Turbid-Water Subsea Infrastructure 3D Reconstruction with Assisted Stereo

This paper studies underwater perception and short-range scene reconstruction. Our work enables autonomous manipulation behaviors that support the autonomous maintenance of subsea infrastructure. We present a system that leverages assisted stereo to reconstruct the geometry of textured or untextured structures immersed in turbid water. Our package projects a random binary pattern in the cameras' field of view, which facilitates stereopsis in areas that are not naturally textured. We discuss the design and assembly of the package, and we quantify the accuracy and coverage of our method in turbid water ranging from 1 to 2.5 NTU.