Panorama stitching with very uniform images

asked 2014-07-07 17:31:05 -0500

MartinArroyo gravatar image

Hello everyone! I am trying to implement a panorama creator using 2.4.9. So far the results have been great, I started with the stitching_detailed.cpp example and then I made some adjustments. However, I am facing a very difficult set of images. Basically I would like to create a panorama of a vertical structure made of concrete (basically, the whole structure is gray). Since the colour is very uniform and there are not a lot of detectable features, the algorithm is incapable of creating a panorama from of the input. I would like to know if any of you has had to deal with this kind of situation, or any suggestions you may have in order to solve this problem. I would truly appreciate your help!

Thanks in advance.

edit retag flag offensive close merge delete



I worked with images of a desert once (lots of sand, not cake) and stitching failed. Simply there were not enough good pairs to calculate the homography. You may try to experiment with different detectors/descriptors though. Some of them may be more sensitive than others. When I was doing my stuff, only SIFT and SURF were around.

Witek gravatar imageWitek ( 2014-07-07 19:31:29 -0500 )edit

I have seen that some features detectors search for features only in the center of the image, like ORB for example... Maybe it helps...

thdrksdfthmn gravatar imagethdrksdfthmn ( 2014-07-08 02:34:00 -0500 )edit

Just an idea that pops in my mind. Should it be possible to introduce extra texture somehow in your images at known positions that can then be matched?

StevenPuttemans gravatar imageStevenPuttemans ( 2014-07-08 04:26:16 -0500 )edit

That is one of the ideas that I've come up with. I will try it if I have no other choice. Thanks!

MartinArroyo gravatar imageMartinArroyo ( 2014-07-08 12:22:35 -0500 )edit

I've read a paper once called "Reconstructing Building Interiors". It uses Manhattan World assumption and computes the main three orthogonal axes of the scene. But it has quite a large pipeline, first computing a sparse 3d point cloud with, then a dense, then Manhattan World... But maybe you can take some ideas out of it.

Duffycola gravatar imageDuffycola ( 2014-07-21 10:50:03 -0500 )edit