"Deep Learning" is one kind of machine learning.
To apply deep learning to count stairs, you would need a number of images each tagged with the number of stairs. The stairs depicted should have different spacings, different angles of view, painted different colors, made of different materials, carpetted different ways (including with patterns that might happen to have straight lines), some should be curved. You would also need a number of images in which there are no stairs at all -- some in which there is nothing remotely resembling stairs and others with things that sort of resemble stairs but are not. For example, curbs are like single-stair staircases and yet are not considered stairs; and rock ledges that happen to be straight-ish are not considered stairs, except for the cases they are considered stairs such as because they are used to climb up from a beach.
