Aircraft localization method based on hierarchical rotation matching of infrared images

Algorithm 1: Hierarchical rotation matching algorithm
Input: Image A and image B
Output: Location of the center of image B in image A
1) RBN-SuperPoint extracting feature points:
Image A$ \to $$ {d^A} $, ${ {{p} }^A}$; image B$ \to $$ {d^B} $, $ {p^B} $
2) L-LightGlue adaptive matching:
① ${f^A} = {\rm{fusion}}({d^A},{PE} ({p^A}))$; ${f^B} = {\rm{fusion}}({d^B}{\text{,PE} }({p^B}))$
② Self-attention: ${ {\boldsymbol{\hat f} }^A} = {{\rm{Self}}} ({ {f} }_i^A,{ {f} }_j^A)$, ${ {\boldsymbol{\hat f} }^B} = {{\rm{Self}}} ({ {f} }_i^B,{ {f} }_j^B)$
Cross-attention: ${ {\boldsymbol{\tilde f} }^{AB} } = {{\rm{Cross}}} ({\boldsymbol{\hat f} }_i^A,{\boldsymbol{\hat f} }_j^B)$
③ Confidence classifier: if $ \left( {\dfrac{1}{{N + M}}\displaystyle\sum\limits_{i \in \left\{ {A,B} \right\}} {\left[\kern-0.15 em\left[ {{c_i} > {\lambda _l}} \right]\kern-0.15 em\right]} } \right) > \alpha $
to step ④
Else:
pruning and to step ②
④ Optimal matching:
$ {{\boldsymbol{P}}_{ij}} = \sigma _i^A\sigma _j^B\mathop {{{\mathrm{Softmax}}} }\limits_{k \in A} {({{\boldsymbol{S}}_{kj}})_i}\mathop {{{\mathrm{Softmax}}} }\limits_{k \in B} {({{\boldsymbol{S}}_{ik}})_j} $
Record the matched pairs: ${ {\boldsymbol{P} }_{AB} } = {\rm{append}}({ {\boldsymbol{P} }_{ij} } > \tau )$
3) ${{\boldsymbol{P}}_{AB}}$$ \to $${{\boldsymbol{H}}_1}$, image B rotation: ${\boldsymbol{C}} = {{\boldsymbol{H}}_1}{\boldsymbol{B}}$
4) Image A and C repeat step (2) to get ${{\boldsymbol{P}}_{AC}}$
5) Feature points rotation: ${{\boldsymbol{P'}}_{AB}} = {{\boldsymbol{H}}_1}^{ - 1}{{\boldsymbol{P}}_{AC}}$, ${{\boldsymbol{P'}}_{AB}}$$ \to $${{\boldsymbol{H}}_2}$
6) Corresponding points in image A: $ Po{s^A} = {{\boldsymbol{H}}_2}{\boldsymbol{x}}_{{\mathrm{center}}}^B $

Table 2. Performance comparison of feature point extraction algorithms

View table

View in Article

Table 2. Performance comparison of feature point extraction algorithms

Algorithms	Scene 1		Scene 2		Scene 3
Algorithms	${p^A}$	${p^B}$	${p^A}$	${p^B}$	${p^A}$	${p^B}$
SIFT	463	481	804	740	410	285
SURF	2922	3291	1608	1571	904	639
ORB	254	274	589	628	554	405
SuperPoint	1286	1381	543	522	424	744
RBN-SuperPoint	3209	3462	2542	2311	1375	1124

Table 3. Performance comparison of matching algorithms

View table

View in Article

Table 3. Performance comparison of matching algorithms

Algorithms	Scenes	${N_{{\rm{all}}} }$	$ {N_{CMP}} $	MA	AME/pixel	Time/s
LoFTR	1	175	3	1.71%	9.98	0.42
	2	220	4	1.82%	10.06	0.42
	3	120	6	5.00%	17.10	0.42
COTR	1	100	3	3.00%	9.73	45.77
	2	58	11	18.97%	22.13	45.05
	3	69	33	47.83%	17.68	45.04
SuperGlue	1	148	142	95.95%	2.86	3.14
	2	132	128	96.97%	2.48	3.10
	3	59	53	89.83%	1.14	3.12
LightGlue	1	78	37	47.44%	12.69	2.24
	2	44	26	59.09%	4.29	2.21
	3	37	25	67.57%	10.26	2.17
This paper	1	462	449	97.19%	2.81	2.29
	2	280	276	98.57%	2.42	2.25
	3	168	165	98.21%	1.07	2.22

Table 4. Performance comparison of aircraft positioning performance

View table

View in Article

Table 4. Performance comparison of aircraft positioning performance

Group	True value	LoFTR	COTR	SuperGlue	LightGlue	This paper
1	364, 182	354, 243	424, 206	285, 104	363, 182	362, 185
2	247, 292	258, 314	248, 307	242, 288	242, 285	243, 287
3	235, 127	265, 100	273, 129	203, 88	249, 207	234, 129
Average errors/pixel		42.26	39.23	55.96	30.27	4.08

Tools

Get Citation

Copy Citation Text

Qingge Li, Xiaogang Yang, Ruitao Lu, Jiwei Fan, Bin Tang, Zhenyu Zhang, Siyu Wang, Shuang Su. Aircraft localization method based on hierarchical rotation matching of infrared images[J]. Infrared and Laser Engineering, 2024, 53(5): 20240086

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: Feb. 29, 2024

Accepted: --

Published Online: Jun. 21, 2024

The Author Email:

DOI:10.3788/IRLA20240086

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology