Pose estimation of a textureless box

I have a box (which has no texture) and I want to estimate its pose. The problem is, that I cant detect any key points, because of the missing texture. Another attempt (to detect the edges) also failed because of to much background noise. The next thing I tried was to use a color threshold to filter out the box. That failed as well, because of the different lighting in different poses.
Now I am out of ideas and ask for your help…
I have a 3d model of the box, if that helps
edit: I should add that I only have a RGB camera → no depth image and stereoscopic setup

Maybe if you took the image in complete darkness, it could make the detection a bit harder… :roll_eyes:

Thanks, that helps a lot
edit: if you think there is no solution, you can delete this post if you want

Anyway, maybe if you posted a photo of your box, that could give us a clue what could be used.


So I tried many other things now and came to the conclusion, that there is no real solution for my problem. I worked around my limitations, so that I can use SIFT features (plus FLANN Matcher) and homography + solvePnP to localize the box. I wrote a little registration script for different boxes, so that the user can easily add new ones if wanted