1. Introduction
Consider a situation where one would like to recover information about a hidden scene in an unknown room without directly peeking inside. Staring at the blank wall of the room from outside may reveal nothing to the naked eye, yet the wall reflects extremely faint but meaningful patterns of light from the hidden scene. We show that by analyzing a video of the blank wall, we can infer information about a person’s activity or classify the number of people in a hidden region of the scene, with no prior calibration or knowledge of the environment. Real-time in-situ use of such a non-line-of-sight (NLOS) method can be critical for search and rescue operations, law enforcement, emergency response, fall detection for the elderly, and detection of hidden pedestrians for intelligent vehicles [6], [29].