In this article, we present a dataset of underwater videos captured through manual dives in a complex and unstructured seabed area dominated by harbor structures and coral reefs. The area is shallow (0.5 - 7.0 m depth) with an enclosed embayment for the harbor area, offering protection from ocean currents and waves. The coral reef area is located in a more open ocean sloping gently toward the deeper seafloor, leading to a more pronounced rolling shutter effect and camera motion. The dataset was collected using a GoPro Hero 10 camera, employing a standard wide lens with a horizontal field of view (FoV) of 109° and 768 × 432 image resolution. The camera is also equipped with an Inertial Measurement Unit (IMU) sensor, comprising a 200 Hz frequency accelerometer and gyroscope. During underwater deployment, the camera is protected with a 5 mm thick flat glass panel. This camera setting hence creates three medium layers of water-glass-air leading to additional refraction distortion. To address the refraction distortion, the dataset has been subject to pre-calibration utilizing flat refractive geometry found in the Pinax camera model. The Pinax camera model for the underwater imagery is calculated by combining the aspects of pinhole calibration parameters with axial camera projection. The main aim of the dataset collection is to facilitate the testing and evaluation of underwater imaging algorithms that are used in underwater robotics, such as computer vision, photogrammetry, and Simultaneous Localization and Mapping (SLAM).
Keywords: Ar-Track marker tracking; Computer vision; Coral reef environment; Natural seabed; Underwater imagery.
© 2023 The Author(s).