Performance of a Feature-Based Algorithm for Displacement Estimation of an Underwater Vehicle

An algorithm for estimating the displacement of an underwater vehicle was developed, which is based on detecting and matching features in seafloor images collected by the underwater vehicle itself. This feature-based algorithm consists of four main steps: (i) correcting radial distortions in an image, (ii) correcting tangential distortion for tilts of the camera image plane, (iii) detecting and matching features in an image sequence and (iv) converting image pixels into units of physical length. The success of image feature detection and matching depends on various factors such as the density of keypoints in the seafloor image, the feature detector, the threshold for Hessian keypoint detector, and the overlap of two adjacent images. This study aimed at investigating the effects of the above factors on the performance of the feature-based positioning algorithm. Two 5-minute seafloor videos with sparse and dense keypoints, respectively, collected off southwestern Taiwan at a depth of about 1000 meters by a deep-towed vehicle are used for testing the proposed feature-based algorithm. Performance evaluations were conducted by comparing the vehicle's trajectory estimated by the feature-based algorithm to the DVL navigation trajectory.