HAUD: A High-Accuracy Underwater Dataset for Visual-Inertial Odometry

Abstract

Visual-Inertial Odometry (VIO) is becoming more and more popular in underwater localization methods because of its high precision and low cost. However, due to the difficulty of obtaining ground truth in open-sea areas, existing underwater VIO datasets normally lack complete and accurate ground truth trajectories, which has limited the evolution of VIO in underwater scenes. This paper proposed an underwater visual-inertial dataset with complete and millimeter accuracy ground truth which was obtained through a motion capture system. To the best of our knowledge, it is the first underwater VIO dataset with millimeter accuracy ground truth. To make the dataset suitable for evaluating different VIO algorithms, different terrains and trajectories were designed in an artificial pool for data collection. Data collection devices include a stereo camera with a built-in IMU. The dataset was validated by several state-of-art vision-based localization algorithms, and the results demonstrate the usability of the dataset. The dataset is publicly available and can be downloaded from: https://bat.sjtu.edu.cn/zh/haud-dataset/.