Quantitative assessment of camera calibration for 3D reconstruction of underwater scenes in the presence of refraction

3D scene reconstruction is a well-researched topic for urban environments. However, this is applicable only to images taken in air. Underwater imagery poses new challenges due to the fact that most cameras are designed to operate in air, and in water they have to be placed in protective housings. Housing viewport distorts imagery by refracting light at two surfaces separating media with different refraction indexes. Thus, correction of imagery for Euclidean 3D reconstruction requires an additional calibration step - determination of refraction-related parameters. The set of these parameters depends on the geometry of a viewport. Most widely used viewports are flat and hemispherical. Refractive parameters are very difficult to measure directly and are known to be notoriously difficult to be determined by computational methods. We propose novel approaches for determination of these parameters and show how 3D reconstruction with refraction taken into account improves the results.