Optimization and Verification of Iterative Closest Point Algorithm Using Principal Component Analysis

Fengyuan Shi; Chunming Zhang; Lihui Jiang; Qi Zhou; Di Pan

doi:10.3788/LOP202259.2211001

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 22 >Page 2211001 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 22, 2211001 (2022)

Optimization and Verification of Iterative Closest Point Algorithm Using Principal Component Analysis

Fengyuan Shi^1,2, Chunming Zhang^3,*, Lihui Jiang^1,2, Qi Zhou^1,2, and Di Pan^1,2

Author Affiliations

¹Shanghai Institute of Aerospace Control Technology, Shanghai 201109, China

²Shanghai Key Laboratory of Space Intelligent Control Technology, Shanghai 201109, China

³Shanghai Academy of Spaceflight Technology, Shanghai 201109, China

show less

DOI: 10.3788/LOP202259.2211001 Cite this Article Set citation alerts

Fengyuan Shi, Chunming Zhang, Lihui Jiang, Qi Zhou, Di Pan. Optimization and Verification of Iterative Closest Point Algorithm Using Principal Component Analysis[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2211001 Copy Citation Text

show less

Fig. 1. Schematic diagram of principal component analysis (PCA) in case of two-dimensional input

Download full size

Fig. 2. Simulation result graph of ICP algorithm

Download full size

Fig. 3. Flow chart of PCA improved algorithm

Download full size

Fig. 4. Initial posture of point cloud. (a) Query point cloud; (b) reference point cloud; (c) relative position relationship between reference point cloud (lower left quarter) and query point cloud (upper right corner)

Download full size

Fig. 5. Simulation result graph of ICP algorithm. (a) Iteration error; (b) registration error; (c) registration result

Download full size

Fig. 6. Decentralized point cloud image. (a) Rough initial value; (b) accurate initial value

Download full size

Fig. 7. Rough initial pose registration result map. (a) Iteration error; (b) registration error; (c) registration result

Download full size

Fig. 8. Accurate initial pose registration result map. (a) Iteration error; (b) registration error; (c) registration result

Download full size

Fig. 9. Top three principal components of point cloud. (a) Reference point cloud; (b) query point cloud

Download full size

Fig. 10. PCA preprocessing registration results. (a) Iteration error; (b) registration error; (c) registration result

Download full size

Fig. 11. PCA iterative registration results. (a) Iteration error; (b) registration error; (c) registration result

Download full size

Fig. 12. PCA iteration+rough initial value registration result. (a) Iteration error; (b) registration error; (c) registration result

Download full size

Fig. 13. PCA iteration+accurate initial value registration result. (a) Iteration error; (b) registration error; (c) registration result

Download full size

ICP algorithm

Input：

$q \in Q$ ：Model point

$p \in P$ ：Query point

$R_{0}$ ：Initial rotation matrix

$t_{0}$ ：Initial translation matrix

Output：

$R$ ：Rotation matrix

$t$ ：Translation matrix

1） $P_{t e m} = R e p r o j e c t (P, R_{0} t_{0})$

2） $f o r$ $i t e r i s 1$ $t o$ $m a x_i t e r$ $d o$

3） $[P_{t e m}] = S e a r c h N N (Q)$

4） $[R t] = E s t i m a t e T r a n s (P_{t e m}, Q, R_{t e m} t_{t e m})$

5） $P' = R e p r o j e c t (P_{t e m}, R t)$

6）end for

7）return $R$ _t

Table 1. Calculation process of ICP algorithm

	Time /s	Iteration times	Registration error	Iteration error	Registration result
ICP	179.262697	100	8.2823	0.8717	Fail
ICP+rough	179.743619	100	10.1698	0.1089	Fail
ICP+accurate	176.654238	100	24.5595	0.1531	Fail
PCA	175.897402	100	25.8340	0.3129	Fail
PCA iteration	169.420848	94	6.6857	0.0001008	Succeed
PCA iteration+rough	52.328031	29	3.6489	0.0009937	Succeed
PCA iteration+accurate	19.427939	10	2.1932	0.0001471	Succeed