In recent years, visual-inertial odometry has been increasingly adopted by researchers for underwater localization. However, due to the difficulty of obtaining ground truth trajectories in open seas, existing underwater localization datasets often lack complete and high accuracy trajectory references, which limits the application of visual‑inertial odometry in underwater scenarios. To address the issues of incomplete ground truth and insufficient accuracy in current underwater datasets, we present a new underwater dataset featuring high precision, complete trajectory ground truth. The data collection environment is illustrated in Figure 1.
Figure 1. Schematic diagram of the data collection environment
During data acquisition, we used the collection device shown in Figure 2. The entire setup consists of a ZED2 stereo camera, an extension pole, and two sets of motion capture markers. The image and IMU data are captured by the ZED2 stereo camera, which was waterproofed to enable normal operation underwater. The stereo camera is fixed at one end of the extension pole, with the two sets of motion capture markers attached to both ends. Together with the motion capture system, these markers provide accurate pose ground truth for the camera:
Figure 2. Core data acquisition equipment
To simulate various underwater scenes, we arranged three regions with different texture complexities in the artificial water pool where data collection took place, as shown in Figure 3.
Figure 3. Three regions with different texture complexity levels in the data collection environment
For more detailed information about the dataset, please visit: https://bat.sjtu.edu.cn/zh/haud-dataset/
