It's a lot easier to detect and blur faces than to detect people, remove them and then try and fill in what was removed.