Multiview 3D Reconstruction of Underwater Scenes Acquired With a Single Refractive Layer Using Structure from Motion

3D reconstruction of underwater objects is a challenging problem in computer vision due to the scattering and absorption of light which leads to contrast and colour degradations of the acquired images. In addition, refraction across media boundaries introduces geometric variations which lead to erroneous correspondence matching between images. In this paper, we present a refraction reconstruction model to compensate for refractive errors. Our model is for images acquired with the camera above the water surface instead of underwater camera images, excluding the possible refraction across glass-water interface. The corrected images are used under a multi-view 3D reconstruction framework to produce the 3D geometry of the underwater objects as well as camera pose.