Passive indoor visible light localization detects changes in illumination patterns in a room to infer the position of a user. We present analytical bounds on positioning error of an object in an indoor scenario and use deep learning to estimate positioning in realistic scenarios.