Yes the sky background has to be subtracted somehow. The same problem crops up in spectroscopy of supernovae and some programs have various ways of fitting a model to the sky background in the surrounding region. In my limited experience of photometry I think programs which model the star point spread function against the background work better in this situation, though there could still be a bright/or dark patch directly under the target of course. The transient survey instruments get round this problem by subtracting an image taken when the transient was not there.
Robin