Wide Range Depth Estimation from Binocular Light Field Camera

Light field camera have been developed to capture spatial and angular information of rays. But limited by the structure of micro-lens, it acts as a short-baseline multi-view camera. Foreground objects have accurate depth map in light field while depth information is missing in the background. To avoid such a drawback in the existing light field depth estimation system, we propose a binocular light field camera and introduce longbaseline stereo matching in it. The system can estimate wide range depth of scene by merging complementary depth map of far scene into depth map from light field camera. We firstly estimate an relative depth map from light field and stereo matching respectively, and present calibration methods that normalize both depth maps to the real depth space. Then we model depth fusion problem as Markov random field which can be solved by graph cuts efficiently. Experiments show that our system have a wider depth sensing ability than either single light field camera or traditional binocular camera.