Global descriptors suitable for a set of depth images

Can you give some examples of global descriptors suitable for a set of depth images (or generally grayscale images) for the purpose of classification?

please try to explain, which problem you want to solve

There are a bunch of depth images from a moving object in which the object is steady in certain frames. The goal is to specify these frames. I am thinking of exploiting the global descriptors in order to encode the similarity of frames to specify the steady frames.

so, this is about detecting movement, not classification ?

what would a “global descriptor” be ? do you mean something like Daisy ?

also, what have you tried, so far, and where do you need help ?