I have a set of images of a pipe taken by a camera that rotates 360 degrees and captures image every x degrees, hence inducing a constant overlap between the images. I need to stitch these images together such that they can be analysed as one big image, perhaps a panoramic image or orthomosaic. Here's a couple of examples:
Because it's a pipe, there's a slight curve in each image, so I am thinking first we can "unroll" the image, and do it for every image. After that, perhaps all those images can be stitched together.
I have tried unwrapping using "six-point" method ( you define cross-section of the cylinder with 3 points from the top and 3 from the bottom) , like you are unwrapping a sticker on a bottle, which is not terrible (can be improved of course). Here's how "unwrapping" looks like:
Second, SIFT is not working well for stitching. I am thinking it's because images are quite similar in nature. But, I am not sure how to best stitch them. This is where I need help. I need to align the crests of the pipe and stitch the images seamlessly - images could be up to 90 or 120. Would love any help here. Thanks.
This is something from a software, which is quite bad:
