A vision-based system for autonomous underwater vehicle navigation

This paper describes the work for the design and development of an autonomous underwater vehicle (AUV). The reference missions are sea bottom surveys and sealines inspections. A vision-based system for the automatic underwater vehicle is presented. The detection of underwater pipeline borders and its symmetry axis is performed. The method adopted for edge detection consists of two steps: 1) a backpropagation neural network is applied to segment the underwater image into different regions; and 2) for each region, the best fit segment and the related parameters are extracted. Since the information on which regions are the right pipeline edges does not depend only on single region characteristics, but also on relations between regions, all the possible regions pairs are analyzed, in order to determine the right one. Satisfactory results are also obtained for pipelines partially covered by sand