OpenCV: When to use the GridAdaptedFeatureDetector?

Question

I am trying to make a descriptor based detector. I use OpenCV, and I saw that there are many types of objects and types of descriptors, as well as types of sockets. Moreover, I also saw that for types of objects there can be such types as Grid or Pyramid.

I did not find a good explanation for them (except Pyramid, which says it’s good " for detectors that are not inherently scalable "). I want to have a short description of each type and each combination (feature-descriptor-matcher) in order to compose a presentation, rather than create an exhaustive work of finding and checking each possible combination.

Does anyone know more information about this?

Answer 1

The term function is usually used for two different things:

• function detectors
• function descriptors.

The detector seeks ... well ... to detect good points of interest, that is, points that are stable when changing the point of view and light and which provide good performance in tasks such as evaluating homography or detecting an object.

The descriptor aims to achieve good consistent performance for detected points under a specified point of view and lighting changes.

. (Moravec, Harris, ) (FAST).

, , SIFT SURF. ( , SIFT HoG ). (.. ).

, , (FAST). FAST . , BRIEF, BRISK, FREAK, ORB... , ( ).

, :

• . : , , , ( ) (SIFT, SURF);
• , . , ,
• SIFT SURF , L2. (BRIEF, BRISK, FREAK, ORB) , .