Depth Estimation with Sonar-based Correction for Low-cost Underwater Drone

Abstract

Listen

Translate article

Underwater drones are expected to survey in areas where human beings cannot go directly. To achieve automation of underwater drone operation, high-performance sensing technology is highly required. Especially, depth estimation plays an important role to confirm the relative position of the target objects and the underwater drone. However, there is still no efficiently trained deep learning model for underwater depth estimation to estimate the depth of underwater objects. In this paper, an underwater depth estimation model and depth correction method for depth maps using sonar is proposed. By enhancing the training dataset, an underwater depth estimation model is generated. Moreover, the generated depth map can be updated using the corrected depth of the target object catched by the assistant sonar system. Simulation results show that the proposed method performs highly accurate depth estimation for underwater images.