Visual Seafloor Mapping and Monitoring

Satellites can only look a few meters beyond the water surface in clear conditions. Murky coastal waters or deeper water obscure the seafloor and visual mapping of underwater installations, cables, pipelines or also geologic settings or biologic habitats can only be done by diving close to the seafloor. On top of difficult visibility, there is often not enough natural light, such that camera platforms, idvers or robots have to use artificial light sources that can creates light cones and strong scattering effects. These light cones move with the camera and so the same 3D point appears darker or brighter when at the boundary or in the center of the light cone, and depending on how close the light is to the surface. On top, water interacts with light and distance variations of a robot or diver from the seafloor leads to more or less greenish or blueish appearance. These effects are called photometric effects, because they affect the colors of the image. On top, cameras are typically mounted in water-proof housings and look through a window into the marine world. Light rays that pass this window change their direction: Air, glass and water have different optical densities, and Snell's law predicts how light rays are bent, in case all parameters are known. These effects can violate the pinhole camera model and lead to biased / bent reconstructions (flat seafloor looks curves, objects have wrong sizes) when off-the-shelf photogrammetry software is used that does not consider underwater effects. Finally, the footprint of a pixel on the ground is typically sub-mm and the angular resolution of a camera is much better than 0.1°. Consequently, the camera position and orientation must be known to this accuracy in order to fully exploit the full image resolution. Our 3D mapping algorithms therefore recover position and orientation from the images themselves, since no external sensor comes anywhere close to this accuracy. 3D points on the seafloor are tracked across many images, allowing to extract camera motion and 3D geometry. Physical or statistical water/light and color models can then be used to reconstruct seafloor colors "without water".