Dense Depth Map Recovery Using Trinocular Stereo Fish-Eye Lenses

摘要

Fish-eye lenses are useful where a large field view is required. When the clearance between the object and the cameras is minimal, stereo fish-eye lenses can provide a full stereo view of the object where conventional stereo lenses fail to do so. This is because fish-eye lenses can provide a field of view which is approximately 180°. Though fish-eye lenses provide for a large field of view, they introduce significant distortion in the image. This paper presents an algorithm for high-accuracy dense depth map recovery by using trinocular stereo vision with fish-eye lens. We employ the normalizing local image, fish-eye distortion correction, area-based matching similarity measure and 2D-3D region growing for obtaining both precise and stable estimation of correspondence. Experimental results obtained on several real stereo images are presented.