IMPROVING 3-D VARIATIONAL STEREO RECONSTRUCTION OF OCEANIC SEA STATES BY CAMERA CALIBRATION REFINEMENT

Studies of wave climate, extreme ocean events, turbulence, and the energy dissipation of breaking and non-breaking waves are closely related to the measurements of the ocean surface. To gauge and analyze ocean waves on a computer, we reconstruct their 3-D model by utilizing the concepts of stereoscopic reconstruction and variational optimization. This technique requires a pair of calibrated cameras cameras whose parameters are estimated for the mathematical projection model from space to an image plane to take videos of the ocean surface as input. However, the accuracy of camera parameters, including the orientations and the positions of cameras as well as the internal specifications of optics elements, are subject to environmental factors and manual calibration errors. Because the errors of camera parameters magnify the errors of the 3-D reconstruction after projection, we propose a novel algorithm that refines camera parameters, thereby improving the accuracy of variational 3-D reconstruction. We design a multivariate error function that represents discrepancies between captured images and the re-projection of the reconstruction onto the images. As a result of the iteratively diminished error function, the camera parameters and the reconstruction of ocean waves evolve to optimal values. We demonstrate the success of our algorithm by comparing the reconstruction results with the refinement procedure to those without it and show improvements in the statistics and spectrum of the wave reconstruction after the refinement procedure.