Real-time Vision-based Station Keeping for Underwater Robots

In this paper we describe the design and implementation of a control system for automatic vision-based station keeping with an underwater ROV, relative to some visual landmark. First, a region-based tracking system is described that tracks naturally textured landmarks in the image plane, based upon full planar-projective motion models. The implementation of the algorithm is such that it allows most of the computation to be done off-line, resulting in fast and accurate tracking at near video rate. Robustness is added to the algorithm by integrating optic flow information and by learning expected image motions. Furthermore, the problem of automatically selecting promising landmarks in underwater images is addressed, based upon the information contained in the texture map. We then turn to the vision-based control problem that uses the information provided by the tracker, to station keep an underwater ROV. A decoupled control design is presented for positioning the vehicle relative to a visual landmark, while maintaining a fixed depth. Additionally, an image stabilization technique is developed, which aims to keep the landmark centered in the image plane during station keeping maneuvers, by using a pan and tilt camera. The system was tested under real conditions and results obtained from station keeping tests at open sea are presented.