Self-supervised underwater depth estimation based on polarization binocular imaging

In this Letter, we introduce a self-supervised depth estimation method based on polarization binocular imaging. First, an end-to-end disparity estimation network is utilized to estimate the left and right disparities from the stereo view images. Next, we design loss functions that facilitate self-supervised training of the network, eliminating the need for labeled data. The self-supervised framework fully leverages the strengths of both binocular and polarization imaging. The effectiveness of the proposed algorithm is validated using real underwater polarized binocular data.