High-Precision Registration of Non-Homologous Point Clouds in Laser Scanning and Photogrammetry

Chunmei Hu; Huajie Fei; Guofang Xia; Xi Liu; Xinjian Ma

doi:10.3788/LOP202259.2415007

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 24 >Page 2415007 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 24, 2415007 (2022)

High-Precision Registration of Non-Homologous Point Clouds in Laser Scanning and Photogrammetry

Chunmei Hu^1,2, Huajie Fei^1,2,*, Guofang Xia³, Xi Liu⁴, and Xinjian Ma⁵

Author Affiliations

¹School of Geomatics and Urban Spatial Informatics, Beijing University of Civil Engineering and Architecture, Beijing 100044, China

²Representative Architecture and Ancient Architecture Database Engineering Research Center of the Ministry of Education, Beijing 100044, China

³China Cultural Relics Information Consulting Center, Beijing 100029, China

⁴Heilongjiang Surveying and Mapping Measuring Instrument Verification Station, Harbin 150081, Heilongjiang, China

⁵Beijing Institute of Surveying and Mapping Design, Beijing 100038, China

show less

DOI: 10.3788/LOP202259.2415007 Cite this Article Set citation alerts

Chunmei Hu, Huajie Fei, Guofang Xia, Xi Liu, Xinjian Ma. High-Precision Registration of Non-Homologous Point Clouds in Laser Scanning and Photogrammetry[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2415007 Copy Citation Text

show less

Fig. 1. Algorithm flow

Download full size

Fig. 2. FPFH calculation principle

Download full size

Fig. 3. Schematic diagram of octree structure

Download full size

Fig. 4. Point cloud visualization. (a) Face point cloud; (b) ear point cloud; (c) buddha head point cloud

Download full size

Fig. 5. Rough registration results of different algorithms. (a) Overall point cloud after down sampling; (b) 4PCS algorithm; (c) RANSAC algorithm; (d) proposed algorithm

Download full size

Fig. 6. Iterative estimation of grid size

Download full size

Fig. 7. Octree grid rendering. (a) Face point cloud; (b) ear point cloud; (c) buddha head point cloud

Download full size

Fig. 8. Corresponding point matching diagrams. (a) Before removing false matching; (b) after removing false matching; (c) the number of point pairs corresponding to the removal of errors varying with the number of iterations

Download full size

Fig. 9. Registration results of different methods. (a) Grid centroid; (b) nearest point of grid centroid ; (c) point with the smallest Euclidean distance in the grid

Download full size

Fig. 10. Registration results of different algorithms. (a) ICP algorithm; (b) proposed algorithm

Download full size

Date	Point cloud	Number of point clouds	Number of down sampled point clouds	Registration time /s	Registration error / $(10^{- 3} m)$
Face	Source point cloud	10818	934	22.01	$3.53$
Face	Target point cloud	158344	1009	22.01	$3.53$
Ear	Source point cloud	20096	1135	24.24	$2.29$
Ear	Target point cloud	114603	1160	24.24	$2.29$
Head	Source point cloud	176315	2280	56.17	$2.39$
Head	Target point cloud	1587694	2456	56.17	$2.39$

Table 1. Rough matching result of model point cloud

Method	Face		Ear		Head
Method	Time /s	$R_{M S E}$ / $(10^{- 3} m)$	Time /s	$R_{M S E}$ / $(10^{- 3} m)$	Time /s	$R_{M S E}$ / $(10^{- 3} m)$
4PCS	25.18	$4.34$	30.29	$2.91$	65.38	$3.34$
RANSAC	41.54	$4.11$	59.73	$2.66$	78.75	$2.78$
Proposed method	22.01	$3.53$	24.24	$2.29$	56.17	$2.39$

Table 2. Comparison of registration results of different coarse registration algorithms

Data	Grid centroid	Nearest point of grid centroid	Point with the smallest Euclidean distance in the grid
Face	$2.32 \times 10^{- 3}$	$1.01 \times 10^{- 3}$	$5.72 \times 10^{- 5}$
Ear	$3.97 \times 10^{- 3}$	$2.04 \times 10^{- 3}$	$1.59 \times 10^{- 4}$
Head	$4.36 \times 10^{- 3}$	$1.53 \times 10^{- 3}$	$1.55 \times 10^{- 4}$

Table 3. Comparison of registration errors of different corresponding point pair selection methods

Algorithm	Face		Ear		Head
Algorithm	Time /s	$R_{M S E}$ /m	Time /s	$R_{M S E}$ /m	Time /s	$R_{M S E}$ /m
ICP	106.44	$4.09 \times 10^{- 4}$	94.76	$4.46 \times 10^{- 4}$	1416.99	$1.71 \times 10^{- 3}$
Proposed algorithm	28.34	$5.72 \times 10^{- 5}$	30.42	$1.59 \times 10^{- 4}$	478.78	$1.55 \times 10^{- 4}$