Target-based calibration of underwater camera housing parameters

This paper presents a method for calibrating a camera for underwater use. We consider the case that the camera is encased in a housing and views the world through a flat port. The method is analogous to common air calibration methods except that it occurs underwater where refraction plays a significant role in imaging. The calibration approach is as follows. The camera intrinsics are first calibrated in air using standard tools. The housed camera then records a video sequence of a moving calibration target underwater. From this sequence, the housing parameters (refractive indices, relevant housing dimensions, glass thickness and inclination angles), as well as the camera poses, are determined through a nonlinear optimization, which minimizes the reprojection errors of calibration target observations. In comparison to other methods, our method uses a moving, rather than static calibration target and we define our reprojection errors directly on the image plane of the camera, in order to accurately capture how noise affects the measurements. We show in a series of experiments that our approach significantly reduces the reprojection errors on an underwater evaluation sequence, when compared to an in air calibration and a model of underwater refraction with nominal parameters.