DEPTH ESTIMATION IN LIGHT FIELD CAMERA ARRAYS BASED ON MULTI-STEREO MATCHING AND BELIEF PROPAGATION

Despite of the rich variety of depth estimation methods in the literature, computing accurate depth in multi-view camera systems remains a difficult computer vision problem. The paper proposes a novel depth estimation method for light field camera arrays. This work goes beyond existing depth estimation methods for light field cameras, being the first to employ an array of such cameras. The proposed method makes use of a multi-window and multi-scale stereo matching algorithm combined with global energy minimization based on belief propagation. The stereo-pair results are merged based on k-means clustering. The experiments demonstrate systematically improved depth estimation performance compared to the use of singular light field cameras. Additionally, the quality of the depth estimates is quasi constant at any location between the cameras, which holds great promise for the development of free navigation applications in the near future.