Detecting linear motion of an object in a sequence of monocular underwater images

Abstract

A method of determining the egomotion of an autonomous underwater vehicle from a sequence of monocular underwater images is described. The algorithm is robust, fast, insensitive to noise and requires no exact knowledge of the camera parameters. The algorithm builds up a scale space pyramid of each of the images. The edges of the observed structures are determined using gradient-based methods. In the subsequent images, correlation-based methods are used to determine motion.