OpenCV Image stiching when camera parameters are known

Question

We have shots taken from an airplane flying over an area with 50% overlap, and use the OpenCV stapling algorithm to stitch them together. This is great for our version 1. In our next iteration, we want to learn a few additional things that I could use in a few comments.

The stitching algorithm is currently evaluating camera parameters. We have camera parameters and a lot of information available from the plane regarding camera angle, position (GPS), etc. Could we benefit from any of this information, as opposed to having the algorithm evaluate everything based on coincident points of the function?

These images are taken with high resolution, and the algorithm currently takes up a lot of RAM, and not a big problem, because we just rotate large machines in the cloud. But in our next iteration, I would like to release the homography from the sampled images and apply it to the larger images later. It will also give us more opportunities to manipulate and visualize other information about the source images and be able to go back and forth between the original and stitched images.

If we are going to parse the stitching algorithm in question 1 to place known information, it simply uses the method findHomographyto obtain information, or are there better alternatives for creating homography when we really know the flat position and angles and camera parameters.

I got a basic understanding of opencv and does a great job with C ++ programming, so its not a problem to write my own booklet, but the theory here is a little rusty.

Answer 1

Because you use homology to deform your images, I assume that you capture areas small enough so you don't have to worry about the effects of the Earth's curvature. In addition, I assume that you are not using the elevation model.

, () , - . , .

(, GPS + IMU). .

Answer 2

, , : , .

2D, H - , ,

H=[[a1 a2 a3] [a4 a5 a6] [a7 a8 a9]]

R T , a9 == 1.

R= [[a1 a2] [a4 a5]], T=[[a3] [a6]]

[a7 a8] - . ( , , ).

, , 3x3, a3, a6 a9=1 cv::warpPerspective cv::warpAffine.

, , .