Fast Anchor Point Matching for Emergency UAV Image Stitching Using Position and Pose Information

. 2020 Apr 3;20(7):2007. doi: 10.3390/s20072007

Algorithm 1: The Pseudo-Code of Stitching Two Adjacent Images

Input: Adjacent images

I_{1}

I_{2};

The camera parameter of images

c a m e r a_{1}, c a m e r a_{2}

1. Use

c a m e r a_{1}, c a m e r a_{2}

to get the rough registration of two images

f

2. Use S-H algorithm [34] and rough registration to get the overlapped area of the two images
3. Use SURF to extract feature points

{p_{i}} (i = 1 \dots m)

in the overlapped area on

I_{1}

4. Dived the overlapped area into

n \times n

grids, and

A r e a F l a g_{i} \leftarrow F A L S E (i = 1, \dots, n^{2})

5. Sort feature points, such that

p_{i} . r e s p o n s e > p_{i + 1} . r e s p o n s e (i = 1, \dots, m)

A n c h o r K e y p o i n t P a i r s \leftarrow {}

7. for

p_{i}

{p_{i}}

:
8. if (

p_{i}

is in the

k t h^{}

grid) and (

A r e a F l a g_{k} = F A L S E

):
9.

p_{i}

is used as anchor point, its corresponding position:

(u, v) \leftarrow f (p_{i} . x, p_{i} . y)

10. let region R be the neighbor window around

(u, v)

I_{2}

with

m a r g i n = 3 σ + p_{i} . F e a t u r e S i z e

11. extract SURF features points in R

\to

{q_{j}} (j = 1 \dots l)

12. for

q_{j}

{q_{j}}

:
13.

Δ_{j} = | p_{i} . d e s c r i p t o r - q_{j} . d e s c r i p t o r |

14. if

Δ_{\min} < t h r e s h o l d

:
15. push

〈 p_{i}, q_{m i n} 〉

into set

A n c h o r K e y p o i n t P a i r s

16.

A r e a F l a g_{k} \leftarrow T R U E

17. if all the

A r e a F l a g = T R U E

:
18. break
19. Solve transform matrix

H

with

A n c h o r K e y p o i n t P a i r s

20. if

| h_{11} - h_{22} | > ε

| h_{12} + h_{21} | > ε

(

h_{i j}

is the term of

H

):
21.

H \leftarrow

use rough registration

f (x, y)

to calculate transform matrix
22. Use H to stitch the images to get the stitched image
Output: a stitched image

I