Underwater image matching with efficient refractive-geometry estimation for measurement in glass-flume experiments

Abstract

Computer vision technology has been employed in numerous ocean engineering experiments to measure motions and shapes of target objects. However, when dealing with underwater objects, conventional image matching strategies are less effective because of refractive distortions. In this paper, a new method is proposed for matching underwater images in glass-flume experiments. It leverages the estimation of refractive geometry to refine potential image correspondences that include mismatches and location errors. By analyzing the light paths, the relationship between corresponding points of underwater images is derived. Then, an efficient estimation algorithm is designed to compute the parameters of refractive planes from image correspondences. Finally, an improved matching process is presented by incorporating this algorithm into an outlier removal framework. Extensive experiments on synthetic data and real images demonstrate that the proposed method is effective in calculating the refractive geometry of glass-flume experiments and leads to significantly higher accuracies in underwater image matching.