Point Cloud Registration in Multidirectional Affine Transformation with Variance Compensation

Chang Wang; Qin Shu; Yunxiu Yang; Shijie Deng

doi:10.3788/AOS201939.0215002

Journals >Acta Optica Sinica >Volume 39 >Issue 2 >Page 0215002 > Article

Acta Optica Sinica
Vol. 39, Issue 2, 0215002 (2019)

Point Cloud Registration in Multidirectional Affine Transformation with Variance Compensation

Chang Wang^1、*, Qin Shu^1、*, Yunxiu Yang², and Shijie Deng²

Author Affiliations

¹ College of Electrical Engineering and Information Technology, Sichuan University, Chengdu, Sichuan 610065, China

² Southwest Institute of Technical Physics, Chengdu, Sichuan 610041, China

show less

DOI: 10.3788/AOS201939.0215002 Cite this Article Set citation alerts

Chang Wang, Qin Shu, Yunxiu Yang, Shijie Deng. Point Cloud Registration in Multidirectional Affine Transformation with Variance Compensation[J]. Acta Optica Sinica, 2019, 39(2): 0215002 Copy Citation Text

show less

Fig. 1. Global vector features. (a) In Ω; (b) in Γ

Download full size

Fig. 2. Initial states of point clouds. (a) Bunny; (b) Horse

Download full size

Fig. 3. Registration effects of point clouds after similarity transformation by different algorithms. (a) Algorithm in Ref. [4]; (b) ICP; (c) Scale-ICP; (d) CPD; (e) MARVC

Download full size

Fig. 4. Initial states after multidirectional affine transformation. (a) Bunny; (b) Horse

Download full size

Fig. 5. Registration effects for different algorithms after multidirectional affine transformation. (a) Algorithm in Ref. [4]; (b) ICP; (c) Scale-ICP; (d) CPD; (e) MARVC

Download full size

Fig. 6. Registration effects of point clouds for different algorithms under random loss and multi-directional affine transformation. (a) Scale-ICP; (b) CPD; (c) MARVC

Download full size

Fig. 7. Registration effects of point clouds with 20 dB random noise for different algorithms under random loss and multi-directional affine transformation. (a) ICP; (b) Scale-ICP; (c) CPD; (d) MARVC

Download full size

Fig. 8. Actual objects scanned by portable laser scanner, obtained data, and registration effects. (a) Three groups of objects; (b) obtained data; (c) registration effects by MARVC algorithm

Download full size

Point cloud	s	Algorithm in Ref. [4]	ICP	Scale-ICP	CPD	MARVC
	1.5	1.8562	1.8216	0.7687	0.3727	5.6550×10^-4
Bunny	2	3.2229	3.1774	0.0013	0.2641	3.6760×10^-4
	3	6.7786	6.7907	2.1157	0.3519	3.1740×10^-4
	1.5	12.3420	13.0710	7.0815	2.2140	7.9900×10^-4
Horse	2	29.2730	37.7740	5.1928	3.1270	8.0440×10^-4
	3	63.0870	62.2310	6.4387	3.0251	2.2180×10^-3

Table 1. RMSE for different algorithms mm

Point cloud	s	Algorithm in Ref. [4]	ICP	Scale-ICP	CPD	MARVC
	1.5	6.2565	649.6780	8.1354	988.3850	2.7407
Bunny	2	5.9417	351.4250	21.6884	952.2650	4.1969
	3	6.5417	257.1220	8.4478	986.4610	3.5356
	1.5	7.1232	296.7850	9.1698	2146.4000	4.6706
Horse	2	6.8912	499.6830	24.2360	2241.3700	0.4924
	3	7.2567	423.3320	14.3790	2549.5600	1.3736

Table 2. Registration time for different algorithmss

Point cloud	Algorithm in Ref. [4]	ICP	Scale-ICP	CPD	MARVC
Bunny	14.5430	9.2450	3.6550	0.1160	1.1030×10^-3
Horse	39.3240	38.8230	10.9120	107.4520	3.1610×10^-3

Table 3. RMSE for different algorithmsmm

Point cloud	Algorithm in Ref. [4]	ICP	Scale-ICP	CPD	MARVC
Bunny	2.3290	1187.5130	24.3850	208.6340	2.8120
Horse	3.3590	502.6340	27.3530	434.2230	0.8770

Table 4. Registration time for different algorithmss

Point cloud	ICP	Scale-ICP	CPD	MARVC
Bunny	8.0910	8.0910	0.3080	0.0302
Horse	38.8170	11.3050	2.1010	0.0763

Table 5. RMSE for different algorithms after random loss of data pointsmm

Point cloud	ICP	Scale-ICP	CPD	MARVC
Bunny	949.7260	31.7890	191.8430	1.3710
Horse	420.9660	103.7960	363.9730	3.2760

Table 6. Registration time for different algorithms after random loss of data pointss

Proportion of lost points		10%			20%
Signal-to-noise ratio		15 dB	20 dB	25 dB	15 dB	20 dB	25 dB
	ICP	0.9852	0.9460	0.9371	0.9886	0.9808	1.005
Bunny	Scale-ICP	1.4010	1.5680	1.6530	1.5020	1.6020	1.6670
	CPD	0.4449	0.2958	0.3768	0.3990	0.3561	0.3141
	MARVC	0.3976	0.2366	0.1414	0.3952	0.2398	0.2104
	ICP	5.5390	5.2670	5.1220	23.3880	9.7710	9.2740
Horse	Scale-ICP	6.1890	4.8690	9.4320	18.5440	9.5680	9.1640
	CPD	5.7380	4.5980	3.2550	5.4770	3.4740	3.5850
	MARVC	4.6890	3.2150	2.1490	4.7420	2.6170	2.1290

Table 7. RMSE for different algorithms under different interference environmentsmm

Proportion of lost points		10%			20%
Signal-to-noise ratio		15 dB	20 dB	25 dB	15 dB	20 dB	25 dB
	ICP	1023.0880	918.7300	914.6390	830.3700	779.6490	846.3800
Bunny	Scale-ICP	31.0410	20.2570	17.1920	31.8360	17.3670	17.8140
	CPD	182.9190	204.9210	210.1450	170.3430	170.8650	171.7440
	MARVC	3.4680	3.3600	4.3160	4.1900	7.5390	4.8620
	ICP	470.2290	428.7340	510.2460	747.7910	232.7110	412.3490
Horse	Scale-ICP	45.0120	35.2130	84.9230	56.1270	44.2320	62.5230
	CPD	397.9310	492.1960	716.1470	405.0180	402.9340	297.3500
	MARVC	1.8410	1.7860	1.7430	1.4060	2.4450	1.5570

Table 8. Registration time for different algorithms under different interference environmentss

Object No.	Parameter
Object 1	RMSE /mm	0.5974	2.4375	3.8148	0.7934
Object 1	Time /s	5.773	634.755	6.608	64.099
Object 2	RMSE /mm	0.5724	7.4200	7.4088	3.1088
Object 2	Time /s	4.835	446.558	4.632	47.154
Object 3	RMSE /mm	0.6431	8.9753	9.4262	0.6919
Object 3	Time /s	6.703	827.352	16.724	242.237

Table 9. Registration effects of three groups of objects for different algorithms

Chang Wang, Qin Shu, Yunxiu Yang, Shijie Deng. Point Cloud Registration in Multidirectional Affine Transformation with Variance Compensation[J]. Acta Optica Sinica, 2019, 39(2): 0215002

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information