Abstract
Passive depth estimation is among the most long-studied fields in computervision. The most common methods for passive depth estimation are either astereo or a monocular system. Using the former requires an accurate calibrationprocess, and has a limited effective range. The latter, which does not requireextrinsic calibration but generally achieves inferior depth accuracy, can betuned to achieve better results in part of the depth range. In this work, wesuggest combining the two frameworks. We propose a two-camera system, in whichthe cameras are used jointly to extract a stereo depth and individually toprovide a monocular depth from each camera. The combination of these depth mapsleads to more accurate depth estimation. Moreover, enforcing consistencybetween the extracted maps leads to a novel online self-calibration strategy.We present a prototype camera that demonstrates the benefits of the proposedcombination, for both self-calibration and depth reconstruction in real-worldscenes.