Enhancement and Evaluation of an AUV's Docking Station Detection with Deep-Learning Based Object-Detection

A specific field of application for resident Autonomous Underwater Vehicles (AUVs) is in under-ice environments, both on the terrestrial and extra-terrestrial settings. In such environments, the underwater docking process is a critical aspect as this allows AUV to return for charging, data transmission, and maintenance, increasing their efficiency and capability in underwater exploration. My thesis will delve into the enhancement of the detection capabilities of AUVs during the autonomous docking process using advanced deep learning models, focusing on YOLO for improved recognition in various visibility conditions. By leveraging a combination of AprilTag marker measurements and Ultra-Short Baseline 2D position data, the research aims to address the limitations of current visual confirmation methods in challenging underwater environments. Utilizing datasets from natural and controlled settings, the thesis will conduct a comparative analysis of model performance, emphasizing adaptability to diverse environmental conditions. The outcome will contribute to the efficiency and operational continuity of AUVs in under-ice explorations and other critical underwater missions.