Application of visual servoing to the dynamic positioning of an underwater vehicle

Abstract

Conventional underwater sensors are not well suited to the task of aiding unmanned underwater vehicles to hover. These sensors suffer from several drawbacks such as low sampling rates, low resolution, complexity of operations, drift and cost. Underwater video cameras, however, can provide local measurements of position with respect to a local object. Underwater vision presents several challenges: it suffers from a limited range and poor visibility conditions. Besides, recovering motion from images requires high computing power, which is a limited resource on-board most underwater vehicles. The main objective of this thesis was therefore to investigate visual control methods to dynamically position a typical underwater vehicle with respect to a fixed object. These methods also had to have computing power’s needs compatible with off-the-shelf embedded computers that can be operated on real vehicles. A hybrid visual servoing technique, adaptated from the 2 1/2 D visual servoing scheme, was proposed. Its performance was assessed in simulations using a six degrees-of-freedom nonlinear dynamic model of an Remotely Operated Vehicle. The effects of sea current disturbances, the target’s orientation, and noise under sparse feature tracking conditions was studied. The proposed method proved stable and took into account the restrictive controllability of the vehicle. A 2-D visual servoing scheme which employed the Shi-Tomasi-Kanade sparse feature tracker on unmarked planar targets in a water tank was then proposed. The scheme controlled a planar Cartesian robot which emulated the dynamic behaviour of the surge and sway degrees-of-freedom of a typical underwater vehicle. The effect of sea current disturbances on the stability and performance of the control scheme was also studied. An underwater experimental evaluation of the Shi-Tomasi-Kanade feature tracker under various conditions of lighting and relative speed between the underwater scene and the camera was also performed.