Localize all people in sight wrt map

In receptionist "the robot must point at a place or location where the guest can sit". We can annotate all possible places to sit, it's more about making sure that we don't point at a place that's already occupied. In order to do this, we need to detect where the people are on the map. OpenPose or YOLO -> pull points from depth cam, transform to map frame is a reasonable first pass at this.