Images for underwater 3D reconstruction of Posidonia seagrass transplantation site

The photographic data, which are the first step for data processing, were acquired with a low-cost action camera (GoPro Hero 9) capable of shooting high quality (20 Mpix images) photos with a specific time-lapse function. To perform the acquisition over a large area (3200 m2) the camera was mounted on a Diver Propulsion Vehicle (DPV) which allowed the SCUBA diver to map the entire area in only 1h and 10’. Consecutive (2 seconds interval) photos were acquired in raw format to allow further manual editing aimed at improving the result via colour grading and correction. Subsequently, each image was processed in a photogrammetric software (Agisoft Metashape) based on Structure from Motion (SfM) processing algorithms to generate 2D raster outputs (orthophoto mosaics and Digital Elevation Models, DEM),3D dense point clouds and 3D triangular mesh with texture. Finally, the open-source software Blender 2.93 was used to render the 3D scene to create a realistic view of the transplanted area by adding water effects and Posidonia leaves that otherwise cannot be well reproduced by only SfM data.

作为数据处理首要步骤的摄影数据集，由一款低成本运动相机（GoPro Hero 9）采集。该相机支持高质量成像（2000万像素图像），并具备专用延时摄影功能。为完成3200平方米大面积区域的数据采集，相机被安装于水下推进器（Diver Propulsion Vehicle, DPV）上，使得水肺潜水员仅用1小时10分钟即可完成整片区域的测绘。采集的连续照片（拍摄间隔2秒）采用原始格式存储，以便后续通过色彩分级与校正开展手动编辑，优化最终成果。随后，基于运动恢复结构（Structure from Motion, SfM）处理算法，使用摄影测量软件（Agisoft Metashape）对每张图像进行处理，生成二维栅格输出成果（正射影像镶嵌图与数字高程模型（Digital Elevation Models, DEM））、三维密集点云以及带纹理的三维三角网格模型。最后，使用开源软件Blender 2.93对三维场景进行渲染，通过添加水体效果与波西多尼亚海草叶片，还原移植区域的真实视觉效果——仅依靠运动恢复结构数据无法很好地复现此类元素。