When Two Cameras Are a Crowd

Vision and robotics systems enabled by cameras that recover 3D scene geometry are revolutionizing several aspects of our lives via technologies such as autonomous transportation, robotic surgery, and'hands-free' user interfaces. Modern 3D cameras are active devices, where a programmable light source emits coded illumination. The emitted light gets reflected from the scene and is received by a sensor to infer the 3D structure of the surroundings. In a multi-camera environment, such active 3D cameras may receive light from the sources of other cameras, resulting in large depth errors. This problem is becoming increasingly important due to the emergence of low-cost and compact active 3D cameras, which are becoming ubiquitous across a wide range of applications, from consumer devices to vehicular vision systems.