Segmentation of underwater images for AUV navigation

The paper considers vision based autonomous navigation of underwater vehicles along one dimensional structures such as sea-lines. The problems we address are the recognition of the sea-line in the image and the reconstruction of the relative pose of the camera with respect to the structure. Images are segmented in order to identify the pipe of the sea-line by performing edge detection and the Hough transform on the posterior probability images computed by a classifier. Each image pixel is classified according to color and texture features. Once, the sea-line has been located, the relative pose and position of the underwater vehicle is computed.