MSPoint: Point Cloud Denoising Network Based on Multiscale Distribution Score

Hao Hu; Qibing Wang; Jiawei Lu; Hongye Su; Jiankun Lai; Gang Xiao

doi:10.3788/LOP222402

Laser & Optoelectronics Progress, Volume. 60, Issue 16, 1615002(2023)

MSPoint: Point Cloud Denoising Network Based on Multiscale Distribution Score

Hao Hu^1、*, Qibing Wang¹, Jiawei Lu¹, Hongye Su², Jiankun Lai³, and Gang Xiao^1、**

Author Affiliations

¹College of Mechanical and Electrical Engineering, China Jiliang University, Hangzhou 310018, Zhejiang, China

²College of Control Science and Engineering, Zhejiang University, Hangzhou 310027, Zhejiang, China

³Zhejiang Xin Zailing Technology Co., Ltd., Hangzhou 310051, Zhejiang, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(20)

Fig. 1. Overall network structure

Download full size

Fig. 2. Feature extraction module

Download full size

Fig. 3. Displacement prediction module

Download full size

Fig. 4. Relation between score and denoising effect

Download full size

Fig. 5. Score estimation unit

Download full size

Fig. 6. Effect of different loss functions on point cloud denoising results. (a) $L_{s}$ ; (b) $L_{a}$

Download full size

Fig. 7. Point cloud dataset. (a) block; (b) blade; (c) column; (d) joint; (e) casting; (f) cube; (g) fandisk

Download full size

Fig. 8. Schematic diagrams of point cloud neighborhood extraction. (a) joint; (b) blade; (c) block; (d) column

Download full size

Fig. 9. Comparison of 0.5% noise casting point cloud model denoising results

Download full size

Fig. 10. Comparison of 0.5% noise cube point cloud model denoising results

Download full size

Fig. 11. Comparison of 0.5% noise fandisk point cloud model denoising results

Download full size

Fig. 12. Point Cloud data of a university library

Download full size

Fig. 13. Comparison of gate steps before and after noise removal. (a) Before noise removal; (b) after noise removal

Download full size

Fig. 14. Comparison of back corridor before and after noise removal. (a) Before noise removal; (b) after noise removal

Download full size

Fig. 15. Comparison of corner before and after noise removal. (a) Before noise removal; (b) after noise removal

Download full size

Fig. 16. Loss convergence of disturbances of different degrees. (a) $σ = 0$ ; (b) $σ = 0.1 %$ ; (c) $σ = 0.5 %$ ; (d) $σ = 1.0 %$

Download full size

Table 1. Comparison of denoising effects of different neighborhood radii using D_CD error as the evaluation standard
View table
Table 1. Comparison of denoising effects of different neighborhood radii using D_CD error as the evaluation standard
$r$ joint blade block column
2% 5.751 5.164 7.222 4.697
5% 2.742 3.661 4.527 2.467
8% 3.826 5.219 4.056 2.926
10% 7.873 5.771 6.745 3.566

Table 2. Comparison of denoising effects of different algorithms using D_CD error as evaluation standard

View table

Table 2. Comparison of denoising effects of different algorithms using D_CD error as evaluation standard

Gaussian noise	D_CD error /×10^-5
Gaussian noise	Noisy	TD	PointCleanNet	Pointfilter	MSPoint
	casting
0.3%	1.916	2.234	1.754	1.401	1.266
0.5%	3.676	3.628	3.672	3.331	2.906
1.0%	9.033	9.581	8.260	7.391	7.113
	cube
0.3%	2.257	1.356	0.963	0.712	0.631
0.5%	4.388	1.691	1.512	1.537	1.236
1.0%	12.57	2.503	2.178	1.861	2.025
	fandisk
0.3%	1.744	2.901	1.631	0.853	0.672
0.5%	3.577	3.219	2.384	1.903	1.898
1.0%	11.13	4.196	3.012	3.599	2.891

Table 3. Comparison of denoising effects of different algorithms using P2F error as the evaluation standard

View table

Table 3. Comparison of denoising effects of different algorithms using P2F error as the evaluation standard

Gaussian noise	P2F error /×10^-3
Gaussian noise	Noisy	TD	PointCleanNet	Pointfilter	MSPoint
	casting
0.3%	2.389	3.258	2.369	1.870	1.774
0.5%	3.915	3.792	3.823	4.387	3.668
1.0%	13.27	14.18	9.337	9.701	8.487
	Cube
0.3%	2.675	1.934	1.430	1.309	0.903
0.5%	4.572	2.187	2.028	1.726	1.499
1.0%	15.92	3.066	2.273	2.837	1.968
	fandisk
0.3%	2.446	3.402	1.726	1.351	0.745
0.5%	4.473	3.776	3.222	2.878	2.182
1.0%	14.93	5.771	4.353	3.884	3.598

Table 4. Comparison of denoising effects using different loss functions
View table
Table 4. Comparison of denoising effects using different loss functions
Loss function 0.3% 0.5% 1%
D_CD P2F D_CD P2F D_CD P2F
$L_{s}$ 1.014 1.531 2.332 2.717 4.274 5.018
$L_{a}$ 0.856 1.140 2.013 2.522 4.069 4.684

Tools

Get Citation

Copy Citation Text

Hao Hu, Qibing Wang, Jiawei Lu, Hongye Su, Jiankun Lai, Gang Xiao. MSPoint: Point Cloud Denoising Network Based on Multiscale Distribution Score[J]. Laser & Optoelectronics Progress, 2023, 60(16): 1615002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Aug. 29, 2022

Accepted: Oct. 13, 2022

Published Online: Aug. 18, 2023

The Author Email: Hao Hu (huhao0127@yeah.net), Gang Xiao (xg@zjut.edu.cn)

DOI:10.3788/LOP222402

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of denoising effects of different neighborhood radii using DCD error as the evaluation standard

Table 1. Comparison of denoising effects of different neighborhood radii using DCD error as the evaluation standard

Table 2. Comparison of denoising effects of different algorithms using DCD error as evaluation standard

Table 2. Comparison of denoising effects of different algorithms using DCD error as evaluation standard

Table 3. Comparison of denoising effects of different algorithms using P2F error as the evaluation standard

Table 3. Comparison of denoising effects of different algorithms using P2F error as the evaluation standard

Table 4. Comparison of denoising effects using different loss functions

Table 4. Comparison of denoising effects using different loss functions

Table 1. Comparison of denoising effects of different neighborhood radii using D_CD error as the evaluation standard

Table 1. Comparison of denoising effects of different neighborhood radii using D_CD error as the evaluation standard

Table 2. Comparison of denoising effects of different algorithms using D_CD error as evaluation standard

Table 2. Comparison of denoising effects of different algorithms using D_CD error as evaluation standard