Many artificial intelligence applications rely on the ability to locate and recognize the people that are using the provided service. The involved tracking and recognition tasks are grossly different from those of the typical security application. The people are recorded from far away and act naturally. On the other hand, there is typically more than one sensor recording them and the decisions are based on some video stream, not on a single frame. This chapter addresses far-field unconstrained person tracking and video-to-video recognition, effectively answering the questions where are the humans, and who are they.