3d representatio of underwater structures using Nerfs

Problem Statement

Existing products that re-create underwater structures using traditional methods such as structure-from-motion (SFM) are costly, power-hungry and heavy. This limits the collection of data to seabed level extraction, since near-surface structures experience turbulent waves, intermittent lighting and high currents. The vehicle does not have enough power to supply the equipment doing SFM since the power is required for the vehicle to fight the waves.

Solution

A low cost solution using existing video recordings was prototyped. Using the positional data and image data, a mixed-bag solution using SFM and a new method called Neural Radiance Fields (NeRFs), example solutions were created and proposed.