Image stitching (create panorama image), calculate homography matrix

[freijon]

Dear NI Vision users,

I'm trying to stitch two images vertically together. I have a fixed configuration of two cameras. I chose their field of view so that there is an overlap zone between the two images and I want to stitch them together so that there is no visible border between the two images. My approach is as follows:

1) Acquire images and remove distortions by using the Distortion Modeling calibration method.

2) Crop an unknown amount of pixels (defined by trial and error method) at the bottom of the top image and at the top of the bottom image until "it fits"

3) Translate/Rotate the bottom image until "it fits" (again, trial and error).

Unfortunately, it never truely fits on the entire width of the stitched image, there are always mismatches (as seen at the middle left of the attachment). I also tried to use the distored image (without calibration) which didn't help either.

I'm wondering if there is a better method to do this. Its quite a common problem IMO and it can't be that hard. Does LabVIEW offer a method to calculate a one-time homography matrix which can be applied on future acquired images? Does anyone have experience with that kind of problem?

Thanks in advance

[NIG_Roman]

Hey Freijon,

I have found an example or your application.

Stitch Images Together in LabVIEW with Vision Development Module

https://decibel.ni.com/content/docs/DOC-4595

But I have no further experience with that kind of poblem.

Regards

Roman Imhoff

[freijon]

Hey Roman,

I used that example before as well. Unfortunately this SubVI doesn't account for overlapping regions in an image. I now use a trial and error method where you define the overlapping zone manually, and also correct rotation and translation errors manually. Maybe this is useful to someone, so I'll attach it. But I'm sure there is a better solution for this kind of problem!