Embedded Visual Detection and Shape Identification of Underwater Umbilical for Vehicle Positioning

Tethers are used in underwater operations to supply power and to transfer data transfer between underwater robots and a surface vessel. The real-time tracking of an underwater cable can be useful to make exploration missions and teleoperation safer by avoiding collisions and entanglement. This paper introduces a new method to detect and to estimate the 3D shape parameters of an underwater umbilical thanks to embedded visual and depth sensors. The tether buoyancy is assumed to be slightly negative, and the tether shape is approximated with a catenary curve whose lowest point is assumed to remain between the attachment points. The estimation of the cable parameters is based on a nonlinear least-square fitting procedure where the cable projection in the image is fitted with the 2D projection of the catenary shape. The underlying Gauss Newton algorithm is improved thanks to an initial guess, and a four step process that handles the calculation of the residual error between the detected points and the fitted curve using saturation, subsampling, M- estimation, and bounded iteration. Simulation is conducted in a test tank using a compact underwater robot connected to a 1.5m orange tether. The results obtained show that the method gives satisfactory results. This tether detection and shape identification can therefore be used by the robot to position itself with respect to the tether or with respect to the vessel at the other end.