Using Visual Sensing for Control of an Underwater Robotic Vehicle

This paper documents research exploring the issues in using visual sensing to control an underwater robot. The local nature of visual sensing makes it useful for underwater situations in which the exact global location of the robot is not accurately known. In these situations, the relative measurement obtained from visual sensing can be used for precise control of the vehicle. To perform visual sensing in an underwater environment, several challenges must be overcome. In addition to standard vision processing issues, extreme lighting variations and marine snow make visual sensing problematic. To handle these di culties, an approach using Laplacian of Gaussian sign correlation hardware has been chosen which o ers robust real-time optical ow and stereo disparity measurements. This approach has been applied to several representative underwater vehicle tasks including: 1) object following, 2) station keeping, 3) mosaicking, and 4) navigation.