Difficulty comparing features

Question

Im working on a 3d reconstruction project in which I have problems setting up to continue the reconstruction. To be more specific, when im matching the image characteristics of matlab examples, I have a high coefficient of correct incorrect matches, but when they match the functions of my own photos taken by the phone’s camera, I have almost only incorrect matches. I tried setting the threshold, but the problem still remains. Any ideas / opinions on what is going wrong?

The im descriptor uses the sift descriptor from vlfeat toolkit

edit: here is the link with the source images, detected main / corner points and matches.

Answer 1

I think your main problems here are the significant difference in lighting between the images and the specular reflections from the plastic case. You also look at the inside of the USB drive through transparent plastic, which does not help.

What features of the detectors / descriptors have you tried? I would start with SURF and then try MSER. It is also possible to use multiple detectors and descriptors, but you must be careful to separate them. Of course, there are also many options for you.

Another thing that may be helpful is to get higher resolution images.

3D-, , ? vision.PointTracker , .