBO; (f) ST-DBO

Download full size

Fig. 9. Registration accuracy changes under different parameter settings

Download full size

Table 0. [in Chinese]

View table

Table 0. [in Chinese]

Algorithm 1 t-distribution perturbation strategy for adaptive probability
Require： Iterations M， Population size N， bound $u_{b}, l_{b}$
Ensure： Optimal position X and its fitness value f_min
1： w₁←0.5； w₂←0.1；
2： while t ≤ M do
3： Calculate adaptive probability p = w₁-w₂×（M-t）/M
4： for j = 1 to N do
5： if p < rand then
6： Generate new position T_emp using t-distribution： T_emp = x（i，：） + x（i，：） × trnd（i）
7： Ensure T_emp is within bounds： T_emp = max［min（T_emp， u_b）， l_b］
8： Calculate fitness value fitvalue = fobj（T_emp）
9： if fitvalue < fit（j） then
10： Update position x（j，：）= T_emp
11： Update fitness fit（j） = fitvalue
12： end if
13： end if
14： end for
14： Update position best X =x（j，：），global best fitness f_min=fit（j）
15： t = t + 1
16： end while
17： Output optimal position best X and its fitness value f_min

Table 1. Comparison of registration accuracy and extraction efficiency of different feature extraction methods

View table

Table 1. Comparison of registration accuracy and extraction efficiency of different feature extraction methods

Algorithm	SIFT		ISS		Harris3d		Proposed
Algorithm	RMSE	Time /s	RMSE	Time /s	RMSE	Time /s	RMSE	Time /s
Bunny	0.00390	1.68	0.00450	24.57	0.00383	9.40	0.00323	1.25
Dragon	0.00202	1.72	0.00219	20.88	0.00212	8.11	0.00191	1.77
Toy	1.66600	27.49	1.50800	39.07	1.90500	10.27	1.47900	2.15
Airplane	0.00294	1.13	0.00280	0.85	0.00673	0.84	0.00207	1.10
Chair	0.00684	1.84	0.00362	0.98	0.00580	1.12	0.00299	1.15

Table 2. Comparison of the results of different registration algorithms

View table

Table 2. Comparison of the results of different registration algorithms

Algorithm	Bunny RMSE	Dragon RMSE	Toy RMSE	Airplane RMSE	Chair RMSE	Improved accuracy /%
ICP	0.00418	0.00231	1.506	0.00581	0.00443	27.74
SSA	0.00347	0.00212	1.521	0.00364	0.00431	18.67
PSO	0.00350	0.00223	1.581	0.00374	0.00500	22.67
GWO	0.00356	0.00211	1.616	0.00543	0.00688	29.13
DBO	0.00364	0.00222	1.768	0.00288	0.00321	15.31
ST-DBO	0.00323	0.00191	1.479	0.00207	0.00299

Table 3. Comparison of registration accuracy and extraction efficiency of different feature extraction methods under the noise point cloud model

View table

Table 3. Comparison of registration accuracy and extraction efficiency of different feature extraction methods under the noise point cloud model

Algorithm	SIFT		ISS		Harris3d		Proposed
Algorithm	RMSE	Time /s	RMSE	Time /s	RMSE	Time /s	RMSE	Time /s
Bunny	0.00432	1.72	0.00551	23.54	0.00405	8.58	0.00330	1.30
Toy	1.70300	26.26	1.54000	44.51	2.00400	10.16	1.52000	1.69
Airplane	0.00368	2.14	0.00410	0.87	0.00722	0.85	0.00239	1.10

Table 4. Comparison of results of different registration algorithms after adding of noise

View table

Table 4. Comparison of results of different registration algorithms after adding of noise

Algorithm	Dragon RMSE	Toy RMSE	Chair RMSE	Improved accuracy /%
ICP	0.00301	1.911	0.01130	41.92
SSA	0.00392	1.755	0.00798	40.78
PSO	0.00285	1.626	0.00818	32.43
GWO	0.00286	1.808	0.01260	40.22
DBO	0.00303	1.866	0.00383	23.01
ST-DBO	0.00200	1.519	0.00320

Tools

Get Citation

Copy Citation Text

Junchao Zhu, Siyuan Song, Fangfang Han, Minghui Zhang. Point Cloud Registration Based on Surface Feature Degree and Improved Dung Beetle Optimization Algorithm[J]. Laser & Optoelectronics Progress, 2025, 62(6): 0615010

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Jul. 9, 2024

Accepted: Sep. 3, 2024

Published Online: Mar. 18, 2025

The Author Email:

DOI:10.3788/LOP241656

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 0. [in Chinese]

Table 0. [in Chinese]

Table 1. Comparison of registration accuracy and extraction efficiency of different feature extraction methods

Table 1. Comparison of registration accuracy and extraction efficiency of different feature extraction methods

Table 2. Comparison of the results of different registration algorithms

Table 2. Comparison of the results of different registration algorithms

Table 3. Comparison of registration accuracy and extraction efficiency of different feature extraction methods under the noise point cloud model

Table 3. Comparison of registration accuracy and extraction efficiency of different feature extraction methods under the noise point cloud model

Table 4. Comparison of results of different registration algorithms after adding of noise

Table 4. Comparison of results of different registration algorithms after adding of noise

微信扫一扫：分享