Vision-based pose estimation for autonomous operations in aquacultural fish farms

There is a largely increasing demand for the usage of Unmanned Underwater Vehicles (UUVs) including Remotely Operated Vehicles (ROVs) for underwater aquaculture operations thereby minimizing the risks for diving accidents associated with such operations. ROVs are commonly used for short-distance inspection and intervention operations. Typically, these vehicles are human-operated and improving the sensing capabilities for visual scene interpretation will contribute significantly to achieve the desired higher degree of autonomy within ROV operations in such a challenging environment. In this paper we propose and investigate an approach enabling the underwater robot to measure its distance to the fishnet and to estimate its orientation with respect to the net. The computer vision based system exploits the 2D Fast Fourier Transform (FFT) for distance estimation from a camera to a regular net-structure in an aquaculture installation. The approach is evaluated in a simulation as well as demonstrated in real-world recordings.