Robustly Removing Deep Sea Lighting Effects for Visual Mapping of\n Abyssal Plains

Abstract

The majority of Earth's surface lies deep in the oceans, where no surface\nlight reaches. Robots diving down to great depths must bring light sources that\ncreate moving illumination patterns in the darkness, such that the same 3D\npoint appears with different color in each image. On top, scattering and\nattenuation of light in the water makes images appear foggy and typically\nblueish, the degradation depending on each pixel's distance to its observed\nseafloor patch, on the local composition of the water and the relative poses\nand cones of the light sources. Consequently, visual mapping, including image\nmatching and surface albedo estimation, severely suffers from the effects that\nco-moving light sources produce, and larger mosaic maps from photos are often\ndominated by lighting effects that obscure the actual seafloor structure. In\nthis contribution a practical approach to estimating and compensating these\nlighting effects on predominantly homogeneous, flat seafloor regions, as can be\nfound in the Abyssal plains of our oceans, is presented. The method is\nessentially parameter-free and intended as a preprocessing step to facilitate\nvisual mapping, but already produces convincing lighting artefact compensation\nup to a global white balance factor. It does not require to be trained\nbeforehand on huge sets of annotated images, which are not available for the\ndeep sea. Rather, we motivate our work by physical models of light propagation,\nperform robust statistics-based estimates of additive and multiplicative\nnuisances that avoid explicit parameters for light, camera, water or scene,\ndiscuss the breakdown point of the algorithms and show results on imagery\ncaptured by robots in several kilometer water depth.\n