A Robust Feature Matching Method for Wide-Baseline Lunar Images

Qihao Peng; Tengqi Zhao; Chuankai Liu; Zhiyu Xiang

doi:10.3788/AOS230498

Acta Optica Sinica, Volume. 43, Issue 24, 2410001(2023)

A Robust Feature Matching Method for Wide-Baseline Lunar Images

Qihao Peng¹, Tengqi Zhao¹, Chuankai Liu^2,3, and Zhiyu Xiang^1,4、*

Author Affiliations

¹College of Information Science & Electronic Engineering, Zhejiang University, Hangzhou 310027, Zhejiang , China

²Beijing Aerospace Flight Control Center, Beijing 100190, China

³National Key Laboratory of Science and Technology on Aerospace Flight Dynamics, Beijing 100190, China

⁴Zhejiang Provincial Key Laboratory of Information Processing, Communication and Networking, Hangzhou 310027, Zhejiang , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Matching algorithm for wide-baseline lunar images DepthWarp-LoFTR

Download full size

Fig. 2. View synthesis based on scene depth

Download full size

Fig. 3. LoFTR network

Download full size

Fig. 4. Disparities of different algorithms. (a) SGBM; (b) GwcNet (sceneflow); (c) GwcNet (Moon)

Download full size

Fig. 5. Results of depth view synthesis. (a) Image at previous site; (b) synthetic image at previous site; (c) image at next site

Download full size

Fig. 6. Image matching results on No. 2 and No. 12 image pairs of LoFTR and ASIFT. (a) LoFTR; (b) ASIFT

Download full size

Fig. 7. Image matching results on No. 1 and No. 5 image pairs. (a) Ground-truth matches; (b) ASIFT; (c) LoFTR; (d) DepthWarp-ASIFT; (e) DepthWarp-LoFTR

Download full size

Fig. 8. Matching failure cases of DepthWarp-LoFTR. (a) No. 8; (b) No. 11

Download full size

Table 1. Accuracy of 3D reconstruction
View table
Table 1. Accuracy of 3D reconstruction
Algorithm 3D reconstruction error /m
3-7 m 8-12 m 13-19 m
SGBM 0.056 0.304 0.709
GwcNet（sceneflow） 0.012 0.073 0.225
GwcNet（Moon） 0.008 0.052 0.166

Table 2. Successful matching results of LoFTR and ASIFT
View table
Table 2. Successful matching results of LoFTR and ASIFT
No. LoFTR ASIFT
Inliners/Matches r_err t_err Inliners/Matches r_err t_err
2 145/751 28.73 35.14 x
4 64/447 64.64 22.67 x
9 124/1009 1.94 0.55 875/5316 2.54 1.56
12 211/822 2.82 3.21 105/1045 13.73 15.34
av — 24.53 15.39 — 8.14 8.45

Table 3. Matching results of DepthWarp-LoFTR and DepthWarp-ASIFT

View table

Table 3. Matching results of DepthWarp-LoFTR and DepthWarp-ASIFT

No.	DepthWarp-LoFTR			DepthWarp-ASIFT
	Inliners/Matches	r_err	t_err	Inliners/Matches	r_err	t_err
1	454/765	1.91	3.42	71/414	67.36	38.11
2	217/660	34.96	31.07	82/301	12.04	4.64
3	284/466	5.48	6.38	x
4	20/133	44.70	80.13	x
5	230/480	2.66	20.31	x
6	107/278	6.53	9.71	x
7	106/308	28.70	48.08	x
8	x			x
9	665/1025	1.47	0.69	2378/5129	0.52	7.44
10	53/281	9.36	22.40	30/121	71.51	14.24
11	x			x
12	1161/1487	0.62	0.58	184/297	3.81	4.05
av	—	13.64	22.28	—	31.05	13.70

Table 4. Running time of different algorithms
View table
Table 4. Running time of different algorithms
Algorithm Disparity prediction Running time Total
View
synthesis
Extraction
and matching
Outlier
rejection
DepthWarp-ASIFT 23.295 4.710 8.312 0.487 36.804
DepthWarp-LoFTR 23.295 4.710 10.844 0.211 39.060

Tools

Get Citation

Copy Citation Text

Qihao Peng, Tengqi Zhao, Chuankai Liu, Zhiyu Xiang. A Robust Feature Matching Method for Wide-Baseline Lunar Images[J]. Acta Optica Sinica, 2023, 43(24): 2410001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites